Rising Inequality in China : Challenges to a Harmonious Society 9781107248236, 9781107002913

Citation preview

more information - www.cambridge.org/9781107002913

RISING INEQUALITY IN CHINA

This book, a sequel to Inequality and Public Policy in China (2008), examines the evolution of inequality in China from 2002 to 2007, a period when the new “harmonious society” development strategy was adopted under Hu Jintao and Wen Jiabao. It fills a gap in knowledge about the outcomes of this development strategy for equity and inequality. Drawing on original information collected from the most recent two waves of nationwide household surveys conducted by the China Household Income Project, this book provides a detailed overview of recent trends in income inequality and cutting-edge analysis of key factors underlying such trends. Topics covered include inequality in education, changes in homeownership and the distribution of housing wealth, the evolution of the migrant labor market, disparities between public and nonpublic sectors, patterns of work and nonwork, gender and ethnic gaps, and the impacts of public policies such as reforms in taxation and social welfare programs. Li Shi is China’s leading specialist on inequality and poverty. He has served as the acting director of the China Institute of Income Distribution at Beijing Normal University since 2011. His numerous published works include Inequality and Public Policy in China edited with Bj¨orn Gustafsson and Terry Sicular (Cambridge University Press, 2008); Unemployment, Inequality and Poverty in Urban China edited with Hiroshi Sato (2006); and numerous articles in Chinese and Western scholarly journals. He has won many academic prizes, including the Sun Yefang Prize for Economic Science (1994 and 2011) and the Zhang Peigang Prize for Development Economics (2010). Hiroshi Sato has published many works on topics related to development economics and inequality in China. He is the coeditor of Unemployment, Inequality and Poverty in Urban China (2006) and author of The Growth of Market Relations in Post-Reform Rural China (2003), and he has contributed to numerous works including Inequality and Public Policy in China. He received the IDE Prize for Research on Developing Economies in 2004 for his Japanese book Shotoku Kakusa to Hinkon (Income Inequality and Poverty, 2003). Terry Sicular is a leading North American specialist on the Chinese economy and has written extensively on inequality, poverty, the labor market, and the rural economy in China. She is a coeditor of and contributor to Inequality and Public Policy in China (2008). Her works have appeared in the Review of Income and Wealth, the Journal of Development Economics, and Economic Journal. She is a recipient of the Zhang Peigang Prize for Development Economics (2010) and the Sun Yefang Prize for Economic Science (2011).

Rising Inequality in China Challenges to a Harmonious Society

Edited by

LI SHI Beijing Normal University, China

HIROSHI SATO Hitotsubashi University, Japan

TERRY SICULAR University of Western Ontario, Canada

cambridge university press Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, S˜ao Paulo, Delhi, Mexico City Cambridge University Press 32 Avenue of the Americas, New York, NY 10013-2473, USA www.cambridge.org Information on this title: www.cambridge.org/9781107002913  C

Cambridge University Press 2013

This publication is in copyright. Subject to statutory exception and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published 2013 Printed in the United States of America A catalog record for this publication is available from the British Library. Library of Congress Cataloging in Publication Data Rising inequality in China : challenges to a harmonious society / Li Shi, Beijing Normal University, China, Hiroshi Sato, Hitotsubashi University, Japan, Terry Sicular, University of Western Ontario, Canada. pages cm Includes bibliographical references and index. ISBN 978-1-107-00291-3 (hardback) 1. Equality – China. 2. China – Social policy – 21st century. 3. China – Economic conditions – 2000– I. Li Shi, 1956– HM821.R57 2013 305.800951–dc23 2012042752 ISBN 978-1-107-00291-3 Hardback Cambridge University Press has no responsibility for the persistence or accuracy of URLs for external or third-party Internet websites referred to in this publication and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.

Contents

List of Tables

page vii

List of Figures

xv

Contributors

xix

Preface

xxi

Abbreviations

xxiii

Glossary

xxv

1

Rising Inequality in China: Key Issues and Findings Li Shi, Hiroshi Sato, and Terry Sicular

2

Overview: Income Inequality and Poverty in China, 2002–2007 Li Shi, Luo Chuliang, and Terry Sicular

3

Housing Ownership, Incomes, and Inequality in China, 2002–2007 Hiroshi Sato, Terry Sicular, and Yue Ximing

85

Educational Inequality in China: The Intergenerational Dimension John Knight, Terry Sicular, and Yue Ximing

142

Inequality and Poverty in Rural China

197

4

5

1 44

Luo Chuliang and Terry Sicular

6

7

The Evolution of the Migrant Labor Market in China, 2002–2007 John Knight, Deng Quheng, and Li Shi

230

A New Episode of Increased Urban Income Inequality in China

255

Deng Quheng and Bj¨orn Gustafsson v

vi 8

9

10

Contents Unemployment and the Rising Number of Nonworkers in Urban China: Causes and Distributional Consequences Bj¨orn Gustafsson and Ding Sai

289

Do Employees in the Public Sector Still Enjoy Earnings Advantages? Yang Juan, Sylvie D´emurger, and Li Shi

332

Redistributive Impacts of the Personal Income Tax in Urban China Xu Jing and Yue Ximing

362

11

Changes in the Gender-Wage Gap in Urban China, 1995–2007 Li Shi and Song Jin

12

Intertemporal Changes in Ethnic Urban Earnings Disparities in China Ding Sai, Li Shi, and Samuel L. Myers, Jr.

384

414

Appendix I. The 2007 Household Surveys: Sampling Methods and Data Description Luo Chuliang, Li Shi, Terry Sicular, Deng Quheng, and Yue Ximing

445

Appendix II. The 2002 and 2007 CHIP Surveys: Sampling, Weights, and Combining the Urban, Rural, and Migrant Samples Song Jin, Terry Sicular, and Yue Ximing

465

Index

487

List of Tables

1.1. 1.2. 1.3. 1.4. 2.1. 2.2. 2.3. 2.4. 2.5. 2.6. 2.7. 2.8. 2.9. 2.10. 2.11. 2.12. 2.13. 2.14. 2.15. 2.16.

Key indicators of redistributive public policies Coverage of the CHIP 2002 and 2007 surveys Comparison of CHIP and NBS household per capita incomes, 2002 and 2007 Key indicators of inequality and poverty in China National mean income and inequality, 2002 and 2007 Decomposition of inequality by income sources, 2002 and 2007 Inequality estimates with and without PPP adjustments, 2002 and 2007 Level, composition, and growth of migrant household per capita income Migrant inequality, 2002 and 2007 Decomposition of migrant income inequality by income source, 2002 and 2007 Urban inequality with and without migrants, 2002 and 2007 The urban-rural income gap, 2002 and 2007 Contribution of urban-rural (between-group) inequality to national inequality (%) Contributions of urban-rural (between-group) inequality to national inequality, with PPP adjustments (%) Regional income gaps, 2002 and 2007 Contributions of between-region inequality to overall inequality (%) Gini coefficients by region, 2002 and 2007 The urban-rural income gap by region, 2002 and 2007 Poverty lines (yuan) Poverty incidence and composition, 2002 and 2007 (%) vii

page 10 22 28 31 54 58 60 61 63 63 64 65 66 67 68 70 71 72 74 74

viii

List of Tables

2.17. The structure of poverty by region (%) 2A.1. Income and inequality with alternative weights, 2002 and 2007 2A.2. Income and inequality with alternative estimates of imputed rental income on owner-occupied housing, 2002 and 2007 2A.3. Mean income per capita by region, 2002 and 2007 (yuan) 3.1. Chronology of housing reform 3.2. Housing tenure for rural, urban, and migrant households, 2002 and 2007 (% of households) 3.3. Mortgage debt among homeowner households, 2002 3.4. Comparisons of housing market value and equity per capita, 2002 3.5. Alternative estimates of imputed rents and income per capita based on market value versus equity value of owner-occupied housing, 2002 3.6. Mean housing wealth per capita, 2002 and 2007 (in yuan and as a percentage of income per capita) 3.7. Average annual increases in per capita housing wealth, 2002 to 2007 (percentage, constant prices) 3.8. Ratios of per capita housing wealth between urban, rural, and migrant households, 2002 and 2007 3.9. Inequality of housing wealth, 2002 and 2007 (Gini coefficients) 3.10. Distribution of housing wealth across income quintiles, 2002 and 2007 3.11. Estimates of per capita imputed rental income from owner-occupied housing, 2002 and 2007 (in yuan and as a percentage of income per capita) 3.12. Imputed rents and income inequality, 2002 and 2007 3.13. Characteristics of urban households used in the analysis of urban housing tenure choice, 2002 and 2007 3.14. Multinomial logit analysis of housing tenure choice in the urban areas, 2002 and 2007 3.15. Characteristics of urban households in the analysis of urban housing wealth, 2002 and 2007 3.16. Determinants of housing wealth in the urban areas, 2002 and 2007 3.17. Characteristics of rural households in the analysis of rural housing wealth, 2002 and 2007

75 79 81 81 88 92 101 101

102 105 106 106 107 108

109 110 113 116 119 122 125

List of Tables 3.18. 3A.1. 3A.2. 3A.3. 3A.4. 3A.5.

3A.6.

4.1. 4.2a. 4.2b. 4.3a. 4.3b. 4.4. 4.5.

4.6.

4.7.

4.8.

Determinants of housing wealth in the rural areas, 2002 and 2007 Relevant housing variables in the 2002 and 2007 CHIP data sets Comparison of urban housing market values from the CHIP and the NBS, 2002 Comparison of urban market rental values of housing from the CHIP and the NBS, 2002 Comparison of the urban rent-price ratio from the CHIP versus that from the NBS, 2002 Formulae for alternative estimates of imputed rental income on owner-occupied housing incorporating costs of ownership, 2002 and 2007 Gini coefficients of household per capita income, calculated using different estimates of imputed rental income, 2002 and 2007 Descriptive statistics for matched individuals and parents in the 2007 CHIP used in the analysis Cross-tabulation of one’s own educational level by the educational level of the father (number of observations) Cross-tabulation of one’s own educational level by the educational level of the mother (number of observations) Average years of education of son by levels of father’s and mother’s education Average years of education of daughter by levels of father’s and mother’s education Regressions of one’s own education as a function of the parents’ average education, all birth cohorts combined Regressions of men’s and women’s own education as a function of their parents’ average education, all birth cohorts combined Regression equations: One’s own education as a function of location, gender, and birth cohort for education-poor and education-rich households Differences in one’s own education between individuals whose parents have no education and individuals whose parents have a junior middle-school or higher education, by cohort Educational inequality and the contribution of parental education

ix

128 132 134 134 135

137

138 159 161 161 166 166 169

170

175

177 182

x

List of Tables

4A.1. Educational levels used in the analysis 4A.2. Conversion of educational levels in the rural questionnaires to codes and years of education used in the analysis 4A.3. Conversion of educational levels in the urban questionnaire to codes and years of education used in the analysis 5.1. Rural household per capita income, 2002 and 2007 5.2. Rural household per capita income, by source 5.3. Estimates of the rural Gini coefficient, 2002 and 2007 5.4. Alternate measures of inequality in rural China, 2002 and 2007 5.5. Gini coefficient decomposition, by income source 5.6. Poverty lines 5.7. Poverty estimates 5.8. Decomposition of changes in poverty, 2002–2007 5.9. Per capita income and its composition for nonpoor and poor households 5.10. Composition of the income difference between nonpoor and poor households 5.11. Percentage of households in each province of the CHIP rural survey reporting wage earnings from migrant employment 5.12. The relationship between migration and poverty 5.13. Taxes and fees paid by rural households (per capita), by deciles 5.14. Taxes and fees paid by poor and nonpoor households (per capita) 5.15. Taxes and fees paid by the poor relative to the poverty gap 5.16. Basic statistics on individuals in dibao versus non-dibao households, from the CHIP rural household survey, 2007 5.17. The relationship between dibao participation and poverty, 2007 6.1. Labor force and employment in China, 1995–2007 6.2. The determinants of migrant log wage income and log self-employment income, 2007 6.3. The determinants of the proportionate change in the migrant wage and self-employment income, 2002–2007 6.4. Decomposition of the increase in the average real migrant wage, 2002–2007: Selective summary. Contribution of change in the mean characteristics to the gross mean wage increase: Percentage

191 191 192 200 201 203 204 206 208 209 211 212 213 216 217 219 220 220 222 225 233 238 241

243

List of Tables Dispersion of migrant average city wage across cities, 2002 and 2007 6.6. Probit equations predicting the probability of migrant status, 2002 and 2007 6.7. Reasons given by nonmigrant workers for not migrating: Distribution of the replies and the relationship of the replies to the probability of migrant status 6.8. Frequency distribution of the number of migrants and nonmigrants by predicted probability of migrating, and “expected value” of migration by nonmigrants, 2002 and 2007 (million) 7.1. Income inequality 1988, 1995, 2002, and 2007, according to various inequality indices 7.2. Absolute poverty in urban China, 1988, 1995, 2002, and 2007 7.3. Relative poverty in urban China, computed using various relative poverty lines, 1988, 1995, 2002, and 2007 7.4. Components and growth of household income per capita, 2002 and 2007 7.5. Household income per capita and its decomposition, 2002 and 2007 7.6. Decomposing differences in the Gini coefficient for 2002 and 2007 by income sources 7.7. Population shares, mean income, income inequality, relative poverty, and proportion of affluence among individuals living in households primarily connected to the state sector, the private sector, and those with no workers, 2002 and 2007 7.8. Population shares, mean income, income inequality, relative poverty, and proportion of affluence among children, adults, and the elderly, 2002 and 2007 7.9. Population shares, mean income, income inequality, relative poverty, and proportion of affluence among individuals living in households with the heads of households having different levels of education, 2002 and 2007 7.10. Income function: Dependent variable, log of household per capita income 7.11. Predicted probabilities of relative poverty and affluence, 2002 and 2007 (percentages) 7A.1. Descriptive statistics 7A.2. Poverty function (poverty line set at 70 percent of the median income)

xi

6.5.

244 245

247

248 263 265 266 268 269 270

273

276

277 278 280 283 284

xii

List of Tables

7A.3. Affluence function, with 200 percent of the median income as the threshold 8.1. Categories of nonworkers 1988, 1995, 2002, and 2007, men ages eighteen to sixty and women ages eighteen to fifty-five 8.2. Nonworkers by category, age, and gender, 1988, 1995, 2002, and 2007 (percentage of persons in different age categories) 8.3. Determinants of various states of nonwork among persons ages eighteen to twenty-nine, in 1995, 2002, and 2007 8.4. Determinants of various states of nonwork among persons ages thirty to fifty-five or sixty, 1995, 2002, and 2007 8.5. Economic dependency of married women in urban China, 1988, 1995, 2002, and 2007 (ages eighteen to sixty for males and ages eighteen to fifty-five for females) 8.6. Personal income and household disposable per capita income among the employed and various categories of nonworkers, 1995, 2002, and 2007 (means and Gini coefficients) 8.7. Adult persons by deciles of personal income and disposable household per capita income, 1988, 1995, 2002, and 2007 8A.1. Descriptive statistics for the sample of young adults, 1995, 2002, and 2007 8A.2. Descriptive statistics for the sample of middle-aged and older workers, 1995, 2002, and 2007 9.1. Definition of ownership categories 9.2. Descriptive statistics on individual characteristics by ownership 9.3. Descriptive statistics on individual earnings by ownership 9.4. Hourly wage functions by ownership, 2002 9.5. Hourly wage functions by ownership, 2007 9.6. Oaxaca-Blinder decomposition of log hourly wages by ownership 10.1. Share of major taxes in total tax revenue in selected years after the 1994 tax reform 10.2. Comparison of household data average tax rates and alternative data average tax rates 10.3. Mean income and proportion of individuals (non)reporting the personal income tax 10.4. Mean business operating income and the proportion of individuals (non)reporting the personal income tax 10.5. Average personal income tax rate by decile 10.6. The MT index and the P index

285 305 306 311 314

318

322 324 328 329 339 340 342 346 347 350 364 370 371 372 377 378

List of Tables 10.7. 10A.1.

11.1. 11.2. 11.3. 11.4. 11.5. 11.6. 11A.1. 11A.2. 11A.3. 12.1. 12.2.

12.3. 12.4. 12.5. 12.6. 12.7. 12.8. AI.1. AI.2. AI.3.

Decomposition of the MT index into the effects of horizontal equity and vertical equity Main elements of the personal income tax in China: Categories of income subject to the personal income tax by category, the time basis for the tax levied, deductions, and the tax schedule Labor-force participation and unemployment Wage structure and the gender-wage gap in urban China, 1995, 2002, and 2007 Regression analysis on the gender-wage gap in urban China Oaxaca’s decomposition analysis for the gender-wage gap, 1995, 2002, and 2007 Decomposition results from the quantile regression analysis Decomposition results for changes in the gender-wage gap Proportion of sample in urban China, 1995, 2002, and 2007 Wage functions in urban China, 1995, 2002, and 2007 (results of linear regression) Wage functions in urban China, 1995, 2002, and 2007 Minority and Han salary or wage income in the same twelve provinces Ratio of minority-to-Han income and ratio of the income of those eighteen to thirty years old to the income of those thirty-one to sixty years old Ordinary least squares estimates of the effects of minority status on ln-earnings Returns to education and employment in state-owned enterprises Descriptive statistics from the 1995, 2002, and 2007 CHIP data Residual difference analysis of ethnic minority versus Han wage and salary income Determinants of changes in the disparities in ethnic earnings Intratemporal and intertemporal decomposition of the disparity measure CHIP sample size for each subgroup, 2007 Samples covered by the CHIP and NBS data (number of households) Distribution of households in the 2007 urban sample, by province

xiii

379

382 388 392 397 400 401 403 404 406 410 428

429 431 433 434 435 437 438 447 449 451

xiv AI.4. AI.5. AI.6. AI.7. AI.8. AI.9. AI.10. AI.11. AI.12. AI.13. AI.14. AI.15. AI.16. AII.1. AII.2. AII.3. AII.4. AII.5. AII.6. AII.7.

List of Tables Gender composition of individuals in the 2007 urban sample, by province Distribution of households in the 2007 urban sample, by household size and province Distribution of individuals in the 2007 urban sample, by age group and province (%) Educational attainment of individuals over the age of fifteen in the 2007 urban sample, by province (%) Distribution of households in the 2007 rural sample, by province Gender composition of individuals in the 2007 rural sample, by province Distribution of households in the 2007 rural sample, by household size and province (%) Distribution of individuals in the 2007 rural sample, by age group and province (%) Educational attainment of individuals over the age of fifteen in the 2007 rural sample, by province (%) Distribution of households and individuals in the 2007 rural-urban migrant sample, by city Gender composition of individuals in the 2007 rural-urban migrant sample, by city Distribution of individuals in the 2007 rural-urban migrant sample, by age group and city (%) Educational attainment of individuals over the age of fifteen in the 2007 rural-urban migrant sample, by city (%) Provinces and their regional classifications in the CHIP samples, 1988 through 2007 Summary of the 2000 census and 2005 mini-census samples before and after reclassification Composition of the CHIP migrant samples, 2002 and 2007 Population frequency by stratum, 2000 (individuals in the 0.095 percent subsample of the 2000 census) Population frequency by stratum, 2005 (individuals in the 20 percent subsample of the 2005 mini census) Population frequency by stratum, 2000 (households in the 0.095 percent subsample of the 2000 census) Population frequency by stratum, 2005 (households in the 20 percent subsample of the 2005 mini census)

451 452 453 454 455 456 456 457 459 461 461 462 462 467 477 479 481 482 483 484

List of Figures

2.1. 2.2. 2.3. 3.1. 3.2. 4.1. 4.2. 4.3. 4.4. 4.5. 4.6. 4.7. 4.8. 4.9. 4.10. 4.11. 4.12.

China’s National Lorenz Curves for Household Per Capita Income, 2002 and 2007. page 56 Income Levels and Growth by Deciles, 2002–2007. 57 Lorenz Curves of Migrant Per Capita Income, 2002 and 2007. 62 Floor Area of Urban Housing, 1990–2007. 91 Changes in Urban Housing Prices, 1998–2007. 93 Primary Net Enrollment and Middle-School Progression Rates, 1952–2008. 147 One’s Own Years of Education and Average Years of Education of Parents, Total Sample. 162 One’s Own Years of Education and Average Years of Education of Parents, Rural Sample. 163 One’s Own Years of Education and Average Years of Education of Parents, Urban Sample. 164 Regression Coefficients and Correlation Coefficients by Cohort, Total Sample. 171 Regression Coefficients and Correlation Coefficients by Cohort, Rural Sample. 172 Regression Coefficients and Correlation Coefficients by Cohort, Urban Sample. 172 Gini Coefficients of Years of Education by Cohort. 178 Squared Coefficients of Variation of Years of Education by Cohort. 179 Standard Deviation of Education Years by Cohort. 180 Contribution of Parental Education to Inequality in Years of Education by Cohort. 184 Contribution of Parental Education to Inequality in Years of Education by Urban versus Rural and by Cohort (%). 184 xv

xvi 5.1. 5.2. 5.3. 5.4. 5.5. 6.1. 7.1.

7.2. 7.3.

7.4. 7.5. 8.1. 8.2. 8.3. 8.4. 8.5. 8.6. 8.7. 8.8.

List of Figures Average Annual Income Growth from 2002 to 2007 for Decile Groups in the Distribution of Income. Growth in Migrant Employment of Rural Labor. Percentage of Households Reporting Wage Earnings from Migrant Employment, by Decile. Wage Earnings from Migration as a Percentage of Household Per Capita Income, by Decile. Percentage of Individuals in Rural Dibao Households, 2007, by Province. The Distribution of the Number of Migrants and Nonmigrants by the Probability of Migrating (Million). Income Growth Curves for the 1988–1995, 1995–2002, and 2002–2007 Periods (annual income growth at various percentiles). Cumulative Distribution of Income, 1988, 1995, 2002, and 2007. Growth Curves for Individuals Living in Households Primarily Connected to the State Sector, the Private Sector, and Those with No Workers, 2002 and 2007. Growth Curves for Children, Adults, and the Elderly, 2002 and 2007. Growth Curves for Individuals Where the Heads of the Household Have Various Levels of Education, 2002 to 2007. City Employment Rates by Deciles, 1988, 1995, 2002, and 2007. Percentage of Workers among Males between the Ages of Sixteen and Thirty, 1988, 1995, 2002, and 2007. Percentage of Workers among Females between the Ages of Sixteen and Thirty, 1988, 1995, 2002, and 2007. Percentage of Workers among Males between the Ages of Thirty and Sixty-Two, 1988, 1995, 2002, and 2007. Percentage of Workers among Females between the Ages of Thirty and Fifty-Seven, 1988, 1995, 2002, and 2007. Unemployment Rates among Men, by Age, 1988, 1995, 2002, and 2007. Unemployment Rates among Females, by Age, 1988, 1995, 2002, and 2007. Predicted Probabilities of Various Rates of Nonwork among Persons between the Ages of Eighteen and Twenty-Nine, 1995, 2002, and 2007.

205 214 215 216 224 249

262 264

274 275 275 300 302 302 303 303 308 308

312

List of Figures 8.9. 8.10. 8.11. 8.12. 8.13. 8.14. 9.1. 9.2. 9.3. 11.1. 11.2. 11.3. 12.1. 12.2.

12.3. AI.1. AI.2. AI.3.

Predicted Probabilities of Employment and Various States of Nonwork among Persons Aged Fifty, 1995, 2002, and 2007. Percentages of Nonworkers by Decile of Disposable Household Per Capita Income, 1988, 1995, 2002, and 2007. Percentage of Various Categories of Nonworkers by Decile of Disposable Household Per Capita Income, 1988. Percentage of Various Categories of Nonworkers by Decile of Disposable Household Per Capita Income, 1995. Percentage of Various Categories of Nonworkers by Decile of Disposable Household Per Capita Income, 2002. Percentage of Various Categories of Nonworkers by Decile of Disposable Household Per Capita Income, 2007. Average Annual Real Wage Trend for Public and Private Sectors, 1995–2007. Kernel Density Estimations for the Distribution of Income by Ownership Category, 2002 and 2007. Juhn-Murphy-Pierce Decomposition of Log Hourly Wages by Ownership. Wage-Age Profile for Male and Female Workers, 1995, 2002, and 2007. Ln-Wage Levels for Male and Female Workers by Ownership Sector, 1995, 2002, and 2007. Gender-Wage Differential Resulting from the Quantile Analysis, 1995, 2002, and 2007. Real Rate of GDP Growth: China. The Ratio of Minority-to-Han Mean and Median Family-Household Total Incomes in Urban China (1995, 2002, and 2007 CHIP data). Ratio of Minority-to-Han Wage and Salary Incomes. Age-Gender Profiles, Urban. Age-Gender Profiles, Rural. Age-Gender Profiles, Migrant.

xvii

316 319 319 320 320 321 334 344 356 398 399 401 415

418 419 453 458 463

Contributors

(Note: Here and elsewhere, Chinese names follow Chinese convention in which the surname precedes the given name.) Sylvie D´emurger, Researcher, Universit´e de Lyon-CNRS-GATE Lyon SaintEtienne Deng Quheng, Associate Professor, School of Economics and Management, Beijing Normal University, and Institute of Economics, Chinese Academy of Social Sciences Ding Sai, Associate Professor, Institute of Ethnology and Anthropology, Chinese Academy of Social Sciences Bj¨orn Gustafsson, Professor, Department of Social Work, University of G¨oteborg, G¨oteborg, Sweden, and Institute for the Study of Labor (IZA) John Knight, Professor Emeritus, Department of Economics, the University of Oxford, and School of Economics and Management, Beijing Normal University Li Shi, Professor, School of Economics and Management, Beijing Normal University Luo Chuliang, Professor, School of Economics and Management, Beijing Normal University Samuel L. Myers, Jr., Professor, Roy Wilkins Center of Human Relations and Social Justice, Hubert H. Humphrey School of Public Affairs, University of Minnesota Hiroshi Sato, Professor, Graduate School of Economics, Hitotsubashi University xix

xx

Contributors

Terry Sicular, Professor, Department of Economics, University of Western Ontario Song Jin, Assistant Researcher, Institute of World Economics and Politics, Chinese Academy of Social Sciences Xu Jing, Graduate Student, School of Finance, Renmin University of China Yang Juan, Lecturer, School of Economics and Management, Beijing Normal University Yue Ximing, Professor, School of Finance, Renmin University of China

Preface

This book is the product of a long-term research effort supported through the years by many individuals and organizations. In the late 1980s, Keith Griffin and Zhao Renwei brought together a team of Chinese and international researchers to organize the first in a series of nationwide household surveys that are now known as the China Household Income Project (CHIP) surveys. Their goal was to collect household survey data that would make possible meaningful empirical analysis of trends in incomes, inequality, and poverty in post-Mao China. In the mid-1990s, Zhao Renwei and Carl Riskin took the lead in organizing a second round of the survey, and in the early 2000s Bj¨orn Gustafsson, Li Shi, and Terry Sicular organized a third round. In the mid-2000s, the editors of this book, together with Meng Xin, organized a fourth round of the survey. The fourth CHIP survey took place in 2008 and gathered data for the year 2007. This round was carried out in conjunction with the Rural-Urban Migration in China and Indonesia (RUMiCI) project. As in earlier rounds, data collection was closely integrated with research analysis. This book contains analyses of incomes, inequality, and poverty based on the 2007 CHIP survey data; most chapters in this volume also use data from one or more of the earlier rounds. We begin our acknowledgments by expressing gratitude to all those individuals who have contributed to and sustained this long-term body of work. Many of the contributors to this book have been trained and inspired by earlier generations of CHIP researchers, and many of the chapters in this book build upon the work of those researchers. We also thank the organizations that have provided ongoing support for the CHIP over the years. Here the Ford Foundation and the National Bureau of Statistics (NBS) in China deserve special mention. The 2007 CHIP survey would not have been possible without substantial financial support from the Ford Foundation, the National Foundation xxi

xxii

Preface

of Social Sciences of China, the Social Sciences and Humanities Research Council of Canada, and AusAid. Additional funds were provided by the University of Western Ontario, Beijing Normal University, Hitotsubashi University, the Ontario Research Foundation, and the Japan Society for the Promotion of Science. We thank these organizations for their generous support. Data collection and survey work were carried out by the NBS Urban and Rural Household Survey Teams. The NBS also provided helpful advice regarding sampling and survey design. We are grateful to all those at the NBS who contributed to the CHIP, and we extend particular thanks to Chen Xiaolong, Sheng Laiyun, Wang Qi, Wei Guixiang, and Yang Junxiong for their efforts. From the initial design of the 2007 CHIP survey through to the completion of this book, we received helpful advice, ideas, and feedback from many individuals, including Cai Fang, Kathleen Hartford, Lai Desheng, Liu Zeyun, Meng Xin, Scott Parris, Scott Rozelle, Sun Zhijun, Wang Dewen, Wang Meiyan, Wang Sangui, Andrew Watson, Wei Zhong, Xing Chunbing, Yin Heng, Zhao Renwei, Zhao Yaohui, and Zhao Zhong. Meng Xin and her team at the Australian National University made great efforts in conducting the migrant household survey as a part of the Rural-Urban Migration in China (RUMiC) survey project. Deng Quheng, Ding Ning, Ding Sai, Huang Mian, Liu Hongbo, Luo Chuliang, Mao Lei, Mu Cuixia, Song Jin, Xiong Liang, Yang Sui, and Zhou Jin spent an enormous amount of time cleaning the data. We thank these individuals, as well as the anonymous referee, for their contributions. One chapter of this book was published previously in a somewhat different form. Chapter 9, “Do Employees in the Public Sector Still Enjoy Earnings Advantages?” by Yang Juan, Sylvie D´emurger, and Li Shi, is a revised version of “Earnings Differentials between the Public and Private Sectors in China: Exploring Changes for Urban Local Residents in the 2000s,” China EcoC 2012 by Elsevier. This chapter nomic Review, 23 (1), 138–153, Copyright  is reprinted with permission to be reproduced in a modified form. We owe special thanks to Nancy Hearst, who carefully read and edited the chapters, put them in publishable form, and kept track of the many revisions and copyediting during the publication process. We also thank Leslie Kostal for assistance with Web-based aspects of the project. As always, we are indebted to the many households that took part in the CHIP survey. Without their cooperation, this project would not have been possible. Li Shi Hiroshi Sato Terry Sicular

January 15, 2013

Abbreviations

CCP CHIP CI CPI CPPCC FDI FGT FIE GAI MLD MOF NBS OECD PE PIE PIT PITL PPP RUMiC RUMiCI SAT SOE SSB TVE UCE VAT WTO

Chinese Communist Party China Household Income Project concentration index consumer price index Chinese People’s Political Consultative Conference foreign direct investment Foster, Greer, and Thorbecke (poverty index) foreign-invested enterprise government agency or institution mean log deviation Ministry of Finance National Bureau of Statistics Organisation for Economic Co-operation and Development private enterprise private or individual enterprise personal income tax Personal Income Tax Law purchasing power parity Rural-Urban Migration in China project Rural-Urban Migration in China and Indonesia project State Administration of Taxation state-owned enterprise State Statistical Bureau township and village enterprise urban collective enterprise value-added tax World Trade Organization xxiii

Glossary

anju gongcheng () welfare-oriented housing projects bingzhen bingcun () merger and reorientation of townships and villages chengfen () class background chengzhen jumin jiben yiliao baoxian zhidu ( ) basic medical insurance program for urban residents chengzhen zhigong jiben yiliao baoxian zhidu ( ) basic medical insurance program for urban employees chun shouru () net income chuzhong () junior middle school cun tiliu () administrative village levy daiye () waiting for employment daxue benke () four-year college daxue zhuanke () junior/specialized college dazhuan () junior/specialized college dianda/hanshou/yuancheng jiaoyu (//) TV/ correspondence/long distance university dibao () minimum living standard guarantee dishouru () low income duoyu shaoqu fanghuo (, , ) giving more, taking less, and allowing more flexibility fanggai fang () housing-reform housing fangwu chanquan dengji () registration system for housing property feigaishui () local levies replaced by formal taxation fuli fenfang () Mao-era system of subsidized rental housing fupin daohu () poverty alleviation given directly to poor villages and households xxv

xxvi

Glossary

fupin kaifa () rural poverty reduction and development of poor areas gaozhong zhongji (xiao zhongzhuan) () () senior middle technical school (junior middle technical school) gongfei yiliao () government employee health insurance program gongwuyuan () civil servant gouzhi nongji butie () subsidy for the purchase of farm machinery hexie shehui () harmonious society Hu-Wen xin zheng () Hu-Wen new policies hukou () household registration jingji kaifaqu () local economic development zones jingji shiyong fang () economically affordable housing jiti gongyijin () collective welfare fund jiuji kuan () relief funds jumin hukou () unified local resident household registration jumin shenfen zheng () resident identification card ke zhipei shouru ( ) disposable income kexue fazhanguan () scientific outlook on development lanyin hukou () blue stamp household registration laobao yiliao () labor health insurance program laonianren butie () subsidy for the elderly liangmian yibu (  ) exemption from tuition/school fees and subsidy for dormitory fees liangshi butie () food grain production subsidy liangzhong butie () subsidy for improved seeds lianzu fang () subsidized rental housing likai hukou dengji di shijian ( ) how much time since he/she left the place of his/her household registration maiduan gongling () work units buy out middle-aged and older employees with a lump sum related to their cumulative future earnings up to regular retirement minsheng () people’s welfare minzu ( ) ethnic group, nationality nongcun shuifei gaige () rural tax and fee reform nongye chanyehua ( ) industrialization of agriculture nongye, nongcun, nongmin wenti (,, ) agricultural, rural, and peasant problems [see also sannong]

Glossary

xxvii

nongye ziliao butie ( ) agricultural input subsidies nongzi zonghe butie () comprehensive subsidy for agricultural inputs pinkuncun () poor village qiye zhigong jiben yanglao baoxian zhidu ( ) basic pension insurance program for enterprise employees sandai tongtang ( ) three-generation family sannong ( ) agricultural, rural, and peasant problems saomang ban () literacy class shangpin fang () commodity housing shehuihua ( ) socialization shequ () neighborhood community siying qiye () privately owned firms that employed eight or more workers tekun () extreme poverty tekun jiuzhu ( ) subsidies for destitute households tiefanwan () iron rice bowl tongchou chengxiang () integrated and balanced urban-rural development toushui qing, ershui zhong, sanshui shi ge wudidong (, ,   ) the first tax is light, the second is heavy, and the third is a bottomless pit tudi caizheng () land-dependent local public budget tudi gufen hezuozhi () land shareholding system tuigeng huanlin () sloping land conversion waichu renkou () individuals who are members of households in a location and have a household registration in that location but were away wubao () five-guarantee program xiagang () workers who are laid off but keep their ties with the work unit xiagang butie () benefits for laid-off workers xiangcai xianguan ( ) direct administration of township government budgets by county governments xiangzhen tongchou () township levy xiaochanquan zhufang () commodity housing built on rural land without a formal deed to use the land xiaochengzhen () small cities and towns xibu dakaifa zhanl¨ue () western development strategy

xxviii

Glossary

xinxing nongcun hezuo yiliao baoxian () new rural cooperative medical insurance program xinxing nongcun shehui yanglao baoxian () new rural pension system yi xian weizhu ( ) county based yihao wenjian ( ) Document Number One yishi yiyi chouzi (  ) one-issue-one-discussion fee collection yulu jihua ( ) Rain and Dew Program za tiefanwan () smashing the iron rice bowl zai 2002 nian nin zonggong zai chengzhen juzhu shijian duoshao yue? ( 2002   ?) how many months did you stay in an urban area in 2002? zai jiuye peixun ( ) retraining zai xiao xuesheng () enrolled students zanzhu renkou () temporary resident zaotui () early retirement zhaijidi () rural land for housing use zhaijidi zhihuan () the exchange of rural-housing land-use rights for urban commodity housing zhengcun banqian yimin (or shengtai yimin) ( ) or ( ) whole-village migration zhengcun tuijin guihua ( ) comprehensive village-level development program zhenxing dongbei ( ) revival of the Northeast strategy zhiye gaozhong () vocational senior middle school zhongbu jueqi () rise of the central region zhongzhuan, zhiye gaozhong (), () specialized (vocational) senior middle school zhuanye hezuo zuzhi () specialized production cooperative zhufang gongjijin ( ) housing provident fund zhufang gongjijin dixi daikuan (   ) low-interest bank loans for housing zhufang shangpinhua ( ) commercialization of housing zijian ziguan ziyong zimie () individually built, individually owned, individually used, and individually abandoned zili kouliang hukou () household registration with own responsibility for food grain zuidi gongzi () minimum wage zuidi shenghuo baozhang (dibao)  () minimum living standard guarantee

Glossary

xxix

zuidi shenghuo baozhangxian () minimum living standard guarantee line zujin gaige () rent reform zuijin 12ge yue nei, zai waichu wugong jingshang di yigong shenghuole jige yue?   ,   ? how many months have you stayed outside your hometown for work or business?

ONE

Rising Inequality in China Key Issues and Findings Li Shi, Hiroshi Sato, and Terry Sicular

I. Introduction More than three decades have passed since China embarked on economic reform and began to transform its economy from a socialist planned economy to a market economy. During these decades China has experienced rapid growth in gross domestic product (GDP) and personal living standards. Growth, however, has been accompanied by widening income inequality. The rise in inequality has been documented in a wide range of studies by individual researchers, international organizations, and government agencies (Benjamin et al. 2008; Griffin and Zhao 1993; Gustafsson, Li, and Sicular 2008; Ravallion and Chen 2007; Riskin, Zhao, and Li 2001; World Bank 2009; Zhang 2010). The various studies give different estimates of the level of inequality, but the authors all agree that since the 1980s inequality in China has increased markedly. In the context of China’s reforms and ensuing rapid growth, it is not surprising that inequality has increased. Growth is often associated with early stages of economic takeoff, which usually begin in particular sectors and regions. As those leading sectors and regions pull ahead of others, income distribution becomes increasingly uneven. Transition from a socialist planned economy characterized by egalitarian wage and income distribution systems also generates inequality. Markets bring income variation arising from risk and uncertainty, and they differentiate among households and individuals based on productivity, human capital, effort, entrepreneurship, and wealth holdings. In a market system, such sources of inequality can play a positive role in providing incentives for innovation, risk taking, effort, and investment. Although rising inequality is associated with initial stages of growth and transition, it need not continue indefinitely. As an economy develops 1

2

Li Shi, Hiroshi Sato, and Terry Sicular

and matures, growth can spread through various linkages to other sectors and regions (Hirschman 1958). Employment can then expand, leading to rising wages that spill over more broadly, and the benefits of growth can spread. Growth can also make possible improvements in health and education. Government policies potentially play an important role here by strengthening positive linkages among sectors and regions and by investing in health and human capital in ways that promote equality of opportunity. Thus, under the right conditions, ongoing growth can be accompanied by a moderation of or a decline in inequality. In the context of economic transition, whether or not initial increases in inequality are followed by a moderation or a decline in inequality also depends on country-specific factors. Milanovic and Ersado (2008), using data from twenty-six post-Communist countries in Europe and Central Asia, examine the impact of different components of economic transition on income inequality. They conclude that the relationship between transition and inequality depends on the nature of the transition process. For example, they find that large-scale privatization tends to have a dis-equalizing effect, whereas small-scale privatization tends to raise the income shares of the bottom deciles. Differences in inequality outcomes among post-Communist countries to some extent may reflect the degree to which politically advantaged groups have benefited disproportionately, and whether their disproportionate benefits are structural and so persistent. Discussion of this issue in economic sociology is referred to as the “market transition debate” between the new institutional and the corporatist views (Keister and Borelli 2012; Nee and Opper 2010; Walder 2003). According to the new institutionalist view, the development of markets causes the rewards for market-based performance (typically returns to human capital) to increase and the rewards for political capital to decrease. The advantages of political capital persist only in the shrinking state-controlled domain, although institutionalists do acknowledge that state intervention in the market continues in China. Corporatists argue that during the economic transition, state control becomes structurally embedded in the market. Politically advantaged groups reorganize and enhance their vested interests through rent seeking, monopolistic business activities, and the mobilization of social networks (social capital). As the existing literature and some chapters in this volume show, empirical evidence on this question is mixed and depends on the definition of the state domain, how human, political, and social capitals are measured, and what kind of outcome measures are used.

Rising Inequality in China

3

This book examines inequality in China from 2002 to 2007, a period covering most of the first decade of the twenty-first century, up to but not including the World Financial Crisis. This period is significant in several regards. First, in the latter half of the 1990s, inequality-moderating processes had emerged, and China’s income inequalities appeared to be stabilizing (Gustafsson, Li, and Sicular 2008). These trends raise the question of whether or not China had reached a turning point between growth with rising inequality and growth with stable or even declining inequality. Analysis of the trends during the 2002–2007 period provides an answer to this question. Second, in the early 2000s China adopted a new development strategy. In the preceding decades, growth in the “productive forces,” that is, GDP and its underlying inputs, was the primary task, as reflected in a government policy agenda that deliberately and successfully promoted rapid GDP growth.1 By the late 1990s, however, China’s rising inequality had become a cause for concern. After Hu Jintao and Wen Jiabao assumed leadership in 2002–2003, these concerns were articulated and incorporated into official policy. The new development strategy, sometimes referred to as the “Hu–Wen New Policies” (Hu–Wen xinzheng) or the “Scientific Outlook on Development” (kexue fazhanguan), emphasized sustainable and equitable growth. With the new strategy came a set of policy measures designed to reduce disparities and to protect the economically vulnerable, including agricultural support policies, social welfare transfers, targeted tax reductions, minimum wage increases, and increased spending on poverty alleviation. The frequency of keywords such as “scientific outlook on development,” “harmonious society” (hexie shehui), “people’s welfare” (minsheng), “integrated and balanced urban-rural development” (tongchou chengxiang), and “agricultural, rural, and peasant problems” (nongye, nongcun, nongmin wenti, or sannong wenti) in policy documents and Hu’s and Wen’s speeches reflect the direction of the new policies.2 Under the umbrella of this new development strategy, China’s economy continued to grow rapidly; indeed, the size of the GDP pie roughly doubled in the ensuing decade. But what happened to the distribution of that pie? Did inequality continue to rise or was it moderated? Did the benefits of growth trickle down to the poor? Did the equalizing processes that had begun to 1 2

See Jiang Zemin’s report to the 14th National Party Congress, October 12, 1992, at http:// cpc.people.com.cn/GB/64162/64168/64567/65446/4526308.html. Accessed June 3, 2012. See, for example, Hu Jintao’s report to the 17th National Party Congress, October 15, 2007, at http://english.people.com.cn/90001/90776/90785/6290120.html. Accessed June 3, 2012.

4

Li Shi, Hiroshi Sato, and Terry Sicular

emerge in the late 1990s continue? Was the new “harmonious society” policy agenda able to influence the direction and pattern of inequality? In this volume we address these questions from an empirical perspective through an analysis of household survey data. The data are from the China Household Income Project (CHIP), which provides a rich source of nationwide, household-level information about incomes and other variables. The CHIP surveys have been ongoing since the late 1980s, with data collected in four waves: 1988, 1995, 2002, and, most recently, 2007. Earlier waves of the CHIP surveys are the subject of three earlier books (Griffin and Zhao 1993; Riskin, Zhao, and Li 2001; Gustafsson, Li, and Sicular 2008) and have provided rich materials for the study of the development-transitioninequality nexus in China. Of particular note, and a motivation for the CHIP, is that these data sets allow fuller measurement of income and analysis at the household level. Analysis of the 2007 CHIP household survey data, with comparisons to 2002, is the main focus of this volume. Some chapters also examine trends going back to 1995 and 1988. Use of this common data source allows for comparability and consistency across chapters, even though the chapters examine different topics and use various approaches. The first several chapters examine nationwide patterns of inequality. Chapter 2 provides an overview of nationwide trends between 2002 and 2007 in income levels, inequality, and poverty, with special attention to the contributions of the different sources of income, the urban-rural income gap, and inter- versus intraregional income differentials. Chapter 3 analyzes homeownership and its implications for the distribution of housing wealth and income. Here the contributors provide a detailed discussion of estimation of imputed rents from owner-occupied housing, a component of income that is included in CHIP income estimates used elsewhere in this book. Chapter 4 looks at inequality of education and its transmission across generations. Chapter 5 discusses incomes, inequality, and poverty in rural China, and Chapter 6 analyzes the impact of migrants and migration. Chapters 7 through 12 examine aspects of inequality in urban China, including but not limited to the roles of employment and unemployment, gender, ethnicity, the state sector, and the personal income tax. A key finding that emerges is that inequality increased between 2002 and 2007. Thus, the stabilization of inequality in the late 1990s and early 2000s was temporary. Between 2002 and 2007 the benefits of growth were not shared equally: richer groups benefited more than did poorer groups. Nevertheless, we do not find evidence of many losers from the growth process during this period. Both the poor and the rich saw their incomes grow, and poverty declined markedly.

Rising Inequality in China

5

The rise in inequality from 2002 to 2007 raises questions about contributing factors. To what extent did the increase in inequality during this period reflect a widening income gap among regions, or between urban and rural areas? What patterns of inequality arose within urban areas and within rural areas? How did the rapid expansion of migration during this period affect levels of inequality and poverty? What particular institutions or aspects of transition shaped trends in inequality? Were new policies such as the minimum living standard guarantee (dibao) program and tax adjustments effective redistributive mechanisms? These and other questions are investigated in the various chapters of this book. In this introductory chapter we provide general background to set the stage for the chapters that follow; in addition, we draw out key findings and highlight some cross-cutting issues. We begin in Section II with an overview of the key redistributive policies adopted during the Hu-Wen period, to provide a policy context for understanding the inequality outcomes. In Section III we discuss the measurement and definition of income, and in Section IV we describe the main features of the CHIP data sets. Section V contains a discussion of the central findings that emerge from the collection of chapters. This is followed by a concluding section, Section VI, in which we reflect on the findings and identify areas for future research.

II. The Policy Context In the early 2000s China adopted a wide range of policy measures that had implications for equity and income distribution. Here we discuss a selection of the policy measures associated with the Hu-Wen New Policies aimed at distributional concerns. Of course, other policies also had distributional consequences, for example, China’s trade liberalization following accession to the World Trade Organization (WTO) at the end of 2001 and regional development programs such as the western development strategy (xibu dakaifa zhanl¨ue). Some of these and other measures are taken up in later chapters.

A. Social Welfare and Social Security Programs In the prereform era, basic politico-economic units (urban work units and rural collectives) assumed the functions of social security. This institutional framework resulted in the absence of an independent social security system and a large urban-rural disparity in well-being that persisted after the reform. It was not until the 2000s that establishment of a coherent

6

Li Shi, Hiroshi Sato, and Terry Sicular

nationwide social welfare and security system was an important aspect of the party and government agenda, encompassing pensions, medical insurance, unemployment insurance, and minimum living standard guarantee programs. Reform of the urban public pension system began in the early 1990s. Originally, it had been a dual-track system with different pension programs for enterprise employees and employees of public sectors (government, party, and other nonbusiness institutions). A substantial reform that established the basis of the current pension program for enterprise employees was implemented between 1995 and 1997 (Feng, He, and Sato 2011; He and Sato 2013). This reform introduced the Basic Pension Insurance Program for Enterprise Employees (qiye zhigong jiben yanglao baoxian zhidu), which changed the financing scheme from pay-as-you-go to a hybrid of pay-as-you-go and newly introduced compulsory individual accounts. The program applied to employees in all urban enterprises, including stateowned, privately owned, and foreign-owned enterprises. An important redistributive consequence of this 1997 pension reform was an increase in intergenerational inequality in pension benefits.3 Since 2005 the government has repeatedly raised the level of pension benefits for retired enterprise employees. These increases have been prompted by the disparities in postretirement well-being between employees of enterprises and employees in the public sector. The latter are part of a pay-asyou-go system, in which benefits have been increased more rapidly with the rapid growth of government revenue. Consequently, public-sector employees continue to enjoy high pension benefits upon retirement, a legacy of the prereform era. In contrast to the urban areas, in rural areas responsibility for old-age security during much of the reform era has rested with the individuals and the family, except for the small proportion of elderly people entitled to wubao (five-guarantee program) or other social assistance programs.4 During the 2000s, the government experimented with some rural pension 3

4

A study based on the CHIP survey suggests that decreased expectations of pension benefits from the new pension program had the effect of raising the urban household savings rate, especially among the younger generation (Feng et al. 2011). Even though in late 2005 the method of calculating pension benefits was adjusted to make it more actuarial, the basic scheme of the 1997 reform remained unchanged. The wubao program, which dates from the 1950s, provides social assistance for food, clothing, medical care, housing, and burial (and education for children). The elderly, handicapped, and minors who cannot work and have no family members/relatives to take care of them are entitled to wubao. The funds for wubao assistance came from the village collective (and township) until the rural tax and fee reform at the beginning of the 2000s,

Rising Inequality in China

7

programs, but it was not until the late 2000s that a nationwide rural pension program formally appeared on the central government’s agenda. In 2009 the State Council began a pilot program – the New Rural Social Pension Insurance (xinxing nongcun shehui yanglao baoxian) – which originally covered 10 percent of the counties, but has since been expanded. Several chapters in this volume examine the distributional implications of these pension policies. Chapter 7 illustrates the role of pension income in inequality and poverty among urban residents, and Chapter 9 compares earnings, including pension income, among ownership sectors. In the prereform era, urban residents were covered by either the Government Employee Health Insurance (gongfei yiliao) Scheme or the Labor Health Insurance (laobao yiliao) Scheme. Reform of urban medical coverage only began in the late 1990s. At the end of 1998, the State Council launched a new medical insurance scheme called the Basic Medical Insurance Program for Urban Employees (chengzhen zhigong jiben yiliao baoxian zhidu) (State Council 1999). In principle, this program covered all urban workers, including employees of enterprises and the public sector, private business owners, and self-employed people. In practice, coverage gradually spread during the first half of the 2000s. In 2007, to reach out to nonemployed urban residents such as students, the State Council introduced a complementary medical insurance program called the Basic Medical Insurance for Urban Residents Program (chengzhen jumin jiben yiliao baoxian zhidu). Although urban medical insurance covers a broader portion of the urban population than the public pension program, it should be noted that there exists a dual track for civil servants (gongwuyuan) versus other urban residents, and a great divide exists between urban residents with local household registration (hukou) and rural migrants.5 The most important progress in the medical insurance system in the 2000s was the launch of the New Rural Cooperative Medical Insurance Program

5

after which it was funded by fiscal transfers (see the revised regulations, in State Council [2006]). In response, in 2009 the State Council issued an instruction to accelerate equalization of medical care. Regarding public servants, the State Council is to commence reform of the Government Employee Health Insurance Program from 2013. As for the divide between local urban residents and rural migrants, the 2009 instruction emphasized the need to improve interregional transfers of health insurance programs and to absorb rural migrants into medical care programs: the Basic Medical Insurance Program for Urban Employees for migrants with labor contracts, and the Basic Medical Insurance Program for Urban Residents or the New Rural Cooperative Medical System for other rural migrants. See “Implication of the Gradual Reform of Public Medical Care,” Social Security Inquiry Network, at http://www.chashebao.com/yiliaobaoxian/8263.html. Accessed February 6, 2012.

8

Li Shi, Hiroshi Sato, and Terry Sicular

(xinxing nongcun hezuo yiliao baoxian) in 2003 (Central Committee of the Communist Party and State Council 2002). Although a cooperative medical care system had operated at the commune level prior to the early 1980s, it had deteriorated after decollectivization. The New Rural Cooperative Medical Insurance Program was meant to fill the vacuum of medical coverage in rural areas. This program differs fundamentally from the commune-based medical care of the prereform era. First, it is an individual-based voluntary insurance program. Second, the program is financed by subsidies from central and local governments as well as by contributions from participants. Third, it is operated at the county level and hence has a larger risk pool than the earlier commune-based scheme. Fourth, it focuses on in-patient care, whereas the prereform scheme focused on basic public health services, including preventive interventions (Wagstaff et al. 2009). Individual enrollments reached 86.2 percent in 2007 and 96.0 percent in 2010 (NBS, China Statistical Yearbook 2011, Table 21–21). Although the level of rural compensation is still very limited compared with urban compensation, the government’s goal of establishing a basic medical insurance system in the rural areas has been achieved. The development of an unemployment insurance program can be divided into two stages: from the latter half of the 1980s to the late 1990s and then the period thereafter. In 1986 and 1993, the State Council issued administrative regulations for unemployment insurance for employees who were “waiting for employment” (daiye). To cope with the rising number of laid-off (xiagang) urban workers who, due to the restructuring of state-owned enterprises in the late 1990s, kept their ties with the work unit, at the beginning of 1999 the State Council promulgated the Regulations for Unemployment Insurance.6 As a result of these regulations, employees began to make contributions for unemployment insurance, and therefore it can be said that an unemployment insurance program in China was launched at the end of the 1990s, even though the program still was not universal.7 In 2006, a pilot program to extend the usage of the unemployment insurance fund to cover social security for unemployed workers and to promote reemployment (e.g., a subsidy for social security fees and an interest subsidy for loans for business promotion) was undertaken in seven provinces/municipalities in the eastern coastal region (Lai et al. 2011). Chapter 7 in this volume provides 6

7

For the process of SOE restructuring in the late 1990s and its distributive impact, see Li and Sato (2006) based on a 1999 urban household survey that focuses on retrenched/ unemployed workers. For example, in 2007 less than 65 percent of the urban unemployed received unemployment benefits (NBS, China Statistical Yearbook 2008).

Rising Inequality in China

9

further policy background regarding unemployment and a discussion of unemployment rates in the urban areas. After several years of regional policy experiments, the minimum living standard guarantee (zuidi shenghuo baozhang, or dibao) system was established in the late 1990s in urban areas and in the mid-2000s in rural areas. The system provides income subsidies to reduce the disparity between the actual income of the poor and the minimum living standard guarantee line (zuidi shenghuo baozhangxian), which is set at the local level with reference to local costs of living and fiscal capacity. The first pilot urban minimum living standard guarantee program was launched in Shanghai in 1993. By the end of 1996, pilot programs had spread across 101 cities. In 1997, the State Council ruled that minimum living standard guarantee programs be established by 1999 in all urban areas including county seats (State Council 1997). Based on the experiences of the regional pilot programs during the 1990s, in 1999 the State Council promulgated the national Regulations for Urban Minimum Living Standard Guarantees. The proportion of the urban population covered increased steeply in the beginning of the 2000s and stabilized in the latter half of the 2000s (see Table 1.1). It is noteworthy that from the mid-2000s, the central government began to take a comprehensive approach that was intended to combine the minimum living standard guarantee with other social policies, such as medical care, for the urban poor (Lei and Wang 2009). The minimum living standard guarantee in rural areas also appeared on the policy agenda in the mid-1990s. However, it was not until 2007, almost a decade after that for the urban areas, that the State Council announced the nationwide establishment of the minimum living standard guarantee program for the rural population. In 1996, the Ministry of Civil Affairs issued an instruction to promote social security in rural areas. By 2003, 2,037 counties in fifteen provinces (approximately 70 percent of China’s 2,861 counties) had introduced such a program. By mid-2007, all of the thirtyone provincial-level administrative units had implemented the program (Lei and Wang 2009). As shown in Table 1.1, the number of rural people receiving a minimum living standard guarantee increased markedly after 2005. For a description of trends in the incidence of poverty estimated from the CHIP surveys and analyses of the effect of the minimum living standard guarantee system on poverty and inequality, see Chapter 2 (for China as a whole), Chapter 5 (for the rural areas), and Chapter 7 (for the urban areas). These chapters suggest that the minimum living standard guarantee programs in urban and rural China have had only a moderate effect on

10

Li Shi, Hiroshi Sato, and Terry Sicular Table 1.1. Key indicators of redistributive public policies

Public pension programs Number of urban enterprise-sector workers participating in the Basic Pension Insurance Program for Enterprise Employees (million persons) Participation rate in the Basic Pension Insurance Program for Enterprise Employees (%)1 Number of rural residents participating in the New Rural Social Pension Program (million persons)2 Medical insurance Number of urban workers participating in the Basic Medical Insurance Program for Urban Employees (million persons) Participation rate in the Basic Medical Insurance Program for Urban Employees (%)3 Number of urban residents participating in the Basic Medical Insurance Program for Urban Residents (million persons) Participation rate in the Basic Medical Insurance Program for Urban Residents (%)4 Number of rural residents participating in the New Rural Cooperative Medical Insurance Program (million persons) Number of counties that have launched the New Rural Cooperative Medical Insurance Program Proportion of counties that have launched the New Rural Cooperative Medical Insurance Program (%)

1995

2000

2002

2005

2007

2010

87.4

104.5

111.3

131.2

151.8

194.0

50.0

49.7

48.6

51.2

54.9

64.2

–

–

–

–

(51.7)

102.8

7.0

28.6

69.3

100.2

134.2

177.9

4.0

13.6

30.3

39.1

48.6

58.9

–

–

–

–

223.1

432.6

–

–

–

–

36.8

64.6

–

–

–

179.0

726.0

836.0

–

–

–

678

23.7

2451

85.7

2678

93.8

Rising Inequality in China

Unemployment insurance Number of workers participating in the unemployment insurance program (million persons) Participation rate in the unemployment insurance program (%)5 Minimum living standard guarantee Number of residents receiving the urban minimum living standard guarantee (million persons) Number of residents receiving the rural minimum living standard guarantee (million persons) Number of residents receiving conventional rural social assistance (million persons)6 Rural poverty reduction Number of rural poor, measured by the 2008 new official poverty standard (million persons)7 Poverty headcount ratio, measured using the 2008 new official poverty standard (%) Poverty line, according to the 2008 new official poverty standard (yuan, per capita)

11

1995

2000

2002

2005

2007

2010

82.4

104.1

101.8

106.5

116.4

133.8

47.1

49.5

44.5

41.6

42.1

44.3

–

4.0

20.6

22.3

22.7

23.1

–

3.0

4.1

8.3

35.7

52.1

–

–

0.9

10.7

6.1

6.2

–

94.2

86.5

64.3

43.2

26.9

–

10.2

9.2

6.8

4.6

2.8

–

865

869

944

1067

1274

Source: NBS, China Statistical Yearbook, various years. Notes: 1. Number of participants (not including retired workers) divided by the urban working population, not including the self-employed. 2. Numbers for 2007 and 2010 are not comparable. The number for 2010 only includes programs that are officially recognized by the State Council as part of the New Rural Social Pension Program. The number for 2007 includes various local-level pilot programs. 3. Number of participants (not including retired workers) divided by the urban working population, not including the self-employed. 4. Number of overall participants divided by the total urban population. 5. Number of participants (not including retired workers) divided by the urban working population, not including the self-employed. 6. Conventional social assistance includes the “five-guarantee” program (wubao), assistance for households in “extreme poverty” (tekun), and other types of social relief. Temporary relief (e.g., natural disaster relief) is not included. 7. The 2008 new official poverty standard was employed beginning in 2008, at which time the official poverty line was increased to the former “low income” (dishouru) line.

12

Li Shi, Hiroshi Sato, and Terry Sicular

absolute poverty, and they have not substantially reduced relative poverty or income inequality.

B. Employment Policy: Labor-Market Policies and Minimum Wage Regulations An important turning point in China’s urban labor policy came in the late 1990s and the beginning of the 2000s (Lai et al. 2011). Beginning in the late 1990s, a set of passive labor-market policies, such as an income guarantee for retrenched workers, unemployment insurance, and an urban minimum living standard guarantee, were implemented to alleviate the impact of largescale state-owned enterprise (SOE) restructuring on urban unemployment and poverty. Thereafter, a series of positive labor-market policies to promote employment were adopted following the 2002 Sixteenth National Congress of the Communist Party (Ministry of Labor and Social Security et al. 2003; State Council 2005, 2008). According to Lai et al. (2011), these policies covered several areas. First, the policies included skill-training programs for rural migrants and retrenched unemployed workers. Second, they included programs to increase jobs, such as promotion of public investment projects, government financing for job positions (e.g., jobs involved in urban environmental sanitation), wage subsidies for eligible employers to help cover wage costs, tax exemptions for family businesses, and interest subsidies for business loans. Third, public employment bureaus provided job information and placement services. Fourth, development of small enterprises was promoted. The Labor Contract Law, adopted at the beginning of 2008 with the aim of giving employees greater job security, is also worth noting. Minimum wage (zuidi gongzi) regulations were first introduced in 1993 in the Regulations for Enterprise Minimum Wages. In 2004, the revised Regulations for Minimum Wages were enforced by the Ministry of Labor and Social Security (the predecessor of the Ministry of Human Resources and Social Security). The most important point of the 2004 revision is that coverage of the program was extended from enterprise employees to all employees, including nonenterprise institutions and small family businesses. Moreover, the minimum wage level was repeatedly and substantially raised in the 2000s.8 However, because the minimum wage regulations 8

For example, the monthly minimum wages in Shanghai, Guangzhou, and Chengdu increased from 535 yuan, 510 yuan, and 340 yuan in 2002 to 840 yuan, 780 yuan, and 650 yuan in 2007, respectively (Bureaus of Human Resources and Social Security of Shanghai, Guangdong Province, and Chengdu City, http://www.12333sh.gov.cn/200912333/

Rising Inequality in China

13

specify a minimum monthly wage, they have rather limited effect for rural migrants, who tend to work more hours than do local urban workers (Du and Pan 2009). Chapter 6 discusses the influence of the minimum wage regulations on migrant wages.

C. Taxation Reforms Reforms of the personal income tax can be divided into three broad phases. Between the early 1980s and the mid-1990s, the personal income tax was put into place, but the number of taxpayers was very limited. From the mid-1990s to the mid-2000s, the basis of the current personal income tax system was established through integration of relevant regulations and extensions of taxable earnings to cover interest income and revenue from self-employment.9 Starting in the mid-2000s, greater consideration was given to the role of the personal income tax as a policy tool to mitigate income inequality. Repeated adjustments of the amount of the tax exemption, for example, an adjustment from 800 to 1,600 yuan per month in 2006, to 2,000 yuan per month in 2008, and to 3,500 yuan per month in 2011, reflect this consideration.10 The actual redistributive effects, however, have been limited (see Chapters 7 and 10). From the standpoint of promoting the redistributive role of the personal income tax, it has been relatively easy for the government to reduce the tax burden on lower-income groups by adjusting the tax exemption threshold or tax rates; in contrast, it has been difficult to capture the taxable income of higher-income groups and to set an effective, optimal tax rate scale. Policy discussions over further reform of the personal income tax continue. Until the early 2000s, China’s fiscal system in the rural areas was multilayered and decentralized with large interregional disparities in the revenueraising abilities of local governments. Regardless, the provision of most basic public services, such as education, public health, and infrastructure, was assigned to local governments, including township and administrative villages, without adequate intergovernmental fiscal transfers. The distributive

9

10

2009bmfw/zcwd/201101/index.shtml; http://www.gdhrss.gov.cn; http://www.cdhrss.gov. cn. All accessed June 3, 2012). The real increases between 2000 and 2007 deflated by the provincial-level urban Consumer Price Index (CPI) were 45.5 percent, 37.4 percent, and 62.6 percent, respectively (the provincial CPI data are derived from NBS [2010]). The Personal Income Tax Law was first adopted in 1980 and the Personal Income Adjustment Tax Law was added in 1987. The Personal Income Tax Law was amended in 1994 to integrate existing regulations on the personal income tax. See Liu (2011) and State Council (2011a) for related amendments of the Personal Income Tax Law.

14

Li Shi, Hiroshi Sato, and Terry Sicular

consequences of this system were, first, the strongly underinstitutionalized rural taxation system and, second, heavy regressivity in the rural tax and fee burden across both income classes and regions (Bernstein and L¨u 2003; Sato, Li, and Yue 2008).11 To redress this situation, the rural tax and fee reform (nongcun shuifei gaige) was undertaken in 2000. The first step of this reform (2000–2003) involved the substitution of formal taxation (newly defined agricultural taxes) for local levies and fees that had been imposed at the township and village levels (tax-for-fee reform, i.e., local levies replaced by formal taxation, feigaishui). The second step (2004–2005), completed in late 2005 and early 2006, was the gradual elimination of rural taxation, including agricultural taxes.12 The elimination of rural taxation has had a favorable distributive effect, but the overall impact on rural income inequality has not been very great (see Chapter 5).

D. Pro-Rural Policies The recent formulation of pro-rural policies to address “agricultural, rural, and peasant problems” (Wen 2003) can be divided into two phases. The first phase corresponds to the period from the end of the 1990s until 2005. The second phase is after 2006. The essence of the pro-rural policies employed in the 2000s is expressed well in the slogan “giving more, taking less, and allowing more flexibility” (duoyu shaoqu fanghuo), which was outlined in the sixth “Document Number One” (yihao wenjian) in 2004 (Central Committee of the Communist Party and State Council 2004). The main component of “taking less” was the previously mentioned tax and fee reform. Another component was the exemption from tuition/school fees and the subsidy for dormitory fees (liangmian yibu) for primary and junior middle schools, which were implemented in 2006 in the western region and were expanded to the central and eastern regions thereafter. This reform, in combination with the introduction of a county-based education budget system during the first phase, marks a turning point in Chinese basic education. We should 11

12

A common saying in rural China, widely quoted in the press until the early 2000s, was “the first tax is light, the second is heavy, and the third is a bottomless pit” (toushui qing, ershui zhong, sanshui shi ge wudidong). (See, e.g., Cheng, Song, and Dang 2012.) The first tax refers to formal state taxation, such as the agricultural tax, whereas the second tax refers to local levies with a certain legal basis, such as the township levy (xiangzhen tongchou) and the administrative village levy (cun tiliu). The third tax refers to other fees such as various administrative fees, compulsory donations, and fines. Fee collection at the administrative level that should be based on villagers’ democratic discussions, called “one-issue-one-discussion fee collection” (yishi yiyi chouzi), remained even after 2006.

Rising Inequality in China

15

also note that after 2001 the merger and reorganization of primary schools progressed alongside the restructuring of the education budget system, with the number of primary schools in the rural areas decreasing from approximately 440,000 in 2000 to 211,000 in 2010 (Rural Socioeconomic Survey Department of the NBS 2011b: 301). The policies to “give more” have consisted of direct agricultural subsidies for rural households, social security and social insurance programs, and the reinforcement of public investments. The direct agricultural subsidies include a food grain production subsidy (liangshi butie), a comprehensive subsidy for agricultural inputs (nongzi zonghe butie), a subsidy for improved seeds (liangzhong butie), a subsidy for the purchase of farm machinery (gouzhi nongji butie), and various crop and region-specific subsidies. We can also classify the subsidy for the sloping land conversion (tuigeng huanlin) as a direct agricultural subsidy policy. The newly introduced social security and social insurance programs consist of the previously mentioned minimum rural living standard guarantee, new rural cooperative medical insurance, and new rural social pension insurance.13 It should be noted that to guarantee the principles of “taking less” and “giving more,” the central government adjusted the local fiscal and administration systems (Fock and Wong 2007a, 2007b; Wong and Fock 2008). The most important adjustment was the reinforcement of intergovernmental fiscal transfers between the central and provincial governments and within the provinces. In 2000, the central government introduced an intergovernmental fiscal transfer to cover the diminished revenue of the county and township governments following the rural tax and fee reform. In 2005, the seventh “Document Number One” required that no less than 70 percent of the annual increase in the local budgets for education, health, and other public services should be below the county level (Central Committee of the Communist Party and State Council 2005). Another inevitable adjustment was the concentration of fiscal responsibility at the county level. From the beginning of the 2000s, the State Council repeatedly demanded the establishment of a county-based (yi xian weizhu) education budget system to guarantee a certain level of spending for rural education. In 2006, the 13

Using microdata of the NBS official 2002, 2004, and 2006 annual household surveys, Z. Wang (2010) estimated the effects of pro-rural policies, including abolition of rural taxation, agricultural subsidies, and health insurance, on rural income inequality and urban-rural income disparity. Using microdata of the NBS 2002 and 2009 national poverty monitoring household surveys, the Rural Socioeconomic Survey Department of the NBS (2011a: 143–155) estimated the direct effects of “giving more, taking less” policies on household income and expenditure.

16

Li Shi, Hiroshi Sato, and Terry Sicular

eighth “Document Number One” proposed the expansion of the direct administration of township government budgets by county governments (xiangcai xianguan) (Central Committee of the Communist Party and State Council 2006). The reinforcement of public investments is closely related to the preceding reforms in the local fiscal and administration systems. Luo, Zhang, and Deng (2008), using panel data on administrative villages in five provinces from 1998 to 2007, find that the number of public investment projects at the village level decreased in the first half of the 2000s following the tax and fee reforms, but subsequently recovered and increased. Administrative village data from the 2007 CHIP survey, which collected data on public investment projects between 2005 and 2007, also reveal that investment in public investment projects increased in 2007 (Sato and Ding 2012).14 The long-term distributive effects of such projects are a topic for future research.

E. Poverty Alleviation The primary focus of China’s poverty alleviation policies has been the reduction of absolute poverty in rural areas (State Council 2001). Compared with other developing countries, these policies have produced impressive achievements (World Bank 2009). As shown in Table 1.1, the poverty headcount ratio in rural areas, measured as those below China’s official poverty line, decreased from 9.2 percent in 2002 to 4.6 percent in 2007.15 Policies for rural poverty reduction and development of poor areas (fupin kaifa) began in the mid-1980s with policies directed to designated nationaland provincial-level poor counties. To avoid biases in resource allocation caused by measuring poverty simply according to the average income in the target areas, “poverty assistance given directly to poor villages and households” (fupin daohu) began to be emphasized after the 1990s. In 2001, the State Council issued the “Outline of the Poverty Alleviation and Development Policies in Rural China 2001–2010” as the relevant policy framework in the 2000s. According to a recent poverty-monitoring report by the National Bureau of Statistics (NBS), in the 2000s the major poverty reduction policies in rural areas were restructured (Rural Socioeconomic Survey 14

15

The increase in funding for social development programs (primary education and public health) and for western region projects contributed much to the overall increase in public investment projects between 2002 and 2007. The official poverty line was adjusted upward in 2008 by using the former “low-income line” as the new standard of absolute poverty.

Rising Inequality in China

17

Department of the NBS 2011a). New elements included a comprehensive village-level development program (zhengcun tuijin guihua), designating 150,000 administrative villages nationwide as “poor villages” (pinkuncun), of which approximately 108,000 had begun a comprehensive development program including infrastructure construction, industry promotion, and social welfare by the end of 2009. A second policy was a skills-training program to promote employment. The nonagricultural skills-training program, called the “Rain and Dew Program” (yulu jihua), with an aggregate participation of 4 million peasants, began in 2004. In order to improve agriculture, a policy called the “industrialization of agriculture” (nongye chanyehua) was adopted, which aimed to improve agricultural technology, strengthen networks of agricultural producers through specialized production cooperatives (zhuanye hezuo zuzhi), and increase vertical integration of agricultural production by enterprises. Migration- and environmentoriented development policies at the village level including whole-village migration (zhengcun banqian yimin or shengtai yimin), the merger and reorientation of townships and villages (bingzhen bingcun), and sloping land conversion were also adopted.

F. Migration and Hukou Reform Reform of the household registration (hukou) system is one of the continuing and as yet unfinished agendas for both central and local governments (Cai 2011; Chan 2009). From different perspectives, most chapters in this volume illustrate how hukou status directly and indirectly affects the income and well-being of Chinese households. This is because the hukou system creates strong institutional complementarity between hukou registration and employment, education, medical care, housing, social security, land-use rights, and various other opportunities. Thus, the hukou system has been the economic and social institutional basis of the urban-rural divide (Knight and Song 1999; Whyte 2010). Hukou reform in the post-reform era can be divided into three broad stages: the late 1980s to the mid-1990s, the late 1990s to the mid-2000s, and the period from the mid-2000s (Cai 2011; Chan 2009). Hukou reform began in the mid-1980s when peasants (those with rural hukou) were allowed to enter nearby townships with the status of “hukou with own responsibility for food grain” (zili kouliang hukou). At the same time, a system of resident identification cards (jumin shenfenzheng) and regulations for the control of “temporary residents” (zanzhu renkou) were implemented. At the end of the 1980s and the early 1990s, controls on employment or self-employment

18

Li Shi, Hiroshi Sato, and Terry Sicular

of peasants in cities were relaxed. Simultaneously, the food rationing system was gradually abolished. From the mid-1990s, with the rapid expansion of rural-urban migration, several hukou reforms were introduced. The first reform was the decentralization of authority to assign rural-urban permanent migration quotas by changing hukou status. In 1992, the Ministry of Public Security allowed small cities and towns (xiaochengzhen) and local economic development zones (jingji kaifaqu) to introduce local-level urban hukou registration. From the mid-1990s, some coastal cities began to introduce a new local urban hukou registration system called “blue stamp household registration” (lanyin hukou) for eligible rural people (e.g., in 1992 in Wenzhou, in 1993 in Shanghai, and in 1995 in Shenzhen). By the late 1990s and early 2000s, based on such policy experiments, relaxation of permanent migration with a reassessment of hukou status in small cities and towns had expanded nationwide.16 The second reform was the relaxation of hukou status transformation for reunification of family members (spouses, minor children, and the elderly). Such reforms, however, did not bring about fundamental change in the hukou system. The system continued to constrain labor migration and to underpin segmentation between the rural and urban labor markets and within urban labor markets (Cai 2011). This is because local governments facing increased rural-urban migration and rising urban unemployment caused by the restructuring of the SOE sector had incentives to protect local urban workers by segregating migrant workers rather than by promoting labor-market development. Moreover, small cities lacking the resources to provide public services had high entry thresholds and thus did not attract eligible people to settle permanently. Since the mid-2000s, an increasing number of cities have abolished the division between urban and rural hukou by introducing a unified local household registration (jumin hukou). By 2008, thirteen provincial governments had issued provincial regulations or government documents for hukou unification. Unification of hukou registration, however, has not necessarily led to improved access to social security and public services for 16

Ministry of Public Security (2001). The general qualifications for transformation of hukou status were stability in working status, housing ownership, education, and skills. Another important channel was the exchange of rural land-use rights for local urban hukou due to the expansion of urban administrative jurisdictions. Before the Ministry of Public Security document was published local urban hukou status was officially granted to those who invested in businesses or purchased housing in cities (Ministry of Public Security 1998).

Rising Inequality in China

19

holders of agricultural hukou. From this perspective, three new factors deserve attention. Due to the emerging labor shortage and the rising wages for migrant workers in many urban areas, especially coastal cities, local governments have begun to introduce stronger pro-reform incentives (Cai 2011). A second factor is the previously mentioned reform of complementary institutional arrangements, such as a more inclusive social security system. A third factor is the emergence of incentives for local governments to expand urban municipalities. It should be noted, however, that this last factor has the potential to lead to forced hukou reform accompanied by encroachment on rural hukou holders’ rights to land use (see Chapter 3 for a related review of housing). Recent developments suggest that the central government will persist with hukou reform; however, such reform will be gradual. The 2010 “Document Number One” states that migrants with rural hukou should be allowed to settle in small- and medium-sized cities and should be entitled to equal social security and public services as residents with local urban hukou status. Yet, the State Council’s instructions on hukou reform released in February 2012 emphasize that the reform should be conducted step by step according to city size and that large cities should proceed with caution (State Council 2011b). The instruction also warns that the land-use rights of rural hukou holders should not be contravened in the process of reform.

III. Measurement of Income The original motivation to undertake the CHIP was to collect householdlevel information on income and its correlates in order to allow accurate measurement of incomes and a better understanding of the consequences of China’s reforms for inequality. Although China’s NBS has greatly improved its collection of household income data and now publishes extensive descriptive statistics from its household surveys, the NBS does not make its household-level survey data available for research use. In addition, NBS income statistics do not follow international measurement standards. The NBS’s measure of urban income, which it refers to as “disposable income” (ke zhipei shouru) includes employee income (wages, salaries, and other compensation), self-employment income, property income, and transfer income from public and private sources, net of taxes and fees. This measure consists primarily of cash income from various sources; income in-kind is not fully captured. Moreover, imputed rents on owner-occupied housing are not included. The NBS measure of rural income, which it refers to as “net income” (chun shouru), includes cash income from employment,

20

Li Shi, Hiroshi Sato, and Terry Sicular

self-employment, and household business and production activities such as farming, as well as the monetary value of self-produced consumption, net of production costs, taxes and fees, and depreciation of productive assets. Net private transfers are also included (NBS 2008). As in the case of urban income, the NBS measure of rural income does not include imputed rents on owner-occupied housing. Khan et al. (1992) have pointed out the shortcomings of the NBS’s income measures. One shortcoming is the exclusion of imputed rent of owneroccupied housing. Another is the understatement of consumption subsidies, mostly to urban households. In the past, consumption subsidies arose due to the provision of low-priced consumer goods and subsidized rental housing under the planned economy. Since the early 1990s, however, the planned allocation of consumer goods, and thus the consumption subsidies, has been largely eliminated. Similarly, the urban housing privatization, which was basically completed by the early 2000s, led to a substantial reduction in, but not the complete elimination of, subsidized urban rental housing. In light of these considerations, and following in the spirit of Khan et al. (1992), we have adopted an alternative, more complete measure of income than that used by the NBS. This income measure, referred to in this volume as “CHIP income,” is equal to NBS income plus imputed rents on owner-occupied housing plus implicit subsidies on subsidized public urban rental housing.17 The rest of CHIP income is the same as the disposable income/net income measure used by NBS. Many, but not all, chapters in this book use CHIP income in their analysis. Some use NBS income, and some, such as Chapter 2, use both to allow comparisons with other studies in the literature, which mainly use NBS income. Most chapters in this book analyze inequality among individuals rather than among households. Each household member is treated as an observation, which gives more weight in the analyses to larger households. Individual income is calculated as household income divided by the number of household members. This simple treatment is common in both international studies and in studies of China. Some researchers recommend the use of equivalent scales, in which different members of the household are given different weights (e.g., children are assigned a weight of 0.5 and so are “equivalent to half an adult”), so as to adjust for differences in consumption among heterogeneous household members (for example, Chen 2006). Equivalent scales are not used here due to the lack of agreed-upon standards for Chinese households, and because most comparable studies 17

The procedure for estimating imputed rents is explained in Chapter 3. Implicit subsidies are equal to the difference between the market rent and actual rent paid on rented housing by urban households that live in subsidized public housing, as reported by those households.

Rising Inequality in China

21

do not adopt them. Analysis of inequality without use of equivalent scales may lead to some overestimation of income inequality. Neither NBS income nor CHIP income fully includes the market value of social welfare and public spending programs. Both income measures include income received through cash-transfer programs, such as unemployment insurance, the minimum living standard guarantee program, and grain production subsidies to farmers. Currently received pension income is also included, but employer contributions to pension and social insurance programs for employees are not. Similarly, the value of subsidized social programs and services such as education are not included. Some researchers have argued that the market value of such programs should be estimated and included as an income component. Due to difficulties in collecting the relevant information and differing views about what should be included and how it should be computed, as well as the fact that most international studies do not include such types of income, this study does not incorporate the market value of social welfare and public spending programs. The exclusion of these income components may cause an underestimation of nationwide income inequality, as low-income groups such as rural and migrant households are not entitled to, or receive lower benefits from, these programs.18 A topic of considerable discussion in recent years is the underreported income of high income groups in urban China, sometimes referred to as “grey income” (Luo, Yue, and Li 2011; X. Wang 2010). This situation is not unique to China. Household survey samples are often characterized by underrepresentation of high-income groups, and certain types of income that tend to be received by higher-income households are typically underreported. With the growth of private businesses and privately owned assets in China, and with the emergence of an ultra-rich segment of society, these problems have become relevant for both the NBS household surveys and the CHIP surveys. The consequences for the estimation of inequality are obvious. Although some studies have proposed methods to solve this problem, here we do not attempt to do so.19 Estimates of inequality reported in this volume should be understood with these considerations in mind. 18

19

For example, Li and Luo (2010) estimate the contribution of social security programs to the income gap between urban and rural households and to national income inequality in China. Their estimates indicate that in 2002 the income gap between urban and rural households would have increased by nearly 40 percent and the Gini coefficient of the national income inequality would have increased by 11 percent if the market value of social security programs had been included in household income. For example, Li and Luo (2011) attempt to combine the wealth data of the richest billionaires with the CHIP data to recompute income inequality in China.

22

Li Shi, Hiroshi Sato, and Terry Sicular Table 1.2. Coverage of the CHIP 2002 and 2007 surveys 2002 survey

Individuals Households Provinces

2007 survey

Urban survey

Rural survey

Rural-urban migrant survey

Urban survey

Rural survey

Rural-urban migrant survey

21,696 6,934 12

34,719 7,998 22

5,327 2,005 12

29,262 10,000 16

51,847 13,000 16

8,404 4,978 9

Sources: Appendix I of this volume; Li et al. (2008).

IV. Data and Surveys Most chapters in this book use data from the 2002 and 2007 CHIP surveys. Appendix I provides a detailed explanation of the 2002 and 2007 CHIP samples and sampling methods.20 Here we provide a brief overview. The 2002 and 2007 CHIP surveys contained three subsamples: a sample of households living in rural villages, a sample of households with rural hukou living in urban areas (rural-urban migrant households), and a sample of households with urban hukou living in urban areas (urban local households). In 2002, the surveys for these three subsamples were conducted by the NBS. In 2007, the NBS conducted the urban and rural surveys, and an independent survey company carried out the rural-urban migrant survey. It should be noted that the 2007 CHIP surveys were carried out jointly and in conjunction with the RUMiCI (Rural-Urban Migrants in China and Indonesia) survey project.21 The provincial coverage and sample sizes of the 2002 and 2007 surveys are shown in Table 1.2. Although the coverage and sample sizes differ in the two years, in both years the samples were constructed to represent four geographic regions: megacities with provincial administrative status, and 20 21

For discussion of the CHIP survey samples from earlier years, see Eichen and Zhang (1993), and Li et al. (2008). In addition to samples of 5,000 rural-urban migrant households and 8,000 rural households and 5,000 urban households covered jointly with the RUMiCI surveys, the CHIP survey samples also included an additional 5,000 rural households and 5,000 urban households that were not included in the RUMiCI surveys. Because the RUMiCI surveys are concentrated in the eastern and central regions, the additional households in the CHIP surveys were located mainly in the western and central regions. The sampling procedure and survey method for the 2007 migrant survey are described in detail in Rural-Urban Migration in China and Indonesia Project Survey Documentation. See http://rse.anu.edu.au/rumici/documentation.php. Accessed July 17, 2012.

Rising Inequality in China

23

the eastern, central, and western regions. The provincial distribution of the sampled households in the surveys can be found in Li et al. (2008) and Appendix I of this volume. Provincial coverage in the two years overlaps considerably. All twelve provinces in the 2002 urban survey are also included in the 2007 urban survey, and all sixteen provinces in the 2007 rural household survey are also in the 2002 survey. For the migrant survey, eight of the nine provinces in 2007 are also included in 2002.

A. The Urban and Rural Household Samples The samples in the CHIP surveys for urban local households and rural households were drawn from the NBS household survey samples. The NBS household surveys cover both rural and urban areas and all the provinces of mainland China. The sample sizes of the NBS surveys were 45,317 urban and 68,116 rural households in 2002 and 59,305 urban and 68,190 rural households in 2007.22 The sampling methodology of the NBS urban and rural household surveys is described in Li et al. (2008); Appendixes I and II of this volume provide details regarding the CHIP urban and rural household samples.

B. The Migrant Samples Rural-urban migrants are not adequately captured in the NBS urban and rural household surveys, and thus they are not adequately captured in the CHIP urban and rural household surveys. In view of the increasing importance of migration in China, in 2002 the CHIP began to include a separate survey of rural migrants living in cities. In this section we discuss the main features of the 2002 and 2007 migrant surveys, with attention to probable sample biases due to the transient, informal, and diverse situations of China’s migrant population. Despite some shortcomings, the CHIP migrant surveys in combination with the urban and rural surveys allow for a fuller picture of incomes and inequality in China than is otherwise possible. Appendix I and Appendix II provide additional discussion of the migrant samples. The 2002 migrant survey includes 2,000 rural-urban migrant households. These households were selected from Beijing and from the capital city plus one medium-sized city in each of the twelve provinces included in the 2002 urban sample, with 200 households selected in Beijing and in 22

See NBS, China Statistical Abstract (2003: 102, 107), and NBS, China Statistical Abstract (2008: 104 and 110).

24

Li Shi, Hiroshi Sato, and Terry Sicular

each of the provinces in the eastern and central regions, and 150 households in each of the provinces in the western region. Within each province, 100 households were selected in the capital city, and the rest were selected in the medium-sized city. The households were selected without regard to whether the individuals had registered in their place of residence, so the sample includes both registered and nonregistered migrants. Within cities, rural-urban migrant households were selected using residential neighborhoods as the sample frame; hence, migrant workers living at construction sites or in factories were less likely to be included in the sample. The exclusion of these migrant workers implies that the 2002 migrant sample was biased toward more permanent migrant households and contained fewer single or solo migrants. Also, because the survey was largely carried out in urban neighborhoods, it does not capture migrants living in suburban areas. City statistical bureaus were instructed to determine the method used to select households within neighborhoods according to local conditions, so the sampling method may have differed from one city to another. It was required that the selected households had lived in the community for more than six months and not have a local urban household registration (hukou). The 2007 migrant survey differs from the 2002 survey in terms of sample size, provincial coverage, and design of the questionnaire. Changes were made in these regards to address some of the shortcomings of the 2002 migrant survey. The 2007 migrant survey includes 5,000 rural-urban migrant households from fifteen cities in nine provinces: Guangzhou, Shenzhen, and Dongguan in Guangdong; Shanghai; Nanjing and Wuxi in Jiangsu; Hangzhou and Ningbo in Zhejiang; Wuhan in Hubei; Chongqing; Chengdu in Sichuan; Hefei and Bengbu in Anhui; and Zhengzhou and Luoyang in Henan. The choice of provinces was based on the major sources and destinations of migrants. The sample provinces include the largest inmigrant and out-migrant provinces, and they account for over 70 percent of the national migrant population. As in the 2002 survey, within each province the sample covers the provincial capital city plus one or two medium-sized cities in which rural-urban migrants are more concentrated. Selection of migrant households within each city followed a new method based on a geographic grid. The first step was to define the city boundaries. Most Chinese cities contain two types of administrative districts, urban districts and rural counties. The city boundary was defined as covering all urban districts; rural counties were not included. The second step was to divide each city into a grid. Then cells of the grid were randomly drawn. The third step was to conduct a census of employers and the self-employed in each

Rising Inequality in China

25

selected cell of the grid to obtain information about the population of ruralurban migrants working in each cell. The fourth step was to select randomly employers and to set a sample size for each selected employer according to the number of migrants employed by that employer. Migrant worker samples were randomly drawn from among the employees of each of the selected employers. The final step was for the enumerators to visit and interview the families of these sampled employee and self-employed migrants.23 Although this sampling procedure addresses some of the shortcomings of the 2002 sampling method, it still has some potential biases. One is that because the census was conducted among employers and the self-employed, unemployed migrants were excluded from the survey frame. Unemployed migrants who were members of the households of the selected employed or self-employed migrants are included in the sample, but nevertheless, it is likely that the sample undercounts unemployed migrants. This may lead to an upward bias in the average income of migrant households, but the magnitude may not be significant. Migrant workers who are unemployed are likely to choose to return to their home villages, so relatively few unemployed migrants would reside in the cities.24 Another bias is an underrepresentation of migrant workers in construction, since workers on infrastructure projects, such as airports, highways, and railways, were less likely to be captured by the sampling grid. This may cause a downward bias in the average income of migrant workers, since workers employed in the large construction projects tend to have relatively stable jobs and higher wages. Finally, as with the 2002 migrant survey, the sampling for the 2007 survey did not cover suburban areas; however, the 2007 survey identified migrants based on place of employment, not place of residence. Therefore, it is possible that the 2007 survey contains migrants who lived in suburban areas if they (or a member of their household) commuted to the city to work. 23

24

A more detailed description of the sampling procedure can be found in Rural-Urban Migration in China, 2008, at http://idsc.iza.org/?page=86&wid=778#documentation. Accessed July 17, 2012. A description of the method used to carry out the census of employers is available in the Rural-Urban Migrant Project Census Manual, 2007, at http://rse.anu.edu.au/rumici/pdf/Census%20manual China English08.pdf. Accessed August 2, 2012. For example, 15 to 20 million rural migrant workers had lost their jobs in urban China at the end of 2008 due to the impact of the international financial crisis. It was reported that about 10 million had returned to their hometowns; see “Nongmin gong shiye diaocha” (Investigation on Unemployment of Rural Migrant Workers), January 19, 2009, http:// finance.qq.com/a/20090120/000698.htm. Accessed July 17, 2012.

26

Li Shi, Hiroshi Sato, and Terry Sicular

We note that the NBS rural survey also includes a subset of rural-urban migrants. The NBS’s rural sampling frame excludes households that have moved in entirety to urban areas, but it includes rural households that contain members who are engaged in migration and migrant work. Members of rural households who are away for more than six months and who do not have a strong economic connection with their household of origin are not included as household members; members who are temporary migrants (away for six months or less) or who are away for more than six months but are substantial economic contributors to their rural families are included. The CHIP rural survey, as a subsample of the NBS survey, follows the same rules. It is likely that the income of migrant household members is understated in the NBS survey, because their income is reported by the household respondents resident in the rural areas, who may not be fully informed about the economic activities and income of absent migrant family members. The CHIP migrant survey does not exclude temporary migrants or migrants who are substantial economic contributors to their rural households. Consequently, the potential for double counting arises in analyses that combine the CHIP migrant and rural surveys. To resolve this problem, as described in Appendix II, we have identified the single and temporary migrants in the CHIP migrant sample. Because such migrants are likely also covered in the rural sample, they are excluded from the migrant sample in analyses that combine migrant and rural samples. Thus, analyses of national inequality in this volume (as in Chapter 2) generally use the subsample of the migrant survey that consists of long-term, permanent migrants who do not maintain a strong economic link with their rural households of origin.

C. Questionnaire Design, Variables, and Sources of Data The CHIP urban and rural survey data come from two sources. The first source is the NBS household survey data, which contain information on personal characteristics of individuals and households, such as age, gender, ethnic minority status, educational attainment, employment status, household size, residence location, and detailed information on household income and expenditures. Income information includes household wage income, business income, property income, and transfer income, and expenditures include spending on food, clothing, transportation and communications, daily consumer goods, and housing maintenance. The second source is the independent CHIP survey, which collected information from each of the households using a separate, independently designed questionnaire. The CHIP questionnaire contained additional

Rising Inequality in China

27

questions relevant to the research objectives of the CHIP project. For instance, the CHIP questionnaires contained detailed questions about employment/unemployment, job mobility, and the work time of individuals. They also contained relevant questions about market values and market rents of owner-occupied housing to permit estimates of housing subsidies and imputed rents on owner-occupied housing (see Chapter 3). The NBS does not conduct a separate household survey of rural-urban migrants. Thus, all information in the CHIP migrant survey data set was collected using independently designed questionnaires. The migrant survey questionnaires were designed to collect information that is consistent with and comparable to the information collected in the urban and rural household surveys, and a majority of the questions in the migrant questionnaire are the same or similar to those in the other surveys. In addition, the migrant surveys contained detailed questions on the components of income and expenditures, to allow for the calculation of income and expenditure values that are consistent with those supplied to the CHIP by the NBS for urban and rural households.25 Although the information collected in the different waves of the CHIP survey is reasonably consistent over time, changes have occurred. The reforms have substantially altered the context in which households operate, necessitating the elimination or modification of some survey questions as well as the addition of others. The changing context has also necessitated change in the design of the survey, most notably the inclusion of a separate rural-urban migrant survey as well as new strategies for collecting information from urban and rural households, who now live increasingly diverse and complicated economic lives. The CHIP surveys have been affected by budgetary, political, and time constraints, which have also changed over time, leading to some differences in the amount and types of information collected. Finally, although there has been substantial continuity in the CHIP project team and its research objectives, over the years the team and the objectives have evolved, with consequent implications for the content and design of the surveys.

D. Weighting The CHIP regional and provincial sample sizes, as well as the urban local, rural, and rural-urban migrant sample sizes, are not proportional to the 25

The 2007 CHIP questionnaires are available at http://rse.anu.edu.au/rumici/documen tation.php, where they are referred to as “first wave” or 2008 questionnaires. Accessed July 17, 2012.

28

Li Shi, Hiroshi Sato, and Terry Sicular

Table 1.3. Comparison of CHIP and NBS household per capita incomes, 2002 and 2007

2002 (yuan)

2007 (yuan)

Average annual growth (%, constant prices)

4,221 4,617 4,140 1.020

6.96 7.44 7.51 0.93

15,469 17,639 13,786 1.122

11.26 11.77 9.77 1.15

Rural CHIP Rural Survey Data NBS income definition CHIP income definition Published NBS Statistics CHIP NBS / Published NBS

2,590 2,771 2,476 1.046 Urban

CHIP Urban Survey Data NBS income definition CHIP income definition Published NBS statistics CHIP NBS / published NBS

8,078 9,002 7,702 1.049

Notes: Incomes are in current prices. CHIP incomes are calculated with weights, and average annual growth is calculated using constant prices deflated using the NBS consumer price indices. CHIP numbers are from Chapter 2. Urban incomes are based solely on the CHIP urban household survey and do not incorporate rural-urban migrant households from the migrant survey. The published NBS income statistics and consumer price indices are those published by the NBS. Rural income is net household income per capita; urban income is disposable household income per capita.

regional, provincial, and urban/rural/migrant shares in the Chinese national population. Thus, weights are needed in order to obtain results that are representative. Appendix II of this book provides a detailed discussion of the weights and related issues.

E. Comparison with NBS Income Statistics Estimates of household incomes based on the CHIP survey data and published in this volume differ from the household income statistics published by the NBS. Table 1.3 provides estimates of per capita rural and urban household income based on the CHIP survey data with comparisons to published NBS income statistics. In this table, urban incomes are calculated using the CHIP urban household survey and do not incorporate data from the migrant survey, so that they are comparable to the NBS urban household survey data. One reason for the difference between the CHIP and NBS income estimates is the income definition. Income estimates calculated using the CHIP definition of income are higher than, and rose faster than, income estimates

Rising Inequality in China

29

calculated using the NBS definition (Table 1.3). The differences between these two income measures are mainly due to the CHIP inclusion of imputed rental income on owner-occupied housing, which between 2002 and 2007 grew faster than other income components. Income definition, however, does not explain the entire difference. This can be seen if we compare the NBS-published income statistics with the mean estimates of NBS income calculated using the CHIP survey data. As shown in Table 1.3, in 2002 the CHIP survey data yielded NBS incomes that are, on average, about 5 percent higher than those published by the NBS for both urban and rural areas. In 2007, the CHIP survey data yielded rural NBS income that is 2 percent higher than that published by the NBS, whereas the discrepancy for urban income is larger, at 12 percent. As the CHIP urban and rural survey samples are drawn from the larger NBS urban and rural household surveys, and as NBS income data in the CHIP surveys are provided to the project by the NBS, these differences are almost certainly due to sampling (and perhaps also to choice of weights). Statistical analysis indicates that the differences are not significant. We cannot reject the hypothesis that the mean urban and rural NBS incomes calculated using the CHIP data statistically are not different from those published by the NBS. In principle, whether our estimates of inequality are higher or lower than those based on the full NBS samples will depend on both the means and the dispersion around the means of the different samples. As mentioned earlier, our statistical tests indicate that the CHIP sample means are not statistically different from the published NBS income levels. Gini coefficients calculated using the CHIP samples are also consistent with NBS estimates. The NBS released estimates of national Gini coefficients for 2003 through 2012 for the first time in 2013. The estimate for 2003 is 0.479, and for 2007 it is 0.484 (Xinhua News Agency 2013). These numbers are comparable to our estimates of the Gini coefficients based on the CHIP data, using NBS income and excluding long-term rural-urban migrants, of 0.456 in 2002 and 0.481 in 2007 (Table 1.4). The NBS publishes the Gini coefficients for rural and urban areas separately, calculated using its full rural and urban household survey samples (Zhang 2010). The NBS estimates of the rural Gini coefficient are unchanged at 0.37 in both 2002 and 2007, as compared to Luo and Sicular’s estimates (Chapter 6), calculated using the CHIP data and NBS income, which are unchanged at 0.36 in both years. The NBS estimates of the urban Gini coefficient are 0.32 in 2002 and 0.34 in 2007. The estimates in Chapter 7, calculated using CHIP data and CHIP income and without incorporating migrants, are 0.30 in 2002 and 0.32 in 2007.

30

Li Shi, Hiroshi Sato, and Terry Sicular

IV. Major Findings The chapters in this volume address a range of topics related to inequality in China. Here we discuss the major findings that emerge from the chapters, with special attention to crosscutting themes. In this discussion we refer to Table 1.4, which summarizes the key indicators of inequality and poverty in the various chapters. Finding 1: Inequality continued an upward trend. Between 2002 and 2007, income inequality in China increased. Table 1.4 shows this volume’s estimates of the Gini coefficient of household per capita income. The estimates differ somewhat depending on the definition of income, treatment of imputed rents, correction for regional differences in the cost of living, and treatment of migrants, but in all cases, inequality increased between 2002 and 2007. In most cases the size of the increase is 5 to 6 percent; for the purchasing power parity, or “PPP,” Gini coefficient, which is calculated using incomes adjusted to correct for regional price differences, the increase is more than 8 percent. In 2007 all but one estimate of China’s Gini coefficient exceeds 0.48. This level of inequality places China among the top 25 percent of countries worldwide ranked by degree of inequality. That is, only one-quarter of countries had Gini coefficients as high as or higher than China.26 Our estimates of China’s Gini coefficient for 2007 place China as the most unequal country in Asia. Inequality in Asia tends to be low, with Gini coefficients generally lower than 0.40. Aside from China, Asia’s highest Gini coefficients are about 0.45 for the Philippines and Malaysia.27 In 2002 China’s Gini coefficient was similar to that for the Philippines and Malaysia, but by 2007 China had moved beyond them. The only estimate of China’s Gini coefficient that is substantially below 0.48 is the PPP estimate, at 0.43. Even this level is moderately high by international standards. The PPP Gini coefficient is not directly comparable to the estimates for other countries reported here, which do not correct for regional price differences. Although inequality in China increased from 2002 to 2007, the Hu-Wen New Policies appear to have had some positive distributional effects. As discussed in Chapter 6, for example, the expansion of migration following 26 27

Based on Gini estimates reported by the World Bank, http://data.worldbank.org/indicator/ SI.POV.GINI/. Accessed July 17, 2012. Gini estimates reported by the World Bank, http://data.worldbank.org/indicator/SI.POV. GINI/. Accessed July 17, 2012.

Rising Inequality in China

31

Table 1.4. Key indicators of inequality and poverty in China 2002 Gini, CHIP income per capita Including migrants Including migrants, alternative estimates of imputed rents PPP, including migrants Excluding migrants Gini, NBS income per capita Excluding migrants Urban-rural CHIP income ratio Including migrants Including migrants, alternative estimates of imputed rents PPP, including migrants Excluding migrants Urban-rural NBS income ratio Excluding migrants Urban-rural income gap’s contribution to national inequality (GE(0)) Including migrants PPP, including migrants Contribution of between region inequality to national inequality (GE(0)) Including migrants Including migrants, PPP Asset income’s share of income Asset income’s contribution to national inequality Poverty rate, absolute (PPP $1.25/day) Poverty rate, relative (50% of median)

2007

2007/2002

0.460 0.464

0.483 0.492

1.050 1.060

0.391 0.462

0.423 0.487

1.082 1.054

0.456

0.481

1.055

3.20 3.30

3.80 4.06

1.188 1.230

2.17 3.25

2.68 3.82

1.235 1.175

3.16

3.66

1.158

44.5% 27.1%

50.9% 37.5%

1.144 1.384

17.6% 11.6% 7.2–9.0% 7.6–10.3%

15.5% 11.3% 11.7–16.0% 13.1–19.2%

0.881 0.974 1.63–1.78 1.72–1.86

18.6% 13.2%

8.0% 13.3%

0.444 1.008

Note: Estimates provided in this table are taken from Chapter 2. All estimates are calculated using weights. Except for estimates of NBS income, income includes imputed rents on owner-occupied housing and subsidies on subsidized urban rental housing. Except where noted otherwise, rural imputed rents are calculated using the rate of return on housing value and urban imputed rents are calculated using estimated rental value. The alternate estimates calculate imputed rents using the rate of return on housing value for both urban and rural households. See Chapter 3 for discussion and details.

the easing of restrictions on mobility, agricultural support policies, and the rural dibao program benefited lower-income and rural groups. Findings in Chapters 7 and 8 indicate that in urban China, government transfer programs and employment policies may also have had positive distributional consequences. These equalizing influences, however, were more than offset

32

Li Shi, Hiroshi Sato, and Terry Sicular

by dis-equalizing factors, which included the widening urban-rural gap and growth of income from assets and property. Finding 2: A widening urban-rural income gap contributed to the rising inequality. China’s urban-rural income gap has been widening since the late 1980s, and by 2002 per capita urban household incomes were, on average, more than three times rural incomes. The already large urban-rural income gap widened by roughly 20 percent during the 2002–2007 period, so that by 2007 urban incomes were 3.7 to 4.0 times those in rural areas (Table 1.4). When calculated using PPP incomes, the size of the urban-rural income gap is smaller, but the increase in the gap between 2002 and 2007 is even larger. The widening gap between incomes in urban and rural China contributed to the rise in national inequality during this period. In 2002 the gap was associated with 45 percent of overall inequality (27 percent in PPP terms); in 2007, it was associated with 51 percent of overall inequality (38 percent in PPP terms; Table 1.4). Thus, the urban-rural income gap remains a central factor underlying national income inequality. What explains China’s large and widening urban-rural income gap? Although the chapters in this volume do not directly investigate this question, they provide some relevant information. First, the widening gap during the 2002–2007 period was not due to stagnant rural incomes, but rather to faster – indeed very rapid – growth in urban incomes. During this period, rural incomes grew at an average annual rate of more than 7 percent, but urban incomes grew even more rapidly at 11 percent (Chapters 2, 5, and 7). This raises the question of why urban incomes increased so much more rapidly than rural incomes. Of particular note in this regard are the very rapid increases in certain urban nonemployment income components such as imputed rents on owner-occupied housing, pensions, and transfers (Chapters 3 and 7). By 2007, imputed rents and pension income together accounted for 30 percent of urban income, up from 23 percent in 2002 (see Chapter 7, Table 7.4). For rural households, however, pension income was basically nonexistent and imputed rents accounted for only 6.5 percent and 8.6 percent of income in 2002 and 2007, respectively (see Chapter 5, Table 5.2). To a large extent, these components of income are associated with government policies and programs that have disproportionately benefited the urban population. Second, the widening gap has a distinct regional dimension. Excluding large municipalities such as Beijing and Shanghai, the urban-rural PPP

Rising Inequality in China

33

income ratio rose by a remarkable 43 percent in the East, as compared to 27 percent in the Center and only 3 percent in the West (Chapter 2). The reason for these regional differences merits further investigation. Additional research is also needed to analyze the extent to which the urban-rural income gap reflects differences between urban and rural household characteristics versus differences in the returns to those characteristics. In this regard, urban-rural differences in levels of education and in the returns to education likely play a role. Chapter 4, examining inequality in education, finds that differences in educational opportunities and outcomes between urban and rural areas have persisted. Studies using the 2002 CHIP data found that education was an increasing source of inequality and contributed to the urban-rural income gap (Gustafsson, Li, and Sicular 2008; Sicular et al. 2007). Finding 3: Intraregional, not interregional, income differentials were the main source of national inequality. As discussed earlier, the provinces covered in the CHIP surveys fall into four distinct regional groups: megacities with provincial status, East, Center, and West. Although the income gap between the richest regional group, megacities with provincial status, and the poorest regional group, the western provinces, remained large, differences in average incomes among the four regions contributed a relatively small and declining share of national inequality. As reported in Chapter 2, the contribution of interregional differences fell slightly between 2002 and 2007, to about 11–12 percent of national inequality (in PPP terms). Analysis of earlier CHIP survey data finds an increase in the contribution of regional inequality between 1988 and 1995, followed by a decline between 1995 and 2002 (Gustafsson, Li, Sicular, and Yue 2008), a continuing trend, as suggested by our findings. The decline in the importance of regional inequality and the relatively stable if not declining interregional income gaps suggest that China’s efforts to reduce inequality through regional development policies have had some success. Of course, the declining contribution of interregional income differentials means that the contribution of intraregional inequality has grown. By 2007 inequality within regions contributed well over 80 percent of the national inequality. Within-region inequality rose markedly from 2002 to 2007 for all regions except the megacities. Inequality is particularly high within western and eastern China. Addressing intraregional income differences may require new policies and strategies at both the national and regional levels.

34

Li Shi, Hiroshi Sato, and Terry Sicular

Finding 4: Income from household assets emerged as a new, dis-equalizing factor. By 2007, Chinese households had become owners of substantial amounts of private property. Perhaps the main factor leading to the rise in private household property ownership was the privatization of urban housing, which began in the 1990s and was more or less completed by the mid-2000s. Several other factors are also relevant, in particular the growth in household savings and the increasing diversification of household investments into not only housing but also financial assets and family and private businesses. Household assets generate income – interest earnings, dividends, rents, and capital gains. These components of income are likely to be underreported in the CHIP survey data, and the degree of the underreporting is probably correlated positively with income. Thus, our estimates of inequality from asset wealth may be understated. Another important component of property income is imputed rent on owner-occupied housing. Although imputed rent on owner-occupied housing is not included in the official NBS measure of household income, it is included in CHIP income estimates. Chapter 3 provides a full discussion of imputed rents and their estimation. Estimates of income from assets, and the contribution of that income to overall inequality, appear in Table 1.4. Imputed rental income is the largest component, accounting for more than 85 percent of asset income in both years. The share of asset income in total income was, on average, 7 to 9 percent of household per capita income in 2002 and 12 to 16 percent in 2007. Asset income is more unequally distributed than other types of income, and its contribution to overall inequality has grown. The importance of asset income to overall inequality is sensitive to the method used to estimate imputed rents (see Chapter 3), but all our estimates indicate that by 2007 this source of income had become increasingly important. In 2002, asset income contributed between 8 and 10 percent of inequality. By 2007, it contributed 13 to 19 percent (Table 1.4). The likely underreporting of asset income, as noted previously, may cause these estimates to understate the full impact of asset ownership on inequality. The evolution of asset ownership by China’s households is still in its early stages. Over time, and with the further development of financial and real estate markets and additional property rights reforms, perhaps including reforms in farmland ownership, household wealth is likely to grow, with ongoing implications for inequality. Income from property and asset income therefore will require close attention in future research.

Rising Inequality in China

35

Finding 5: Migration began to play an equalizing role through its impact on rural household incomes. The CHIP 2007 survey provides a second year of migrant data, which allows comparisons over time. Using the 2002 and 2007 migrant survey data, the contributors to this book have calculated estimates of inequality including and excluding long-term rural-urban migrants as part of the urban population. Including long-term migrants lowers measured inequality in both years, but only minimally. Similarly, including long-term migrants in the calculation of the urban-rural income gap reduces the size of the gap in both years, but again only minimally (see Table 1.4). The small impact of long-term rural-urban migrants on national inequality is somewhat surprising, but, as noted in Gustafsson, Li, and Sicular (2008), such calculations do not capture the full impact because they do not take into account how migration may have more generally changed wages and incomes among the rest of the rural and urban populations. Measuring the full impact of migration would require counterfactual information on what the level of inequality would have been if long-term migration had not occurred. Another reason why including long-term migrants does not substantially change measured inequality is because as a percentage of the national population the number of migrants who have succeeded in moving to urban areas on a long-term basis remains fairly small. Although incorporating long-term migrants in the inequality calculations suggests that migration has not had much of an impact, information about short-term migration collected in the rural household survey tells a different story (see Chapter 5). The 2002 and 2007 rural surveys contain information on rural household employment and income from shortterm migrant jobs. Participation in short-term migrant work grew rapidly between 2002 and 2007. In 2002 one-third of rural households reported earnings from short-term migrant work; by 2007, more than 40 percent reported such earnings. Indeed, in 2007 the number of rural households members who had participated in short-term migration greatly outnumbered the number of long-term rural-urban migrants. Between 2002 and 2007 income from short-term migrant work grew very rapidly, more than 17 percent a year in real terms, and by 2007 it was an important source of income for rural households. In 2007 these earnings constituted, on average, nearly 20 percent of rural household income per capita, almost on par with earnings from local wage employment. Income from short-term migrant employment had a substantial and moderating impact on inequality. Without earnings from migrant jobs,

36

Li Shi, Hiroshi Sato, and Terry Sicular

rural household incomes would have grown more slowly, and consequently the urban-rural income gap would have been even larger. Furthermore, earnings from short-term migrant jobs were equalizing within rural areas, thus helping to explain why inequality in rural China remained relatively stable during this period. These findings indicate the importance of shortterm migration as a key player in the inequality equation. Finding 6: Absolute poverty continued to decline, but relative poverty remained unchanged. Absolute poverty in China declined substantially between 2002 and 2007. Calculated using the World Bank international poverty line of $1.25 a day, the poverty rate fell from 19 percent to 8 percent, implying that the number of poor in China fell from about 244 million to 106 million (Table 1.4). This decline is remarkable, especially given that substantial reductions in poverty had already occurred in prior years, and that the remaining poverty in 2002 was relatively dispersed, transitory, and more difficult to resolve. Much of the decline in absolute poverty between 2002 and 2007 reflects marked poverty reduction in rural areas. Poverty also declined for the formal urban and migrant populations, but less so (Chapters 2, 5, and 7). The reduction in absolute poverty during this period likely reflects the benefits of the new social welfare policies adopted during the Hu-Wen period, such as health insurance, unemployment insurance, and the dibao program. With growth and development, attention turns from absolute poverty to relative poverty. In this context, poverty is no longer associated with meeting subsistence requirements, and disadvantages associated with relatively low socioeconomic status become more important. Several chapters in this book present estimates of relative poverty in China. Using a relative poverty line set at 50 percent of median income, Chapter 2 reports that relative poverty in China in 2002 and 2007 remained more or less unchanged at 13 percent (Table 1.4). Although absolute poverty was predominately a rural phenomenon, this was not the case for relative poverty. In rural areas relative poverty remained at about 13 percent, but in urban areas relative poverty continued its historical upward trend, reaching 12 percent in 2007 (Chapters 5 and 7). These findings are indicative of the changing nature of poverty in China and point to new policy challenges.

V. Conclusion The first decade of the twenty-first century was an important juncture in the evolution of incomes and inequality in China. It followed more

Rising Inequality in China

37

than a decade of rapid growth and rising inequality, which prompted a shift in the early 2000s to a policy agenda that placed greater emphasis on distributional concerns. During the years under examination in this volume, 2002–2007, inequality in China resumed its upward trend, reflecting in part the continued widening of the urban-rural gap and the emergence of asset income as a new source of inequality. Nevertheless, the Hu-Wen New Policies had some beneficial effects. Migration began to play a moderating role, and incomes of poorer groups grew. These factors, however, were insufficient to fully offset other, dis-equalizing factors. Measurement of inequality requires detailed information about income and its correlates, preferably at the household-level. The CHIP surveys, a unique, household-level data set, provide the basis for the analyses in this book. With careful analysis and interpretation, the CHIP data sets yield important insights into the evolution of incomes and inequality in China. In this chapter we have discussed the main features of the CHIP surveys and have pointed out potential biases in the data. Further discussion of the data appears in the individual chapters, as well as in the two data appendices. The chapters in this book provide an analysis of the main trends in household incomes, inequality, and poverty in urban and rural China, as well as nationally. Individual chapters provide in-depth analysis of topics ranging from intergenerational transmission of education, to housing ownership, migrant labor, urban gender and ethnic income gaps, and patterns of work and nonwork. Attention is paid to policy interventions, including social welfare programs such as the minimum living standard guarantee program, taxation, regional and rural development programs, and the hukou reforms. The coverage of topics is by no means exhaustive or complete, and our findings raise questions for further study. One area of analysis that is not fully investigated is the contribution of individual and household characteristics to inequality. Chapter 11 examines the impact of gender in the urban sectors and Chapter 12 examines the impact of ethnicity in the urban sector, but neither examine their contributions to inequality in rural China or nationwide; Chapter 4 looks at inequality of education, but it does not analyze how the distribution of education translates into inequality of income. The impact of age on incomes and inequality is another interesting area of research, especially in view of China’s aging population, the differences in age composition between the urban and rural areas, and recent developments in pension policies. The chapters in this book identify several factors that are playing increasingly important roles in driving inequality. One is asset ownership. Another is migration. A third is income mobility across generations, which is related

38

Li Shi, Hiroshi Sato, and Terry Sicular

to wealth accumulation among high-income groups as well as persistent segmentation across sectors and between urban and rural areas. The survey design and data analysis of the CHIP paid special attention to these topics, but the data still have limitations. Future data collection and analysis should be mindful of these issues. Several chapters discuss aspects of wage and earnings inequality relevant to the competitiveness of China’s labor markets. Chapter 9 provides evidence that intersectoral earnings inequality declined in the period under study, whereas Chapter 11 finds that discrimination against female workers worsened. These findings provide somewhat contradictory conclusions regarding whether the Chinese labor market has become more competitive. More studies are needed to better understand the relationship between the functioning of China’s labor markets and patterns of inequality. The chapters in this book highlight the important role of government policies in both moderating and exacerbating inequality. More in-depth analysis is needed to gain a detailed understanding of the impact of specific policy measures that are directly aimed at distributional concerns such as social welfare programs and hukou reforms, as well as other policies with distributional consequences. Relevant here, in our view, are short-term government stimulus measures in response to the world financial crisis, as well as long-term policy measures regarding land rights, enterprise ownership, international trade and investment, taxation, and the financial system. The findings in this book are based on data through 2007. In the ensuing years, China’s economic policies have evolved, partly in response to the world financial crisis and its aftermath. Official data indicate that in recent years the overall level of inequality and the urban-rural income gap have declined somewhat, and there has been some narrowing of income differences within rural areas. These trends may reflect the equalizing effects of rising wages for rural migrant workers and increases in public transfers to low-income rural households. Further data collection and analysis are needed to explore these new developments fully, and to determine whether they reflect short-term changes following the world financial crisis or longterm trends. The questions of whether and how inequality in China will reach a turning point remain relevant for the foreseeable future. References Benjamin, D., L. Brandt, J. Giles, and S. Wang (2008), “Income Inequality During China’s Economic Transition,” in L. Brandt and T. Rawski, eds., China’s Great Economic Transformation, 729–775, New York: Cambridge University Press. Bernstein, T.P. and X. L¨u (2003), Taxation without Representation in Contemporary Rural China, Cambridge: Cambridge University Press.

Rising Inequality in China

39

Cai, F. (2011), “Hukou System Reform and Unification of Rural-Urban Social Welfare,” China and World Economy, 19(3), 33–48. Central Committee of the Communist Party and the State Council (2002), “Zhonggong zhongyang Guowuyuan guanyu jinyibu jiaqiang nongcun weisheng gongzuo de jueding” (Decision of the Central Committee of the Chinese Communist Party and the State Council on Further Strengthening Public Health Work in Rural Areas), October 19, 2002, Guowuyuan gongbao, no. 33 (November 30), 4–8. Central Committee of the Communist Party of China and the State Council (2004), “Zhonggong zhongyang Guowuyuan guanyu zujin nongmin zengjia shouru ruogan zhengce de yijian” (Opinion of the Central Committee of the Chinese Communist Party and the State Council on Several Policies for Promoting an Increase in Farmers’ Income), December 31, 2003, Guowuyuan gongbao, no. 9 (March 30), 4– 10. Central Committee of the Communist Party of China and the State Council (2005), “Zhonggong zhongyang Guowuyuan guanyu jinyibu jiaqiang nongcun gongzuo tigao nongye zonghe shengchan nengli ruogan zhengce de yijian” (Opinion of the Central Committee of the Chinese Communist Party and the State Council on Reinforcing Rural Work and Improving Comprehensive Agricultural Productivity), December 31, 2004, Guowuyuan gongbao, no. 9 (March 30), 4–10. Central Committee of the Communist Party of China and the State Council (2006), “Zhonggong zhongyang Guowuyuan guanyu tuijin shehuizhuyi xin nongcun jianshe de ruogan yijian” (Opinion of the Central Committee of the Chinese Communist Party and the State Council on Promoting the Building of a New Socialist Countryside), Guowuyuan gongbao, no. 11 (April 30), 4–12. Central Committee of the Communist Party of China and the State Council (2010), “Zhonggong zhongyang Guowuyuan guanyu jiada tongchou chengxiang fazhan lidu jinyibu hangshi nongye nongcun fazhan jichu de ruogan yijian” (Opinion of the Central Committee of the Chinese Communist Party and the State Council on Reinforcing Equal Consideration of Urban and Rural Development and on Further Consolidating the Foundation for Agricultural and Rural Development), December 31, 2009, Guowuyuan gongbao, no. 4 (February 10), 5–13. Chan, K.W. (2009), “The Chinese Hukou System at 50,” Eurasian Geography and Economics, 50(2), 197–221. Chen, Z. (2006), “Measuring the Poverty Lines for Urban Households in China: An Equivalence Scale Method,” China Economic Review, 17(3), 239–252. Cheng, G., H. Song, and G. Dang (2012) “ ‘Huangliang guoshui’ zhongjie 2600 nian lishi” (“The Emperor’s Grain and the State’s Tax” Ended Its 2,600-Year History), Renmin ribao, June 28. Du, Y. and W. Pan (2009), “Minimum Wage Regulation in China and Its Applications to Migrant Workers in the Urban Labor Market,” China and World Economy, 17(2), 79–93. Eichen, M. and M. Zhang (1993), “Annex: The 1988 Household Sample Survey: Data Description and Availability,” in K. Griffin and R. Zhao, eds., The Distribution of Income in China, 331–346, New York: St. Martin’s Press. Feng, J., L. He, and H. Sato (2011), “Public Pension and Household Saving: Evidence from Urban China,” Journal of Comparative Economics, 39(4), 470–485. Fock, A. and C. Wong (2007a), “China: Improving Rural Public Finance for the Harmonious Society,” Report No. 41579-CN, The World Bank, Washington, DC.

40

Li Shi, Hiroshi Sato, and Terry Sicular

Fock, A. and C. Wong (2007b), “China: Public Services for Building the New Socialist Countryside,” Report No. 40221-CN, The World Bank, Washington, DC. Griffin, K. and R. Zhao, eds. (1993), The Distribution of Income in China, New York: St. Martin’s Press. Gustafsson, B. A., S. Li, and T. Sicular, eds. (2008), Inequality and Public Policy in China, New York: Cambridge University Press. Gustafsson, B. A., S. Li, T. Sicular, and X. Yue (2008), “Income Inequality and Spatial Differences in China: 1988, 1995, and 2002,” in B. A. Gustafsson, S. Li and T. Sicular, eds., Inequality and Public Policy in China, 35–60, New York: Cambridge University Press. He, L. and H. Sato (2013), “Income Redistribution in Urban China by Social Security System: An Empirical Analysis Based on Annual and Lifetime Income,” Contemporary Economic Policy, 31(2), 314–331. Hirschman, A. (1958), The Strategy of Economic Development, New Haven, CT: Yale University Press. Keister, L. A. and E. P. Borelli (2012), “Market Transition: An Assessment of the State of the Field,” Sociological Perspectives, 55(2), 267–294. Khan, A.R., K. Griffin, C. Riskin, and R. Zhao (1992), “Household Income and Its Distribution in China,” China Quarterly, no. 132, 1029–1061. Knight, J. and L. Song (1999), The Rural-Urban Divide: Economic Disparities and Interactions in China, Oxford: Oxford University Press. Lai, D., D. Meng, C. Li, and Y. Tian (2011), “Zhongguo jiuye zhengce pingjia: 1998– 2008” (An Evaluation of China’s Employment Policy: 1998–2008), Beijing shifan daxue xuebao (Shehuikexue), no. 3, 110–124. Lei, X. and Y. Wang (2009), “Zhongguo zuidi shenghuo baozhang zhidu xianzhuang yu huigu” (Status and Review of the Minimum Living Standard Security System), Shehui baozhang yanjiu, no. 2, 45–55. Li, S. and C. Luo (2010), “Re-estimating the Income Gap between Urban and Rural Households in China,” in M. K. Whyte, ed., One Country, Two Societies: Rural-Urban Inequality in Contemporary China, 105–121, Cambridge, MA: Harvard University Press. Li, S. and C. Luo (2011), “Zhongguo shouru chaju jiujing you duoda?” (How Unequal Is China?), Jingji yanjiu, no. 4, 68–78. Li, S., C. Luo, Z. Wei, and X. Yue (2008), “Appendix: The 1995 and 2002 Household Surveys: Sampling Methods and Data Description,” in B. A. Gustafsson, S. Li, and T. Sicular, eds., Inequality and Public Policy in China, 337–353, New York: Cambridge University Press. Li, S. and H. Sato (2006), Unemployment, Inequality and Poverty in Urban China. New York: Routledge. Liu, Z. (2011), “Geren suodeshui zhidu gaige ‘shiyiwu’ huigu yu ‘shierwu’ zhanwang” (Reform of the Personal Income Tax: Retrospect of the Eleventh Five-Year Plan and Prospects for the Twelfth Five-Year Plan), Caizheng yanjiu, no. 10, 27–29. Luo, C., X. Yu, and S. Li (2011), “Dui Wang Xiaolu huise shouru gusuan de zhiyi” (A Critique of Wang Xiaolu’s Estimation of Grey Income), Bijiao, no. 52, 146–158. Luo, R., L. Zhang, and M. Deng (2008), “Nongcun gonggong wupin touzi cel¨ue de shizheng fenxi” (Empirical Analysis of Rural Public Goods Investment Strategy), Zhongguo kexue jijin, no. 6, 325–330.

Rising Inequality in China

41

Milanovic, B. and L. Ersado (2008), “Reform and Inequality During the Transition: An Analysis Using Panel Household Survey Data, 1990–2005,” World Bank Policy Research Working Paper No. 4780, The World Bank, Washington, DC. Ministry of Labor and Social Security et al. (2003), “Guanyu guanche luoshi zhonggong zhongyang Guowuyuan guanyu jinyibu zuohao xiagang shiye renyuan zaijiuye gongzuo de tongzhi ruogan wenti de yijian” (Joint Opinions on Comments for the Thorough Enforcement of the Communiqu´e of the Central Committee of the Chinese Communist Party and the State Council on Making Further Improvements in the Reemployment Policy for Laid-off and Unemployed Workers), October 15, 2002, Zhongguo jiuye, no. 3, 4–6. Ministry of Public Security (1998), “Guanyu jiejue dangqian hukou guanli gongzuozhong jige tuchu wenti yijian” (Opinion on the Settlement of Current Critical Issues concerning Hukou Administration), Guowuyuan gongbao, no. 21, 827– 829. Ministry of Public Security (2001), “Guanyu tuijin xiaochengzhen huji guanli zhidu gaige de yijian” (Opinion on Promoting Hukou Administration Reform in Small Cities and Towns), March 19, Guowuyuan gongbao, no. 15 (May 30), 13–15. National Bureau of Statistics (NBS) (various years), Zhongguo tongji nianjian (China Statistical Yearbook), Beijing: Zhongguo tongji chubanshe. National Bureau of Statistics (NBS) (various years), Zhongguo tongji zhaiyao (China Statistical Abstract), Beijing: Zhongguo tongji chubanshe. National Bureau of Statistics (NBS) (2008), Zhongguo nongcun zhuhu diaocha nianjian 2008 (China Yearbook of Rural Household Survey 2008), Beijing: Zhongguo tongji chubanshe. National Bureau of Statistics (NBS) (2010), Xin Zhongguo 60 nian tongji ziliao huibian (Collection of Sixty Years of New China’s Statistical Materials), Beijing: Zhongguo tongji chubanshe. Nee, V. and S. Opper (2010), “Political Capital in a Market Economy,” Social Forces, 88(5), 2105–2132. Ravallion, M. and S. Chen (2007), “China’s (Uneven) Progress against Poverty,” Journal of Development Economics, 82(1), 1–42. Riskin, C., R. Zhao, and S. Li, eds. (2001), China’s Retreat from Equality: Income Distribution and Economic Transition, Armonk, NY: M.E. Sharpe. Rural Socioeconomic Survey Department of the National Bureau of Statistics (2011a), Zhongguo nongcun pinkun jiance baogao 2010 (Poverty Monitoring Report of Rural China 2010), Beijing: Zhongguo tongji chubanshe. Rural Socioeconomic Survey Department of the National Bureau of Statistics (2011b), Zhongguo nongcun tongji nianjian 2011 (China Rural Statistical Yearbook 2011), Beijing: Zhongguo tongji chubanshe. Sato, H. and S. Ding (2012), “Local Public Goods Provision in the Post-Agricultural Tax Era in Rural China,” Global COE Hi-Stat Discussion Paper Series No. 222, Hitotsubashi University, Tokyo. Sato, H., S. Li, and X. Yue (2008), “The Redistributive Impact of Taxation in Rural China, 1995–2002,” in B. A. Gustafsson, S. Li, and T. Sicular, eds., Inequality and Public Policy in China, 312–336. New York: Cambridge University Press. Sicular, T., X. Yue, B. Gustafsson, and S. Li (2007), “The Urban-Rural Income Gap and Inequality in China,” Review of Income and Wealth, 53(1), 93–126.

42

Li Shi, Hiroshi Sato, and Terry Sicular

State Council (1997), “Guowuyuan guanyu zai quanguo jianli chengshi jumin zuidi shenghuo baozhang zhidu de tongzhi” (Communiqu´e of the State Council on Establishing a Minimum Living Standard Guarantee Program for City Dwellers across the Country), September 2, 1997, Guowuyuan gongbao, no. 31(October 17), 1398–1400. State Council (1999), “Guowuyuan guanyu jianli chengzhen zhigong jiben yiliao baoxian zhidu de jueding” (Decision of the State Council on Establishing a Basic Medical Care Insurance System for Urban Workers), December 14, 1998, Guowuyuan gongbao, no. 33 (January 13), 1250–1254. State Council (2001), “Zhongguo nongcun fupin kaifa gangyao (2001–2010)” (Outline for Chinese Poverty Relief and Development in Rural Areas [2001–2010]), August 20, 2001, Guowuyuan gongbao, no. 23 (June 13), 34–39. State Council (2005), “Guowuyuan guanyu jinyibu jiaqiang jiuye zaijiuye gongzuo de tongzhi” (Communiqu´e of the State Council on Reinforcement of Employment and Reemployment Policies), November 4, 2005, Guowuyuan gongbao, no. 35 (December 20), http://www.gov.cn/gongbao/content/2005/content 129498.htm. Accessed December 28, 2012. State Council (2006), “Nongcun wubao gongyang gongzuo tiaoli” (Regulations on the Five-Guarantee Social Assistance Program), January 21, 2006, Guowuyuan gongbao, no. 7 (March 10), http://www.gov.cn/gongbao/content/2006/content 219932.htm. Accessed July 29, 2012. State Council (2008), “Guowuyuan guanyu zuohao jiuye gongzuo de tongzhi” (Communiqu´e of the State Council on Bringing Success to Employment Promotion), February 3, 2008, Guowuyuan gongbao, no. 8 (March 30), 5–8. State Council (2011a), “Guanyu xiugai Zhonghua renmin gongheguo geren suodeshuifa shishi tiaoli de jueding” (Decision of the State Council on the Amendment of the Regulations for Enforcement of the Personal Income Tax Law), July 19, 2011, Guowuyuan gongbao, no. 22 (August 10), http://www.gov.cn/gongbao/content/2011/ content 1918912.htm. Accessed July 29, 2012. State Council (2011b), “Guowuyuan bangongting guanyu jiji wentuo huji guanli zhidu gaige de tongzhi” (Communiqu´e of the General Office of the State Council on Positively and Temperately Promoting Hukou Administration Reform), issued in February 2011, http://www.gov.cn/zwgk/2012-02/23/content 2075082.htm. Accessed February 27, 2012. Wagstaff, A., M. Lindelow, S. Wang, and S. Zhang (2009), Reforming China’s Rural Health System. Washington, DC: The World Bank. Walder, A. G. (2003), “Elite Opportunity in Transitional Economies,” American Sociological Review, 68(6), 899–916. Wang, X. (2010), “Huise shouru yu guomin shouru fenpei” (Grey Income and National Income Distribution), Bijiao, no. 48, 1–29. Wang, Z. (2010) “Xinnongcun jianshe de shouru zaifenpei xiaoying” (The Redistribution Effects of New Countryside Construction Policies), Jingji yanjiu, no. 6, 17–27. Wen, J. (2003), “Wei tuijin nongcun xiaokang jianshe er fendou” (Let Us Struggle for the Construction of a Well-off Countryside), Renmin ribao, February 8. Whyte, M. K., ed. (2010), One Country, Two Societies: Rural-Urban Inequality in Contemporary China. Cambridge, MA: Harvard University Press. Wong, C. and A. Fock (2008), “Financing Rural Development for a Harmonious Society in China: Recent Reforms in Public Finance and Their Prospects,” World Bank Policy Research Working Paper No. 4693, The World Bank, Washington, DC.

Rising Inequality in China

43

World Bank (2009), “From Poor Areas to Poor People: China’s Evolving Poverty Reduction Agenda: An Assessment of Poverty and Inequality in China,” Poverty Reduction and Economic Management Department, East Asia and Pacific Region, Report No. 47349-CN, The World Bank, Washington, DC. Xinhua News Agency (2013), “China Gini Coefficient at 0.474 in 2012,” China Daily, January 18, http://www.chinadaily.com/bizchina/2013-01/18/content 16140018.htm. Accessed March 9, 2013. Zhang, D., ed. (2010), Zhongguo jumin shouru fenpei niandu baogao 2010 (China Income Distribution Annual Report 2010), Beijing: Jingji kexue chubanshe.

TWO

Overview Income Inequality and Poverty in China, 2002–2007 Li Shi, Luo Chuliang, and Terry Sicular

I. Introduction It has been more than three decades since China started to transform its economy institutionally and structurally. The economic transformation has stimulated rapid economic growth in both GDP and personal incomes. From 1978 to 2007 the annual growth of GDP averaged close to 10 percent and that of household per capita income more than 7 percent. The rate of economic growth was even more impressive in later years, including the period under study in this chapter. From 2002 to 2007, annual growth of GDP was 11.6 percent and of rural and urban household per capita income 6.8 and 9.6 percent, respectively.1 Although the reforms were successful in promoting GDP growth, by the early 2000s, concerns about rising disparities and sustainability prompted the government to announce a new development strategy emphasizing sustainable, harmonious growth. A new policy program, referred to as the “scientific outlook on development” (kexue fazhanguan), or the “Hu-Wen New Policies” (Hu-Wen xin zheng), aimed to promote development in urban and rural areas, reduce regional disparities, narrow income inequalities, and establish a social protection network with broad coverage over most of the population. As discussed in Chapters 1 and 5, the new policy program contained a series of pro-rural measures. These included the elimination of agricultural taxes, which had been in place for almost sixty years, and the adoption of new farm subsidies, for example, for grain production, purchase

1

These statistics are based on data published by the NBS (National Bureau of Statistics 2008b). As discussed in Chapter 1 and later in this chapter, the NBS statistics yield somewhat different rates of growth in household income than the CHIP data.

44

Overview

45

of agricultural inputs, and farm insurance (Lin and Wong 2012).2 By the end of 2007, Chinese rural households were no longer paying agricultural taxes, and total agricultural production subsidies from the central government exceeded 50 billion yuan (Lin and Wong 2012; Ministry of Agriculture 2007). The pro-rural policies also addressed social welfare concerns. In the early 2000s, the government initiated programs that reduced the costs of education in poor areas, and in 2006–2007 the central government announced a policy of free education in rural areas through junior middle school, eliminating all fees for the first nine years of education (see Chapter 4). During the same time frame, subsidized rural cooperative health care and a rural medical care relief fund were put in place. Although these measures did not have an immediate effect on household earnings, they reduced household outlays on education and health and encouraged schooling, which in the long term can enhance incomes. The rural minimum living standard guarantee (zuidi shenghuo baozheng, or dibao) program was another important component of the rural policy program. The number of rural people supported by dibao increased enormously, from 4 million in 2002 to 36 million in 2007. On average, in 2007 each individual received about 480 yuan, equivalent to 60 percent of the official poverty line in rural areas (Ministry of Civil Affairs 2007; see Chapters 1 and 5 in this volume). During this period, the Chinese government also maintained or expanded policies benefiting lower-income urban households, such as the urban dibao program and the provision of low-cost housing. Some steps were also taken to improve the situation of poor rural-urban migrants, for example, regulations issued in 2003 regarding the treatment of vagrants and beggars, which provided social services to poor individuals regardless of their place of origin (Li 2004; State Council 2003). The impact of such programs on urban inequality, however, has been mixed. Analyses of the urban dibao program, for example, reveal that it played an important role in alleviating urban poverty but did not substantially reduce urban income inequality (Li and Yang 2009; Ravallion, Chen, and Wang 2006). Moreover, the number of urban households benefiting from the program did not increase significantly during the period under study here. 2

The total amount of agricultural subsidy funds, including grain subsidies, reached 52.6 billion yuan in 2007; see “Nongyebu: Guojia jiang baochi zhi nonghui nong zhengce de wendingxing lianxuxing” (Ministry of Agriculture: The State Will Maintain Stable and Continuous Policy Support for Agriculture), September 13, 2007, http://www.china.com .cn/news/2007-09/13/content 8869413.htm. Accessed August 22, 2011.

46

Li Shi, Luo Chuliang, and Terry Sicular

China’s economic growth is closely related to urbanization. The share of the urban population in China’s total population has increased almost one percentage point each year since 1990. By the end of 2007, the share of the urban population in the total population was 45 percent. Rural-tourban migration has been an important part of the urbanization process. According to the Second National Agricultural Census, in 2006 the number of rural-urban migrant workers who were employed in urban areas for more than six months per year was about 132 million. Although rural migration can contribute to the growth of household income in rural areas, it can also create competition in urban labor markets that potentially affects urban incomes and inequality as well. In China, rural-urban and regional divisions in terms of economic and social development are substantial. These spatial divisions were significant during the planning period (D´emurger et al. 2002) and have persisted into the reform era. Concerns about the urban-rural income gap prompted many of the rural support policies outlined earlier. Similarly, differential economic growth between coastal and inland regions led the Chinese government to adopt regional balancing policies. In 1999, the central government implemented the western development strategy (xibu dakaifa zhanl¨ue) and increased investment in infrastructure and fiscal transfers to western provinces (Fang, Zhang, and Li 2007). This was followed by further programs supporting other lagging regions, such as the revival of the Northeast strategy (zhenxing dongbei) in 2003 and the rise of the central region (zhongbu jueqi) scheme aimed at the central provinces in 2006 (Chung, Lai, and Joo 2009; Yao 2009). Such policies could have an impact on regional income disparities. Using data from the 2002 and 2007 waves of the China Household Income Project (CHIP) survey, in this chapter we measure and analyze income inequality and poverty during the 2002–2007 period. Here we report overall nationwide patterns and trends. The findings reported in this chapter establish the groundwork for the later chapters in this volume, which provide in-depth analyses of particular sectors, programs, and policies. We begin in the next section with a brief review of the main findings in the recent literature on changes in China’s income inequality and summarize the results from the previous volume based on the 2002 CHIP survey (Gustafsson, Li, and Sicular 2008b). In Section III we explain key features of our data. In Section IV we present our central findings regarding levels and trends in China’s national income inequality, and we examine the sources of income. Despite substantial growth in mean incomes between 2002 and 2007 and despite the various policies adopted to promote

Overview

47

harmonious growth, during this period nationwide inequality continued an upward trend. This conclusion is robust to the choice of income definition, weights, and inequality index, and to the treatment of migrants. A growing number of rural people have moved to the cities, but they are not fully captured in the official National Bureau of Statistics (NBS) household surveys. This leads to a potential bias in estimations of income growth and inequality among Chinese households. Other chapters in this volume examine the income and inequality of the rural and formal urban populations, but not that of rural-urban migrants. Therefore, in this chapter we include a separate section on income and inequality among rural-urban migrants. Following the approach used by the NBS to classify location of residence, we define migrants as those individuals who have a rural household registration but who reside in a city on a long-term, stable basis. Short-term, temporary migrants are treated as members of their rural households of origin and are included in the rural survey data set (see further discussion in Chapter 1 and Appendix II). Our analysis shows that between 2002 and 2007, the incomes of longterm, stable rural-urban migrants grew rapidly, and inequality among migrants declined. Including migrants in our calculations of inequality reduces inequality within the urban areas, but because of the relatively low share of this group in the national population, it does not substantially alter the national levels of inequality. Temporary and short-term migration, however, contributed to income growth of rural households and thus likely moderated the income gap between the urban and rural areas (see also Chapter 6). The increase in China’s national inequality between 2002 and 2007 reflects changes in the spatial structure of China’s income distribution, as discussed in Sections VI and VII of this chapter. The continued widening of the urbanrural income gap is of particular concern because the urban-rural divide remains a major source of inequality. Analysis of inequality among geographic regions reveals that regional income differentials in fact contribute a relatively small share of national inequality. The overwhelming majority of national inequality is associated with inequality within regions, including urban-rural gaps within regions. Finally, in Section VIII we examine nationwide trends in poverty (later chapters in this volume examine rural and urban poverty separately). Between 2002 and 2007 national poverty, as measured using an absolute poverty line, continued to decline and reached historically low levels. Relative poverty, however, remained unchanged. We comment on these and other findings in a concluding section.

48

Li Shi, Luo Chuliang, and Terry Sicular

II. Main Findings of Previous Studies The rise in income inequality in China during the reform era has been widely documented. Past studies have found that nationwide inequality rose rapidly between the late 1980s and the mid-1990s but then tapered off from the mid-1990s through the early 2000s. Estimates by Ravallion and Chen (2007) and the World Bank (2009a) show income inequality rising from the late 1980s through 1994, dipping a bit in the late 1990s, and then edging upward thereafter, so that by the early 2000s inequality was only slightly higher than it had been in the mid-1990s. Analyses based on the 1995 and 2002 CHIP surveys similarly report that inequality remained more or less unchanged between 1995 and 2002 (Gustafsson, Li, and Sicular 2008a; Khan and Riskin 2008). Gustafsson et al. (2008a) identify several equalizing processes that emerged in the late 1990s that might explain these trends. These include the spread of wage employment in the rural areas, the catching up of lowerincome provinces with higher-income provinces in some regions, shared macroeconomic growth, and, within urban areas, broader implementation of the urban housing reforms. The emergence of these equalizing processes raised the possibility that inequality in China had turned the corner. Findings based on the 2007 CHIP data reported here, however, show that after 2002 inequality in China resumed its upward trajectory. The analysis in this and later chapters finds evidence that some equalizing processes continued to operate from 2002 through 2007, but they were insufficient to offset the stronger dis-equalizing forces. Spatial income differentials figure large in the literature on inequality in China. The widening gap between urban and rural incomes is consistently cited as an important factor underlying national inequality (e.g., Kanbur and Zhang 2009; Ravallion and Chen 2007; Sicular et al. 2010; World Bank 2009a). This finding is robust across numerous studies using different measures of income and inequality. Regional income differences between the eastern, central, and western regions have also received attention, although several recent studies conclude that regional differences are not as important as within-region and urban-rural inequality (Fan, Kanbur, and Zhang 2010; Wan 2007; Yao 2009). In the following, we explore urban-rural and regional income differentials using the 2007 CHIP data; our findings are generally consistent with these other studies. China has an enviable record of poverty reduction (Chen and Ravallion 2008; Ravallion and Chen 2007; World Bank 2009a). Although various

Overview

49

studies differ in their choices of poverty measures and poverty lines, they agree on broad trends over time. During the early and mid-1990s, poverty in China declined substantially, but then in the late 1990s to the early 2000s, the downward trend stalled (Minoiu and Reddy 2008; Ravallion and Chen 2007; World Bank 2009a). Some recent studies suggest that after 2001 poverty reduction once again accelerated (World Bank 2009a). Our estimates of absolute poverty show progress in terms of poverty reduction from 2002 through 2007. Most of the literature on poverty in China measures poverty using an absolute poverty line based on the cost of basic food and nonfood consumption needs. As countries develop, deprivation is associated more with relative than with absolute living standards. In view of China’s transformation from a low- to a middle-income country, we extend the analysis of poverty and measure relative poverty. By such a measure, China’s poverty record in recent years is less encouraging. Poverty, like inequality, has spatial dimensions: it is primarily rural, and its incidence is higher in western China than elsewhere (Ravallion and Chen 2007; World Bank 2009a). As the overall level of poverty has declined, however, the remaining poor have become increasingly dispersed. The spatial pattern of poverty is important in terms of the design of poverty alleviation programs, which in China have relied heavily on geographic targeting (World Bank 2009a). Therefore, in the following analysis, we also investigate regional aspects of poverty.

III. Data and Sample Weights The data used in this chapter come from the last two waves of the CHIP household surveys, 2002 and 2007. The surveys cover three types of households: urban households, rural households, and rural-urban migrant households. The CHIP samples of urban households and rural households are subsamples of the large NBS urban and rural household survey samples. In 2002, the NBS samples included 680,000 households in rural areas and 40,000 households in urban areas.3 In 2007, the NBS urban sample increased to 59,000 households, but the size of the rural sample remained more or less unchanged.4 The NBS urban and rural household survey samples cover all 3 4

See the introduction to the sampling procedure for the NBS household survey in 2002 (NBS 2003: 339–340). See the introduction to the sampling procedure for the NBS household survey in 2007 (NBS 2008a: 313–314).

50

Li Shi, Luo Chuliang, and Terry Sicular

of China’s provinces. The CHIP survey covers a subset of provinces selected to obtain representation of China’s major regions. The 2002 wave of the CHIP rural survey selected 9,200 households from the NBS rural household survey. These households contain 37,969 individuals from 120 counties of twenty-two provinces. For the rural sample, the provinces include Beijing (representing the large metropolitan cities with provincial administrative status); Hebei, Liaoning, Jiangsu, Zhejiang, Shandong, and Guangdong (representing the eastern region); Shanxi, Jilin, Anhui, Jiangxi, Henan, Hubei, and Hunan (representing the central region); and Chongqing, Sichuan, Guizhou, Yunnan, Guangxi, Shaanxi, Xinjiang, and Gansu (representing the western region).5 The provincial statistical bureaus were given autonomy to decide the number of counties in the CHIP subsample, but they were required to select counties and villages representative of different income levels. The 2002 urban survey selected 6,835 households. These households contain 20,632 individuals surveyed in seventy cities in eleven of the twenty-two provinces of the rural survey, including Beijing (large municipality); Liaoning, Jiangsu, and Guangdong (eastern); Shanxi, Anhui, Henan, and Hubei (central); and Chongqing, Sichuan, Yunnan, and Gansu (western). These households are largely formal urban residents with local household registration (hukou). A detailed description of the 2002 survey can be found in Li et al. (2008). The 2002 rural and urban household questionnaires were designed for the purpose of deriving household income that could be comparable internationally. The households were asked questions regarding wage and other income components for each of their working members and regarding income from family businesses. In order to estimate the imputed rent of owner-occupied private housing, several housing-related questions were included, such as the self-estimated market value and the market rent of owner-occupied housing. The 2002 CHIP survey also included a separate, add-on sample of 2,000 rural-urban migrant households, which were selected from the capital city plus one middle-sized city in each province represented in the CHIP urban survey. From each of the provinces in the eastern and central regions, 200 households were selected, and from each of the provinces in the western region, 150 households. Within each province, 100 households were 5

Note that Chongqing did not become a provincial-level municipality until 1997 and it is markedly less urbanized and less economically developed than China’s other provinciallevel municipalities (Beijing, Shanghai, and Tianjin). Therefore, in this and other chapters, Chongqing is included as part of the western region.

Overview

51

allocated to the capital city and the remainder to other cities. Within the cities, rural-urban migrant households were selected from residential communities; hence, the migrant workers living in construction sites and factories were excluded from the sample. Because in our analyses we only use the subsample of migrants who are long-term, stable residents of cities, this aspect of the 2002 sample selection is not overly problematic. The migrant questionnaires include questions regarding wage, business income, consumption, and job characteristics of individual members and households. The 2007 CHIP surveys of rural and urban households were conducted in sixteen provinces, including Beijing and Shanghai (representing the large metropolitan cities with provincial administrative status); Fujian, Guangdong, Liaoning, Jiangsu, and Zhejiang (eastern); Anhui, Hebei, Henan, Hubei, and Shanxi (central); and Chongqing, Sichuan, Yunnan, and Gansu (western). The survey of rural-urban migrant households covered nine of the preceding sixteen provinces. The 2007 CHIP surveys cover 13,000 rural households, 10,000 urban local households, and 5,000 ruralurban migrant households. As in 2002, the 2007 surveys of rural households and urban local households took subsamples from the large NBS sample, whereas the rural-urban migrant survey was conducted separately. For the 2007 migrant survey, sampling was carried out using a geographical grid. Cells from the grid were chosen randomly; within each selected cell, the survey team identified all employers and workplaces and drew up a list of all their migrant employees. Migrants were then selected randomly from this list of employees. The CHIP migrant survey sample is composed of the selected migrants and their household members. This approach is different from that used to construct the 2002 migrant sample. The change in the sampling method for migrants may affect comparisons across the two years; however, to some extent the consequences are mitigated by the fact that in our analysis we include only those migrants who are long-term, stable residents, and by the use of population weights when incorporating the migrant subsamples into our urban and national calculations. More details about the 2007 survey are provided in Chapter 1 and Appendix I of this volume. The questionnaires for the 2007 surveys include many but not all of the same questions as the 2002 surveys. New questions regarding migration status and behavior were added for the purpose of migration analysis. The CHIP survey samples have several characteristics that may lead to an estimation bias if the samples are used without population-based sample weights. A detailed discussion of weights can be found in Appendix II

52

Li Shi, Luo Chuliang, and Terry Sicular

of this volume and in Li et al. (2008). The key issues are (a) the CHIP sample was designed to be representative of four distinct regions (large municipalities with provincial status, eastern China, central China, and western China);6 (b) not all provinces are included in the samples, and provincial coverage changed between 2002 and 2007; (c) provincial sample sizes are not proportional to their populations; and (d) the urban, rural, and migrant sample sizes are not proportional to their populations. In view of these features, when subsamples are combined among groups and regions, and for comparison over time, population weights are needed to make the samples representative and comparable across years. As discussed in Appendix II of this volume, two alternative approaches are recommended for sample weights. The first is to use two-level weights based on the population shares of each group (urban, rural, and, where relevant, migrant) within each region. The second is to use three-level weights based on the population shares of each group (urban, rural, and, where relevant, migrant) within each province and region. In general, we use three-level weights, but to show the sensitivity of the estimation results to the weighting methods, in Table 2A.1 we present estimates of national incomes and inequality calculated using alternative weights. With respect to income, our preferred measure is net disposable household per capita income. The NBS calculates an estimate of net disposable household income that is published in the official sources and is provided in the CHIP data sets. As discussed elsewhere (Gustafsson et al. 2008a; Khan and Riskin 1998), the NBS calculation of net disposable income omits certain components of income. For this reason, we prefer an alternative calculation of income based on that outlined in Khan et al. (1992) and Khan and Riskin (1998), but adapted in light of recent shifts in the structure of income and data availability. Specifically, we calculate income as NBS income, plus imputed subsidies on subsidized rental housing, plus the imputed value of rental income on owner-occupied housing. The CHIP surveys contain information on estimated market rents and market housing values that are used to calculate these additional income components. For imputed rental income of owner-occupied housing, we use the estimates 6

The geographic areas used to construct the CHIP sample frame are (1) large municipalities with provincial status (Beijing, Tianjin, and Shanghai, treated together as a separate geographic area [Chongqing is treated as part of Sichuan in western China for consistency with earlier rounds of the survey]) and (2) eastern China (Hebei, Liaoning, Jiangsu, Zhejiang, Fujian, Shandong, Guangdong, and Hainan), central China (Shanxi, Jilin, Heilongjiang, Anhui, Jiangxi, Henan, Hubei, and Hunan), and western China (Inner Mongolia, Guangxi, Chongqing, Sichuan, Guizhou, Yunnan, Tibet, Shaanxi, Gansu, Qinghai, Ningxia, and Xinjiang).

Overview

53

explained in Chapter 3.7 We refer to this alternative, broader measure of income as “CHIP income.” For purposes of comparison over time, we deflate the 2007 incomes using the consumer price indices published by the NBS to obtain values in constant 2002 prices. For national calculations, we use the average national consumer price index. For separate analyses of the urban and rural areas, we use the separate urban and rural consumer price indices (the urban consumer price index is used for rural-urban migrants). Between 2002 and 2007 the consumer price indices show that, on average, consumer prices rose by 13.9 percent nationwide, by 12.3 percent in urban areas, and by 16.4 percent in rural areas.8 Several studies note that differences in costs of living among regions and provinces can lead to an overstatement of real inequality (Brandt and Holz 2006; Sicular et al. 2010). To obtain income that is comparable among regions in terms of purchasing power parity (PPP), we use the PPP-adjusted deflator from Brandt and Holz (2006) to correct for differences in living costs between urban and rural areas and among provinces. Brandt and Holz (2006) provide the PPP deflators for 2002 that we apply to the 2002 CHIP data. For 2007, we update the Brandt and Holz PPP deflators using the official consumer price indices for urban and rural areas by province, as published by the NBS.

IV. National Household Income Inequality: Main Findings Table 2.1 shows mean national household per capita income and income inequality calculated using three commonly used inequality indices, the Gini coefficient and two Theil indices. Although less common than the Gini coefficient, the Theil indices have desirable properties and, unlike the Gini coefficient, can be decomposed to analyze inequality between and within groups, which is useful for us to examine the role of urban-rural 7

8

Chapter 3 provides two alternative estimates of imputed rents, one in which all imputed rents are calculated using the rate of return approach, and the other in which urban imputed rents are calculated based on the market rent approach and rural imputed rents based on the rate of return approach. The former approach shows higher urban incomes and more rapid growth in urban incomes as it is more sensitive to housing price appreciation in urban China. In this chapter we use the latter approach, which gives lower estimates of national inequality. Table 2A.2 provides comparisons of the results calculated using the two approaches. See Chapter 3 for additional discussion and comparisons between the two approaches. See China Statistical Yearbook 2008, http://www.stats.gov.cn/tjsj/ndsj/2008/indexch.htm. Accessed August 22, 2011.

54

Li Shi, Luo Chuliang, and Terry Sicular Table 2.1. National mean income and inequality, 2002 and 2007

2002 Excluding Migrants

% Change, 2002–2007, Constant 2002 Prices

2007

Including Migrants

Excluding Migrants

Including Migrants

Excluding Migrants

Including Migrants

NBS Income Mean income Gini GE(0) GE(1)

4,467

4,530

8,932

9,165

75.61

77.69

0.456 0.362 0.360

0.455 0.361 0.356

0.481 0.414 0.398

0.478 0.413 0.392

5.48 14.36 10.56

5.05 14.40 10.11

CHIP Income Mean income Gini GE(0) GE(1)

4,958

4,902

10,072

10,277

78.42

84.13

0.462 0.370 0.370

0.460 0.368 0.366

0.487 0.424 0.409

0.483 0.420 0.401

5.41 14.59 10.54

5.00 14.13 9.56

Notes: 1. All estimates are calculated using three-level weights, i.e., urban/rural × regional × provincial population shares. 2. Estimates are calculated using data from all provinces covered by the CHIP surveys. 3. Mean incomes for each year are calculated using current-year prices, and the change between 2002 and 2007 is calculated using constant 2002 prices (deflated using the national average consumer price index). 4. The inequality indices shown in this table are all scale-invariant. Consequently, the level of inequality is the same for both the current year and constant prices (if deflation is carried out using the same price index for all individuals). 5. Here and elsewhere, incomes less than or equal to zero have been dropped for calculation of the GE(0) and GE(1) inequality indices. In all, fewer than 30 observations (individuals) were dropped in 2002 and fewer than 225 in 2007.

inequality.9 We also show the Lorenz curve, which gives a graphical depiction of inequality and is closely related to the Gini coefficient.10 9

10

The Theil indices, like the Gini coefficient, have a minimum value of 0 and increase with inequality. G(0), sometimes referred to as the Mean Log Deviation (MLD), is more sensitive to income differences at the low end of the income distribution. G(1), sometimes referred to as the Theil index, places equal weight on income differences across the income distribution. More information about the Theil measures of inequality can be found in Cowell (2011) and at http://siteresources.worldbank.org/INTPA/Resources/ tn measuring inequality.pdf and http://siteresources.worldbank.org/INTPGI/Resources/ Inequality/litchfie.pdf. Both accessed June 5, 2012. The Lorenz curve is a plot of the percentage of total income in society accruing to the bottom x percentage of the population. In the case of perfect equality (all members of

Overview

55

Our preferred estimates are calculated using the CHIP definition of income, including migrants, and with three-level population weights (urban/rural/migrant group × region × province). As our preferences may not be universally shared, and for ease of comparison with other studies, we also present estimates calculated using the NBS definition of income, and excluding migrants. Table 2A.1 gives estimates calculated using alternative weighting methods, and Table 2A.2 gives estimates calculated using alternative estimates of imputed rents on owner-occupied housing. On average, incomes increased markedly between 2002 and 2007. Regardless of the income definition, treatment of migrants, or choice of weights, mean income increased more than 70 percent during the five years (calculated using constant 2002 prices), implying an average annual growth in income of 12 to 13 percent. Income growth was more rapid for the CHIP definition of income than for the NBS definition, reflecting growth in imputed rents on owner-occupied housing and the expansion of urban homeownership, as discussed in Chapter 3. The inclusion of migrants modestly increases the mean income levels in 2007 and yields more rapid growth in income. On balance, growth in mean income should reduce inequality: if mean income increases while the distribution of income around the mean stays unchanged, then measured inequality will decline. Despite the substantial growth in national mean income, however, inequality in China increased. From 2002 to 2007, China’s Gini coefficient rose by 5.0 to 5.5 percent. For our preferred calculation (CHIP income, including migrants), the Gini coefficient rose by 5.0 percent, from 0.46 in 2002 to 0.48 in 2007. This level of inequality is moderately high by international standards. Increases in the Theil measures of inequality were larger, ranging from 9.5 percent for G(1) to nearly 14.6 percent for G(0). Differences in inequality trends among the three measures reflect that each measure emphasizes different sections of the income distribution. The Gini coefficient emphasizes income differences in the middle of the distribution, the GE(0) places the population have equal income), the Lorenz curve is coincident with the 45-degree line. The farther the Lorenz curve is from the 45-degree line, the greater the inequality. In the case of perfect inequality (one person has all the income and everyone else has zero income), the Lorenz curve is a right angle and coincides with the axis. The Gini coefficient is calculated as the ratio of the area between the Lorenz curve and the 45-degree line to the total area under the 45-degree line. The minimum value of the Gini coefficient is zero, which occurs when there is perfect equality, and the maximum value is one, which occurs when there is perfect inequality. In practice, the Gini coefficient for most countries generally falls between 0.2 and 0.7. See Cowell (2011) for a fuller discussion of inequality measurement.

56

Li Shi, Luo Chuliang, and Terry Sicular

.8 .6 .4 .2 0

Cumulative Income Share

1

Lorenz Curves

0

.2

.4

.6

.8

1

Cumulative Population Share 2002

2007

Figure 2.1. China’s National Lorenz Curves for Household Per Capita Income, 2002 and 2007. Note: Includes all provinces covered by the CHIP surveys, CHIP income definition, weighted by province, region, and urban/rural populations. Calculated using incomes in current-year prices.

more weight on income differences in the lower tail of the distribution, and the GE(1) places even weight on income differences across the income distribution. A graph of the Lorenz curves reveals the pattern of change in income distribution that underlies the increases in these inequality indices (Figure 2.1). The Lorenz curve for 2007 is everywhere lower than that for 2002, which is consistent with an increase in inequality as measured by the inequality indices in Table 2.1. Figure 2.2 shows the distribution of income across income decile groups, ordered from the poorest 10 percent to the richest 10 percent. The height of the light grey bars gives mean income by decile in 2002, and the height of the dark gray bars gives mean income by decile in 2007 (in constant 2002 prices). The black line shows the percentage increase in income between 2002 and 2007 (in constant prices) for each decile. It is clear from Figure 2.2 that income increased for all decile groups, but the increase was smaller for the poorer deciles than for the richer deciles. The income of the bottom decile increased by 401 yuan, or 45 percent (in constant 2002 prices). This is a substantial increase, but in both absolute

Overview

57

35,000

100

30,000

yuan

20,000

60

15,000

40

10,000

20

5,000

0 2002

percent

80

25,000

1 883

2

3

4

5

6

7

8

9

10

0

1,428 1,869 2,346 2,902 3,652 4,743 6,375 8,814 16,575

2007 (constant prices) 1,284 2,239 2,997 3,847 5,012 6,674 8,982 11,935 16,503 30,792 % change

45.4

56.7

60.3

64.0

72.7

82.7

89.4

87.2

87.2

85.8

Figure 2.2. Income levels and growth by deciles, 2002–2007. Note: Includes all provinces covered by the CHIP surveys, CHIP income definition, weighted by province, region, and urban/rural populations. Calculated using incomes in constant 2002 prices.

and relative terms it lags far behind that of the higher-income groups. The income of the top decile, for example, increased by more than 14,000 yuan, or 86 percent. Do these patterns of inequality reflect changes in the composition of income? Clues about the role of different income sources can be found in Table 2.2, which shows the income shares, Gini concentration ratios, and contributions to overall inequality of each component of per capita income. The contributions to inequality are calculated using the standard inequality decomposition by factor components (Shorrocks 1982). Looking first at urban incomes, one can see that the concentration ratio of urban household incomes is much higher than the Gini of the total income distribution, implying that on balance urban income was concentrated among higher-income groups. This was especially true for urban income from assets and imputed rent on owner-occupied housing. More generally, the numbers in Table 2.2 reveal the emergence of private property as a new and increasingly important source of inequality. Nationally, including both rural and migrant households, the contribution of assets and imputed rent to total inequality rose from 8 percent in 2002 to 13 percent in 2007. If calculated using alternative estimates of imputed rent, the contribution of assets and imputed rent to total inequality rose from 10 to 19 percent.11 11

The alternative estimates are calculated as the rate of return times the estimated market value of housing. This yields values of imputed rents that are higher and that increased more rapidly than the base estimates, which are calculated using estimated market rents. The alternative estimates give higher contributions of urban imputed rents to overall inequality: 8.80 percent in 2002 and 17.45 percent in 2007. See Chapter 3 for further discussion.

Table 2.2. Decomposition of inequality by income sources, 2002 and 2007 2002

2007

Contribution Contribution Concentration Share to total Concentration Share to total ratio or Gini (%) inequality (%) ratio or Gini (%) inequality (%) Rural total Wages from migrant jobs Other wages Net farm Net from nonfarm activities Assets Net transfers Imputed rent on owneroccupied housing Urban total Wages Pensions Net from individual businesses Assets Net transfers In-kind subsidies on public rental housing Imputed rent on owneroccupied housing Other in-kind income Migrants total Wages Net from individual businesses Assets Net transfers Imputed rent on owneroccupied housing National total

0.011 −0.066

35.87 4.07

0.89 −0.59

−0.101 −0.185

25.30 4.47

−5.30 −1.71

0.156 −0.129 0.206

8.77 14.23 4.69

2.97 −3.98 2.10

−0.017 −0.191 0.126

5.09 9.24 2.58

−0.18 −3.65 0.67

0.410 0.071 −0.013

0.24 1.51 2.35

0.22 0.23 −0.06

0.185 −0.089 −0.108

0.66 1.08 2.17

0.25 −0.20 −0.49

0.717 0.717 0.718 0.583

60.56 42.19 9.89 2.01

94.32 65.74 15.42 2.55

0.684 0.679 0.664 0.687

69.65 45.78 12.58 5.39

98.49 64.36 17.28 7.66

0.783 0.678 0.742

0.72 −0.39 1.68

1.23 −0.57 2.70

0.875 0.697 0.645

1.09 −3.75 0.41

1.98 −5.41 0.55

0.739

3.76

6.03

0.714

7.68

11.34

0.808

0.70

1.22

0.778

0.46

0.74

0.618 0.554 0.652

3.57 1.38 2.02

4.79 1.66 2.86

0.652 0.626 0.695

5.05 3.42 1.50

6.81 4.43 2.16

0.413 0.719 0.763

0.01 0.09 0.08

0.01 0.14 0.13

0.886 0.885 0.796

0.03 0.02 0.08

0.05 0.04 0.13

0.460

100

100

0.483

100

100

Note: CHIP income definition, including migrants, using three-level weights. Includes all provinces covered by the CHIP surveys. Calculated using incomes measured in current-year prices.

Overview

59

The negative contribution of urban net transfers (including both government and private transfers) is also noteworthy, especially in 2007 when they reduced total inequality by 5 percent. The increasingly equalizing role of urban net transfers likely reflects the expansion of government urban welfare programs, such as the urban minimum living standard guarantee program (see Chapter 7) and income taxes (see Chapters 7 and 10). The concentration coefficient of migrant income was similar to that of urban income, but owing to the small population and income share of migrants, the overall impact on national inequality remained small, although it increased over time. In Section V we discuss the income and inequality of migrants in more detail. In contrast, the concentration ratio of rural household income was close to zero in 2002 and became negative in 2007, implying that rural household income had an increasingly equalizing effect on total inequality. Income from farming was the most equalizing source of rural income. Income from short-term migrant work by rural household members was also equalizing and became more equalizing from 2002 to 2007. In-depth analysis of rural incomes and inequality can be found in Chapter 6. Most analyses of inequality in China do not adjust for differences in the cost of living among regions. The cost of living is typically higher in wealthier areas, therefore measured inequality will be overstated as it reflects price differentials as well as real differences in purchasing power. Table 2.3 presents a comparison of inequality estimates calculated with and without adjustments for PPP. In all cases, PPP adjustments reduce the measured level of inequality. For example, adjusting for PPP reduces the 2007 Gini coefficient by 12 percent, from 0.483 to 0.423. Although the measured level of inequality is lower with the PPP adjustment, it remains moderately high compared to inequality estimates for other countries (which typically are not adjusted for domestic price differentials). The 2007 Gini coefficient, for example, remains well above 0.40 regardless of whether it is calculated using NBS or CHIP income. Moreover, PPP adjustments do not alter the conclusion that inequality rose substantially between 2002 and 2007. In fact, the increase in PPP inequality was 8 percent, which is greater than the 5 percent increase for our non-PPP estimates.

V Household Income Growth and Inequality of Rural-Urban Migrants Because other chapters in this volume do not fully explore incomes and inequality among rural-urban migrants, here we include a separate analysis

60

Li Shi, Luo Chuliang, and Terry Sicular

Table 2.3. Inequality estimates with and without PPP adjustments, 2002 and 2007

2002 Without PPP

% change, 2002–2007

2007 With PPP

Without PPP

With PPP

Without PPP

With PPP

0.421 0.315 0.302

5.1 14.4 10.1

8.2 18.9 17.1

0.423 0.317 0.305

5.00 14.13 9.56

8.18 19.62 17.76

NBS Income Gini GE(0) GE(1)

0.455 0.361 0.356

0.389 0.265 0.258

0.478 0.413 0.392

CHIP Income Gini GE(0) GE(1)

0.460 0.368 0.366

0.391 0.265 0.259

0.483 0.420 0.401

Notes: 1. Includes all provinces covered by the CHIP surveys. 2. Calculated using three-level weights and including migrants. Incomes are in current-year prices. 3. For PPP estimates, incomes have been adjusted for differences in cost of living between urban and rural areas and among provinces using the Brandt and Holz (2006) geographic price indices for 2002 and updated to 2007 using the provincial rural and urban price indices published by the NBS. 4. Incomes less than or equal to zero have been dropped for calculation of the GE(0) and GE(1) inequality indices. See notes to Table 2.1.

of incomes and inequality for this group. Our analysis draws on data from the CHIP migrant surveys carried out in 2002 and 2007. As mentioned earlier, in our analysis we include only long-term, stable rural-urban migrants. Following the criteria used to classify individuals in the NBS household surveys (on which the CHIP surveys are based), we define long-term, stable rural-urban migrants as individuals whose origins are in rural areas, who have lived in cities for more than six months, and who are either single or living with a spouse. A detailed explanation of the classification criteria can be found in Appendix II to this volume. We note that limiting our analysis to long-term, stable migrants reduces the potential bias due to differences in the sampling methods used for the 2002 and 2007 migrant surveys. As noted earlier, the 2002 survey does not capture migrants who live in temporary or employer-provided housing. This group is largely composed of short-term, temporary migrants, whom we exclude from our long-term, stable migrant sample (but who are represented in the rural sample). Table 2.4 shows the level and composition of household per capita income of migrants. The mean income of the migrants falls between that of rural

Overview

61

Table 2.4. Level, composition, and growth of migrant household per capita income Mean income per capita (yuan)

Wage income Individual business net income Asset income Net transfer income Imputed rent on owner-occupied housing Total income

2002

2007

Annual income growth (constant 2002 prices)

2,768 4,050 13 177 159

11,294 4,953 99 75 252

29.4% 1.7% 47.3% −17.7% 7.1%

7,167

16,673

15.7%

Note: Includes all provinces covered by the migrant surveys, CHIP income definition, weighted by province and region using population shares of long-term stable migrants (see Appendix II to this volume). In current-year prices except for the real growth rates, which are deflated using the urban consumer price index.

and urban households. On average, in 2002 migrant income was 2.6 times rural per capita income and 80 percent of urban per capita income. In 2007, migrant per capita income was 3.6 times rural per capita income and 95 percent of urban per capita income. Migrants enjoyed rapid income growth between 2002 and 2007. On average, migrant per capita income in real terms grew at an annual rate of 16 percent, exceeding the growth rates of both rural and urban incomes. Thus, between 2002 and 2007 migrant income moved closer to that of urban households. To some extent, the higher growth rate of migrant income may be due to a self-selection process. Lowincome migrants are more likely to choose to return to their original homes, whereas high-income migrants are more likely to remain in the cities on a long-term basis. Looking at the growth by income component, we find that the wage income of migrants grew at a very rapid annual rate of 29 percent, so that its share of total migrant income rose from 39 percent in 2002 to 68 percent in 2007. As shown in Table 2.4, almost 90 percent of the total income growth can be attributed to the growth of wage income. Growth of income from individual businesses was slow, less than 2 percent annually. The rapid growth of wage income and slow growth of individual business income shown here to some extent may be due to the change in the migrant survey sampling procedure in the two years. In 2002 the survey was conducted in neighborhood communities (shequ) and did not include any migrant workers living in construction sites or factory dormitories; in 2007 migrants were selected based on employer records of migrant employees. This could lead to an underrepresentation of wage employees and an overrepresentation of

62

Li Shi, Luo Chuliang, and Terry Sicular

.8 .6 .4 .2 0

Cumulative Income Share

1

Lorenz Curves

0

.2

.4

.6

.8

1

Cumulative Population Share 2002

2007

Figure 2.3. Lorenz Curves of Migrant Per Capita Income, 2002 and 2007. Note: Includes all provinces covered by the CHIP migrant surveys, CHIP income definition, weighted by province and region using population shares of long-term, stable migrants (see Appendix II in this volume). Calculated using incomes in current-year prices.

self-employed migrants in 2002 as compared to 2007. Nevertheless, as discussed in Chapter 6, rapid growth in migrant wage income at this time likely also reflected real economic factors, in particular, growth in labor demand and increased reservation wages associated with higher farm earnings. Due to the increase in wages, which are relatively equally distributed, as well as growth in incomes overall, income inequality for migrants declined from 2002 to 2007, as shown by the Lorenz curves in Figure 2.3 and the inequality indices and inequality decomposition reported in Tables 2.5 and 2.6. Again, changes between 2002 and 2007 may in part reflect differences in the sampling procedures.12 How does the inclusion of long-term, stable rural-urban migrants affect national inequality? As shown in Table 2.1, the inclusion of these migrants reduces national inequality only slightly, by less than 1 percent in both years. Including migrants reduces national inequality because migrants tend to 12

If the share of each income component had remained the same in 2002 and 2007, the inequality of total migrant income would have increased by 4 percent. The analysis in Chapter 6 in this volume, however, suggests that some of the change in the structure of migrant income was likely due to real economic factors, not merely a sample bias.

Overview

63

Table 2.5. Migrant inequality, 2002 and 2007

Gini GE(0) GE(1)

2002

2007

% change, 2002–2007

0.336 0.200 0.193

0.289 0.144 0.154

−14.0% −28.0% −20.2%

Note: Includes all provinces covered by the migrant surveys, CHIP income definition, weighted by province and region using population shares of longterm stable migrants (see Appendix II to this volume). Calculated using current-year prices, but the level of inequality is the same for the current year and constant prices if deflation is carried out using the same consumer price index for all individuals.

fall in the center of the income distribution, but the reduction is minimal because the population share of long-term, stable migrants is relatively small, although increasing. According to data from the 2000 census, this group constituted 2.5 percent of the national population and 7.4 percent of the urban population. According to data from the 2005 mini census, this Table 2.6. Decomposition of migrant income inequality by income source, 2002 and 2007 2002

2007

Contribution Contribution to total to total Concentration Share inequality Concentration Share inequality ratio or Gini (%) (%) ratio or Gini (%) (%) Wage income Individual business net income Asset income Net transfer income Imputed rent on owneroccupied housing Total income

0.219 0.398

38.63 56.5

25.2 66.9

0.224 0.404

67.74 29.71

52.5 41.5

0.014 0.537

0.18 2.48

0.0 4.0

0.797 0.805

0.59 0.45

1.6 1.3

0.597

2.21

3.9

0.590

1.51

3.1

0.336

100

100

0.289

100

100

Note: Includes all provinces covered by the migrant surveys, CHIP income definition, weighted by province and region using population shares of long-term stable migrants (see Appendix II to this volume). Calculated using incomes in current-year prices; the level of inequality is the same for the current year and constant prices if deflation is carried out using the same consumer price index for all individuals.

64

Li Shi, Luo Chuliang, and Terry Sicular Table 2.7. Urban inequality with and without migrants, 2002 and 2007 2002

Gini GE(0) GE(1)

2007

Without

With

Without

With

0.331 0.182 0.186

0.305 0.156 0.157

0.340 0.193 0.199

0.317 0.169 0.174

Note: Includes all provinces covered by the surveys in both years, CHIP income definition, weighted by province and region using the population shares of urban locals and long-term stable migrants (see Appendix II to this volume). Calculated using incomes in current-year prices.

group constituted 3.2 percent of the national population and 7.6 percent of the urban population (see Appendix II in this volume). If we limit our attention to the urban sector, within which the migrants constitute a larger share of the population, the inclusion of long-term, stable migrants when estimating inequality has a greater impact (Table 2.7). In 2002 the inclusion of migrants reduced urban inequality by 8 percent, and in 2007 by 7 percent. We note that the difference between inequality calculated with and without migrants is not the same as measuring the full impact of migration on inequality. Migration can influence income levels of urban and rural households, and likely has different impacts in richer and poorer areas. Fully analyzing the impact of migration would require estimating the counterfactual income levels that would have prevailed had migration not taken place. Our calculations use only the actual income levels.

VI. The Structure of Inequality: The Urban-Rural Income Gap Analyses of inequality in China typically highlight the widening gap between urban and rural household incomes. Most studies, including those based on earlier rounds of the CHIP survey, find that the urban-rural income gap has widened over time and that it has contributed to the increase in overall inequality. Here we examine changes in the urban-rural income gap between 2002 and 2007. In our analysis we use the NBS and CHIP definitions of income. We note that these measures of income do not fully capture implicit subsidies that are disproportionately enjoyed by urban residents, and which if included would widen the urban-rural differential (Li and Luo 2010). We do, however, show estimates of the urban-rural gap that are adjusted for

Overview

65

Table 2.8. The urban-rural income gap, 2002 and 2007 Mean income per capita (yuan) 2002

2007

Average annual income growth (constant 2002 prices)

Urban-rural income ratio

Urban-rural income ratio (PPP adjusted)

2002

2007

2002

2007

NBS Income Urban, without migrants Urban, with migrants Rural

8,078 15,469

11.26%

3.16

3.66

2.13

2.61

8,005 15,537

11.56%

3.13

3.68

2.10

2.60

2,590

4,221

6.96% CHIP Income

Urban, without migrants Urban, with migrants Rural

9,002 17,639

11.77%

3.25

3.82

2.21

2.71

8,875 17,570

12.00%

3.20

3.80

2.17

2.68

2,771

4,618

7.44%

Note: Unadjusted current-year prices unless noted otherwise. Includes all provinces covered in the CHIP surveys; calculated using regional and provincial population weights. PPP estimates are calculated using incomes that have been adjusted for differences in cost of living between urban and rural areas and among provinces using the Brandt and Holz (2006) geographic price indices for 2002, and updated to 2007 using the provincial rural and urban price indices published by the NBS.

differences in the cost of living between urban and rural areas. The adjustment reduces the size of the urban-rural income gap (Sicular et al. 2010). We measure the urban-rural income gap as the ratio of average disposable income per capita of households in the urban survey, or in the combined urban and migrant surveys, to average net income per capita of households in the rural survey. Average incomes of each sector are calculated using weights. We find that the urban-rural income gap continued to widen between 2002 and 2007 (Table 2.8). The widening gap is not due to slow growth in rural incomes – rural incomes in fact grew rapidly during this period (see Chapter 5) – but reflects even faster growth in urban incomes. Calculated using CHIP income and including migrants, the gap increases by about 20 percent from 3.2 to 3.8. This urban-rural gap is high by international standards. Available estimates for other countries indicate that urban-rural income ratios above 3.0 are rare. For India, Bangladesh, Indonesia, and Malaysia the ratio is less

66

Li Shi, Luo Chuliang, and Terry Sicular Table 2.9. Contribution of urban-rural (between-group) inequality to national inequality (%) NBS income definition

Without migrants GE(0) GE(1) With migrants GE(0) GE(1)

CHIP income definition

2002

2007

2002

2007

43.1 44.0

49.3 48.0

45.6 46.3

52.0 50.7

42.9 43.5

49.6 48.1

44.5 44.5

50.9 48.5

Note: Calculations with migrants include in the urban sector long-term, stable migrants from the rural areas. Three-level weights are used. Calculated using incomes measured in current-year prices. See Shorrocks (1980) for a discussion of the decomposition methodology.

than 2.0; for Thailand and the Philippines the ratio is about 2.3. Only for a few countries, such as South Africa and Zimbabwe, does the ratio exceed 3.0 (Knight and Song 1999: 138; World Bank 2009b). Alternative calculations change the size of the gap, but in all cases the gap widens from 2002 to 2007. Excluding migrants increases the size of the income gap somewhat but does not substantially change the trend. The income gap is smaller for NBS income than for CHIP income, but in both cases the gap widens over time. Adjusting for cost-of-living differences substantially reduces the magnitude of the urban-rural income gap. Measured using PPP-adjusted CHIP incomes and including migrants, in 2002 the urban-rural income ratio is 2.2, and in 2007, 2.7. Again, the urban-rural income ratio widened, increasing by 24 percent between the two years. The widening urban-rural gap was a factor underlying rising national inequality. Table 2.9 presents summary results of a standard inequality decomposition by population subgroup using the Theil inequality measures (Shorrocks 1980).13 This method disaggregates overall inequality into the contributions of inequality between groups and within groups. In our application, the groups are urban and rural. Between-group inequality is the component associated with the urban-rural income gap. 13

The Gini coefficient is not decomposable by population subgroup.

Overview

67

Table 2.10. Contributions of urban-rural (between-group) inequality to national inequality, with PPP adjustments (%) NBS income definition

Without migrants GE(0) GE(1) With migrants GE(0) GE(1)

CHIP income definition

2002

2007

2002

2007

25.7 27.2

35.4 35.9

28.3 29.9

38.6 39.0

25.2 26.6

35.6 35.8

27.1 28.4

37.5 37.3

Note: The note to Table 2.9 applies. PPP adjustments for 2002 use the Brandt and Holz (2006) price deflators; for 2007, the Brandt and Holz (2006) deflators are updated using the NBS provincial urban and rural consumer price indices.

We report the results for the two Theil measures of inequality, for both the NBS and CHIP income definitions, and without and with migrants.14 In all cases, the share of national inequality contributed by between-group inequality increased from 2002 to 2007. In 2002, between-group inequality contributed 43 to 46 percent of overall inequality. In 2007, between-group inequality contributed 48 to 52 percent of overall inequality, an increase of about 5 percentage points over 2002. Thus, by 2007 the urban-rural income gap was associated with roughly half of the national inequality in China. PPP adjustments reduce the contribution of the urban-rural gap to inequality, but exacerbate the increase in the contribution of the urbanrural gap to inequality over time (Table 2.10). For the CHIP measure of income and including migrants, in 2002 the urban-rural gap contributed 27 to 28 percent of PPP inequality, and by 2007 it contributed more than 37 percent.

VII. The Structure of Inequality: Regional Income Differences Previous studies note large regional disparities in household incomes in China. Analysis of the 2002 CHIP data identified large regional gaps, but with some evidence of a regional catch-up (Gustafsson et al. 2008a). To investigate regional income inequalities between 2002 and 2007, we conduct several computations. Following the CHIP sampling approach as well as the 14

We also carried out the decomposition using alternative weights. The results are similar, so we do not report them here.

68

Li Shi, Luo Chuliang, and Terry Sicular Table 2.11. Regional income gaps, 2002 and 2007 PPP unadjusted 2002

Region Large municipalities Eastern Central Western

2007

Urban

Rural

Migrant

All

Urban

Rural

Migrant

All

2.30

2.69

1.40

4.03

2.12

3.34

1.39

3.54

1.37 0.92 1.00

1.98 1.22 1.00

1.37 0.89 1.00

1.86 1.10 1.00

1.67 1.05 1.00

1.82 1.21 1.00

1.23 0.85 1.00

2.05 1.17 1.00

PPP adjusted 2002 Region Large municipalities Eastern Central Western

2007

Urban 1.56

Rural 1.70

Migrant 0.95

All 2.51

Urban 1.58

Rural 2.23

Migrant 1.04

All 2.43

1.12 0.90 1.00

2.00 1.29 1.00

1.03 0.87 1.00

1.63 1.12 1.00

1.43 1.03 1.00

1.77 1.21 1.00

1.00 0.84 1.00

1.76 1.15 1.00

Note: Income gaps are equal to the ratio of each region’s income per capita to that in the western region. In this table long-term stable migrants are shown separately, and urban excludes migrants. CHIP income definition; calculated using three-level weights for all and regional × provincial weights for the urban, rural, and migrant subgroups; current-year prices. See notes to previous tables regarding PPP adjustments.

official classification of regions, in these computations we divide China into four regions: large, provincial-level metropolitan cities; the eastern region; the central region; and the western region. Table 2.11 shows the relative incomes of the four regions, calculated as a ratio using the mean income of the western region as the denominator. All calculations use the CHIP income definition (see Table 2A.3 for mean incomes per capita by region). We present alternative estimates using unadjusted prices (current-year prices, no adjustments for regional cost-of-living differences) and PPP prices (current-year prices, adjusted for regional cost-of-living differences). Costs of living are generally higher in more developed regions, so using the PPP prices reduces the income differences between the richer and poorer regions. As shown in Table 2.11, PPP adjustments markedly reduce regional income ratios between the large municipalities and the western regions and between the eastern and western regions, but they do not substantially change the income ratio between the central and western regions.

Overview

69

Looking at the PPP estimates, we find the largest income ratio to be between the large municipalities and the western region. In 2002 per capita incomes in the large municipalities were on average 2.5 times those in the western region; in 2007 the ratio narrowed slightly to 2.4. The ratio between the eastern and western regions was smaller but also substantial; the ratio between the central and western regions was fairly small. The regional structure of PPP incomes differs somewhat for the urban, migrant, and rural subpopulations. Regional income differences are largest for rural residents. With the exception of large municipalities, rural regional income differences narrowed between 2002 and 2007. The narrowing of rural regional income differences might reflect the equalizing effects of migration, or the effects of increased returns to farming (see Chapter 5), which could narrow the gap between areas with more and less nonagricultural development. In urban areas, the regional income gaps all widened. Our estimates indicate that income growth of urban households in the western provinces lagged behind that of urban households in other regions during the period under study. Regional income differences among urban-based migrant households are small. Even between the large metropolitan cities and the western region, in 2007 the income gap is less than 5 percent. There is almost no regional income gap for migrants between the eastern and western regions, and migrant incomes in central China are 6 percent lower than those in western China. The lack of substantial regional income differences for migrants may reflect the equalizing effect of migration among regions as migrants move in response to real differentials in their wages. Overall, then, it appears that the widening of the overall regional income gaps in China between 2002 and 2007 was largely driven by regional trends among urban areas, and between the large municipalities and the rest of China. Income gaps among other regions and groups were relatively stable or narrowed. How important is interregional inequality to overall inequality in China? We address this question using standard inequality decomposition analysis by population subgroup of the Theil inequality indices. Here the relevant groups are the four regions. The contribution of between-group inequality captures the importance of regional income differences to overall inequality in China. Table 2.12 shows estimates of the contribution of between-group (interregion) inequality to inequality for China as a whole (“all”) and for the urban, rural, and migrant populations. The table reports estimates calculated with and without PPP adjustments, but our discussion focuses on the

70

Li Shi, Luo Chuliang, and Terry Sicular

Table 2.12. Contributions of between-region inequality to overall inequality (%) PPP unadjusted 2002

GE(0) GE(1)

2007

Urban

Rural

Migrant

All

Urban

Rural

Migrant

All

19.4 20.5

20.7 20.0

7.8 7.6

17.6 19.9

18.0 17.6

16.9 17.2

7.3 6.5

15.5 17.1

Migrant 1.3 1.1

All 11.3 12.1

PPP adjusted 2002 GE(0) GE(1)

Urban 6.9 7.2

Rural 19.1 18.6

Migrant 1.0 1.0

2007 All 11.6 12.5

Urban 9.1 8.9

Rural 13.5 13.6

Note: The contributions of the differences in mean incomes among the four regions to national inequality are shown in the column titled “All.” The other columns report the contributions of income differences among the four regions to inequality in each of the urban, rural, and migrant subgroups. CHIP income definition; calculated using three-level weights for all and regional × provincial weights for the urban, rural, and migrant subgroups; current-year prices.

PPP estimates, for which incomes are more comparable among regions and between urban and rural areas. For China as a whole, the share of between-region inequality is relatively low, contributing 11 to 12 percent of overall inequality, and with a very slight decrease between 2002 and 2007. In other words, in both years within-region inequality accounts for the overwhelming majority of national inequality. As one might expect, regional income differences are most important for rural inequality, although over time their contribution declined. In 2002 between-region inequality contributed 19 percent, and in 2007 less than 14 percent of rural inequality. The declining contribution of regional income differentials to rural inequality likely reflects the spread of nonagricultural employment opportunities from the eastern areas to the central and western areas, as well as the increased migration by rural workers in the western region. For the formal urban population, between-region differences contributed a smaller but growing share of inequality. These results could reflect continuing or perhaps increasing segmentation of the formal urban labor markets, as well as regional immobility caused by rapidly rising housing costs in the large metropolitan cities. Regional inequality is unimportant among migrant households. As shown in Table 2.12, between-region income inequality as a percentage of total inequality among migrants was only about 1 percent in both years.

Overview

71

Table 2.13. Gini coefficients by region, 2002 and 2007 PPP unadjusted

Large municipalities Eastern Central Western

2002

2007

0.321 0.418 0.398 0.456

0.315 0.456 0.428 0.471

PPP adjusted Large municipalities Eastern Central Western

2002 0.311 0.352 0.346 0.422

2007 0.307 0.400 0.381 0.421

Note: CHIP income definition; incomes are in current-year prices. Calculated using provincial and rural/urban weights. Long-term, stable migrants are included in these calculations.

The findings in Table 2.12 indicate that national inequality is driven more by inequality within regions than by inequality between regions. Table 2.13, which shows the levels of inequality within regions, reveals that within-region inequality has remained particularly high in western China. Within-region inequality increased in eastern and central China between 2002 and 2007, but the increase was most marked – more than 13 percent – in eastern China. Inequality within regions is in part a reflection of the large urban-rural income gap discussed in the previous section. In both 2002 and 2007 the urban-rural income gap was largest in the western region, about 3 with the PPP adjustments (3.7 to 3.9 without the PPP adjustments; Table 2.14). In the eastern and central regions the urban-rural gap was moderate in 2002 but increased substantially between 2002 and 2007. In large metropolitan cities the urban-rural income gap shrank between 2002 and 2007, so that by 2007 the large metropolitan cities had the smallest urban-rural income ratio, although it still remained at 2.0.15 This decline may reflect the economic development of rural districts within the large metropolitan cities and their increased urban integration. 15

Urban administrative areas in China often include not only urbanized districts but also farmland and rural populations, following the Mao-era practice of incorporating surrounding rural areas into city administration. See Chan (2010).

72

Li Shi, Luo Chuliang, and Terry Sicular Table 2.14. The urban-rural income gap by region, 2002 and 2007 PPP unadjusted

Large municipalities Eastern Central Western

2002

2007

3.08 2.58 2.81 3.73

2.34 3.44 3.32 3.85

PPP adjusted Large municipalities Eastern Central Western

2002 2.62 1.64 2.08 2.97

2007 2.00 2.32 2.52 2.96

Note: See notes to Table 2.13 and notes to previous tables regarding PPP adjustments. CHIP income definition. Migrants are included as urban residents in the calculations.

Based on the preceding regional analysis, we conclude that income differences between the eastern, central, and western regions are not a major source of nationwide inequality. Within-region income differences are much more important, although less so in the large metropolitan cities than in the eastern, central, and western regions. Urban-rural inequality appears to be a contributing factor to the rising inequality in the latter three regions.

VIII. Poverty During the reform era China has achieved dramatic and ongoing reductions in poverty. By 2002 the poverty rate was already quite low, and further poverty reduction became more challenging due to several factors, for example, the fact that a high proportion of the remaining poverty was geographically dispersed and transient, and because poverty had become less responsive to macroeconomic growth (World Bank 2009a). Policies adopted after 2002, such as the minimum living standard guarantee program, the new rural cooperative medical insurance system, and the new rural pension system, have addressed some of these factors. Here we examine trends in poverty between 2002 and 2007 to understand the net effects of policies and growth on poverty. Studies of poverty have used

Overview

73

different poverty lines and poverty measures. We present three alternative estimates of poverty, two using absolute poverty lines and one using a relative poverty line. For all estimates we use the NBS definition of income, which does not include imputed rents on owner-occupied housing. We exclude imputed rents because the poverty lines are set without reference to imputed rents. The first absolute poverty line is the international PPP poverty threshold of $1.25 per day per person, which we convert to yuan using the PPP exchange rate of 3.46 yuan to the U.S. dollar in 2005 (Chen and Ravallion 2008). The second absolute poverty line is the Chinese government’s official poverty line for rural areas. In view of past criticisms that the Chinese official poverty line in 2007 and earlier years was too low, and because the official poverty line was increased substantially in 2008, we use the higher 2008 official poverty line of 1,196 yuan. We treat both of these poverty lines as rural poverty lines and convert them to 2002 and 2007 prices using the NBS consumer price index for rural areas. We set the urban absolute poverty lines equal to the rural poverty lines adjusted by the urban-rural cost-of-living differential (taken from Brandt and Holz [2006] and, for 2007, updated using the NBS consumer prices indices). Relative poverty lines are used fairly often, especially in higher-income countries where few households experience absolute deprivation but where individuals at the lower end of the income distribution are nevertheless disadvantaged (Osberg 2000; Ravallion 1992). In view of the substantial growth in personal incomes in China in recent decades, the concept of relative poverty has become increasingly relevant. Following common practice in the literature, we use a relative poverty line equal to 50 percent of the median income. The relative poverty lines are set at 50 percent of the median income in each of the rural and urban sectors, with long-term, stable migrants included in the urban sector. Table 2.15 shows our poverty lines expressed in current prices for each year. We note that Chapters 5 and 7 in this volume provide more detailed, separate analyses of poverty in the rural and urban sectors. Due to differences in calculation, in some cases the levels of poverty reported in these chapters may differ from those reported here; however, the overall trends between 2002 and 2007 are similar. Our estimates of poverty incidence appear on the top half of Table 2.16. For China as a whole, absolute poverty declined quite substantially between 2002 and 2007. Using the PPP $1.25 poverty line, for example, the poverty rate fell from 19 percent to 8 percent. Underlying this reduction is a marked decline in rural poverty. Absolute poverty in the formal urban and migrant populations also declined, but was already low in 2002.

74

Li Shi, Luo Chuliang, and Terry Sicular Table 2.15. Poverty lines (yuan) Official poverty line

PPP $1.25/day

50% of median income

2002

2007

2002

2007

2002

2007

964 1,338

1,123 1,503

1,451 2,013

1,689 2,260

1,051 3,379

1,714 6,412

Rural Urban, with migrants

Notes: 1. For the official poverty line we use the 2008 official poverty line, adjusted for inflation (see the text). The international PPP poverty threshold of $1.25 per day per person is converted to yuan using the PPP exchange rate of 3.46 yuan to the U.S. dollar in 2005 (Chen and Ravallion 2008). 2. We treat both the official poverty line and the PPP $1.25/day poverty line as rural poverty lines and convert them to 2002 and 2007 prices using the NBS rural consumer price index. Urban absolute poverty lines are equal to the rural poverty lines adjusted by the urban-rural cost-ofliving differential of 1.3876 in 2002 and 1.3382 in 2007 (taken from Brandt and Holz [2006], and for 2007 updated using NBS consumer price indices). 3. The relative poverty lines are calculated separately for urban and rural. Median incomes for rural and urban (including migrants) are calculated using regional × provincial weights and the NBS income definition. 4. All poverty lines are in current-year prices.

Table 2.16. Poverty incidence and composition, 2002 and 2007 (%) Official poverty line 2002

2007

PPP$1.25/day 2002

50% of median income

2007

2002

2007

13.88 0.44 0.17 0.42 8.00

13.69 11.88 18.57 12.34 13.21

14.32 12.37 7.00 11.98 13.30

97.70 2.23 0.07 2.30 100

66.52 30.01 3.47 33.48 100

60.63 37.73 1.64 39.37 100

Poverty incidence Rural Urban Migrants Urban + migrants Total

11.22 0.55 2.43 0.68 7.44

5.59 0.12 0.08 0.12 3.20

27.49 2.34 5.80 2.58 18.57

Poverty composition Rural Urban Migrants Urban + migrants Total

96.72 2.48 0.80 3.28 100

98.35 1.57 0.08 1.65 100

95.02 4.21 0.77 4.98 100

Note: Calculated using three-level weights for total and regional × provincial weights for subgroups. NBS income definition; current-year prices.

Overview

75

Table 2.17. The structure of poverty by region (%) Official poverty line 2002

2007

PPP$1.25/day 2002

50% of median income

2007

2002

2007

0.35 3.74 7.47 14.77 8.00

0.89 7.73 14.21 20.49 13.21

1.87 7.78 12.81 21.99 13.30

0.14 16.51 30.94 52.40 100

0.21 21.19 34.91 43.69 100

0.44 20.65 31.94 46.96 100

Poverty incidences Large municipalities Eastern Central Western Total

0.07 3.77 6.98 13.53 7.44

0.09 1.59 2.74 6.07 3.20

0.70 8.80 19.87 31.64 18.57

Poverty composition Large municipalities Eastern Central Western Total

0.03 18.33 30.42 51.22 100

0.09 17.59 28.41 53.91 100

0.12 17.16 34.71 48.00 100

Note: Calculated using three-level weights for total and regional × provincial weights for subgroups. NBS income definition; current-year prices. A single relative poverty line calculated as 50 percent of the national median income is used for all regions.

In contrast, the rate of relative poverty in China as a whole remained more or less unchanged at 13 percent. Stagnant relative poverty rates suggest that households at the lower tail of the income distribution were not catching up to the median. This is consistent with our finding of increased inequality, as discussed earlier. Relative poverty rates are fairly similar for the rural and urban areas, except for migrants within the urban areas. For this group, relative poverty was higher in 2002, but by 2007 it had declined and was below the relative poverty rates for the rural and formal urban populations. For all poverty lines, the overwhelming majority of the poor were rural (as shown in the bottom half of Table 2.16). Using absolute poverty measures, more than 95 percent of the poor were rural. Using the relative poverty measure, the share of the rural poor is lower, although still high at 60-plus percent. Because the urban relative poverty lines are equal to 50 percent of the median urban income, and thus higher than the rural relative poverty lines, it is not surprising that by this measure a greater proportion of the relative poor than the absolute poor are located in the cities. Moreover, the share of the relative poor located in the cities increased noticeably between 2002 and 2007. Poverty rates differed greatly among regions. As shown in Table 2.17, the incidence of absolute poverty in the large municipalities was extremely

76

Li Shi, Luo Chuliang, and Terry Sicular

low; in the eastern region it was also relatively low, especially in 2007. The incidence of absolute poverty was higher in the central region and highest in the western region, although in both places it declined substantially between 2002 and 2007. In the western region the rate of absolute poverty, measured using PPP$1.25 per day, declined from 32 percent to 15 percent. Relative poverty was very low in the large municipalities, somewhat low in the eastern region, moderate in the central region, and highest in the western region, where more than 20 percent of the population fell below the relative poverty line. Relative poverty nationwide and in all regions was fairly stable between 2002 and 2007. Note that we use the same relative poverty line (50 percent of national median income) for all regions. By all measures, China’s poor are heavily concentrated in the West. As shown on the bottom half of Table 2.17, half of China’s absolute poor and well over 40 percent of the relative poor live in the western region. Moreover, from 2002 to 2007 the western region’s share of the poor increased. Less than 1 percent of China’s poor live in the large municipalities; 15 to 20 percent live in the eastern region; and about one-third live in the central region. This regional structure suggests the need for focused attention on poverty alleviation in the western and central regions. We note further that within all regions poverty was largely rural. For example, in 2007 in all regions including the western region, rates of absolute poverty measured using $1.25 per day for formal urban residents and for long-term migrants were all less than 1 percent. In the large municipalities the rate of rural poverty was also less than 1 percent. In contrast, in the eastern, central, and western regions the rates of rural poverty were 7 percent, 12 percent, and 22 percent, respectively. Again, this pattern has implications for the design of poverty alleviation programs.

IX. Conclusion Despite official policies emphasizing shared growth during the Hu Jintao– Wen Jiabao period, between 2002 and 2007 income inequality in China resumed its upward trajectory. By 2007 the level of inequality in China was moderately high by international standards. With a Gini coefficient of approximately 0.5, China was in the same ballpark as countries in South and Central America such as Mexico (0.51), Nicaragua (0.52), and Peru (0.48), although the level of inequality was still below that of the high-inequality countries such as Brazil and Honduras in the range of 0.56 to 0.57.16 16

The Gini coefficients for the other countries reported here are for 2005 and are measured over household income per capita. They are from the UNU-WIDER WIID2c database,

Overview

77

Our analysis reveals some old and some new factors that have contributed to this increase in inequality. An old factor is China’s already large urban-rural income gap. The urban-rural gap widened further between 2002 and 2007. Even after adjusting for differences in costs of living, the difference between urban and rural incomes was high by international standards and contributed a substantial share of national inequality. A new factor contributing to the rising inequality was income from property and assets. By 2007, with the completion of the urban housing privatization and the development of urban residential real estate markets, expansion of stock and capital markets, growth of private enterprises, and other property rights reforms, income from property and assets was beginning to be important. We find that in 2007 asset and property income contributed to both the urban-rural income gap and to overall inequality nationwide. In the future, the importance of asset and property income is likely to grow and may continue to drive up inequality in China. Inequality in these sources of income is potentially a hot-button issue, as in China the institutions that shape the distribution of assets are not yet transparent or equitable. We find evidence that some equalizing factors have also been at work. Although they did not fully offset the dis-equalizing factors, they nevertheless moderated the upward trend. In 2007 urban net transfers began to have a modestly equalizing impact. This category of income includes public transfers, thereby suggesting that the expansion of urban social welfare programs has played a positive role. Rapid growth in rural incomes, even if not as rapid as urban income growth, also moderated inequality. From the perspective of inequality, growth in rural incomes from farming and short-term migration was especially important. Some dimensions of regional inequality narrowed, for example, between-region rural inequality. These findings suggest that farm supports and regional development programs may have moderated income disparities, especially in rural China. We note that our estimates likely understate the real trends in inequality because high-income urban households are increasingly underrepresented in the NBS urban survey sample and because the income of high-income households is likely understated. These are common problems in household surveys. The problem is relatively recent in China, and future sampling methods and analytical approaches will need to adapt. A preliminary study by Li and Luo (2011) indicates that adjustments to correct for the undercounting of income of high-income urban households would increase the at http://www.wider.unu.edu/research/Database/en GB/wiid/. Accessed August 12, 2011. Note that the Gini coefficients for Brazil and Honduras are the highest among all countries listed in this database for 2005–2006.

78

Li Shi, Luo Chuliang, and Terry Sicular

Gini coefficient by 8 percentage points in urban areas and by 5 percentage points nationwide. Between 2002 and 2007 China achieved major gains in poverty reduction. Despite new challenges in poverty alleviation, during this period absolute poverty continued its downward trend. Relative poverty, however, did not decline, indicating that households at the bottom of the income distribution were not catching up with those at the middle or the top. As China’s economy matures and the number of absolute poor shrinks, relative poverty will become an increasingly important social indicator. In summary, then, we find that although households in all income groups, sectors, and regions continued to enjoy substantial income growth during this period, income growth was faster for richer households than for poorer households. The resulting increase in inequality reflected shifts in the structure of the income distribution and the emergence of some new underlying mechanisms. China thus faces ongoing challenges in its efforts to promote growth with equity. In the future, China’s distributional policies will need to evolve accordingly.

APPENDIX Table 2A.1. Income and inequality with alternative weights, 2002 and 2007 2002

Urban

Rural

National (excluding migrants)

2007

Migrant

National (including migrants)

Urban

Rural

National (excluding migrants)

17,527 0.340 0.193 0.196

5,106 0.377 0.239 0.250

9,587 0.480 0.407 0.397

16,048 0.308 0.163 0.173

9,982 0.475 0.400 0.385

5,106 0.377 0.239 0.250

10,322 0.476 0.405 0.386

16,048 0.308 0.163 0.173

10,501 0.472 0.400 0.380

Migrant

National (including migrants)

No Weights 79

Mean income Gini GE(0) GE(1)

8,504 0.325 0.176 0.180

2,773 0.364 0.225 0.238

4,791 0.454 0.360 0.355

6,180 0.349 0.214 0.212

4,858 0.450 0.355 0.348

Weight I (Urban/Rural) Mean income Gini GE(0) GE(1)

8,504 0.325 0.176 0.180

2,773 0.364 0.225 0.238

4,740 0.455 0.360 0.356

6,180 0.349 0.214 0.212

4,776 0.453 0.357 0.352

17,527 0.340 0.193 0.196

(continued)

Table 2A.1 (continued) 2002

Urban

Rural

National (excluding migrants)

2007

Migrant

National (including migrants)

Urban

Rural

National (excluding migrants)

Migrant

National (including migrants)

4,659 0.367 0.227 0.236

9,746 0.479 0.411 0.394

16,785 0.295 0.149 0.159

9,966 0.476 0.408 0.388

10,072 0.487 0.424 0.409

16,673 0.289 0.144 0.154

10,277 0.483 0.420 0.401

Weight II (Urban/Rural × Region)

80

Mean income Gini GE(0) GE(1)

8,800 0.326 0.177 0.180

2,815 0.365 0.227 0.239

4,862 0.458 0.366 0.362

6,691 0.343 0.206 0.203

4,907 0.456 0.364 0.358

16,805 0.337 0.190 0.196

Weight III (Urban/Rural × Province × Region) Mean income Gini GE(0) GE(1)

9,002 0.331 0.182 0.186

2,771 0.354 0.213 0.226

4,958 0.462 0.370 0.370

7,167 0.336 0.200 0.193

4,902 0.460 0.368 0.366

17,639 0.340 0.193 0.199

4,617 0.358 0.216 0.226

Notes: 1. Includes all provinces covered by the CHIP. Calculated using current-year prices and CHIP income. 2. The inequality indices shown in this table are all scale-invariant. Consequently, the level of inequality is the same for both the current year and constant prices (if deflation is carried out using the same price index for all individuals). 3. Incomes less than or equal to zero have been dropped for calculation of the GE(0) and GE(1) inequality indices (fewer than 30 observations [individuals] were dropped in 2002 and fewer than 225 in 2007).

Table 2A.2. Income and inequality with alternative estimates of imputed rental income on owner-occupied housing, 2002 and 2007 2002

Mean value of urban imputed rents on owner-occupied housing Mean value of urban income per capita Urban-rural income ratio

2007

A

B

A

B

558

860

1945

3229

9,002 3.25

9,303 3.36

17,638 3.82

18,922 4.10

0.340 0.193 0.199

0.337 0.190 0.197

0.483 0.420 0.401

0.492 0.440 0.416

Inequality within urban areas (migrants excluded) Gini G(0) G(1)

0.331 0.182 0.186

0.327 0.178 0.182

National inequality (migrants included) Gini G(0) G(1)

0.460 0.368 0.366

0.464 0.375 0.372

Note: Column A contains estimates that use the rate of return approach to calculate rural imputed rents and the market rent approach to calculate urban imputed rents. Estimates reported elsewhere in this chapter follow this approach. Column B contains alternative estimates that use the rate of return approach for both rural and urban areas. (See Chapter 3 for a detailed discussion of the two approaches.) Three-level weights, CHIP income, and current prices are used in all calculations.

Table 2A.3. Mean income per capita by region, 2002 and 2007 (yuan) PPP unadjusted 2002

2007

Region

Urban Rural Migrant

Large municipalities Eastern Central Western

17,022 10,155 6,790 7,390

5,267 3,869 2,391 1,955

All

Urban

Rural

Migrant

8,206 13,902 27,780 11,436 8,052 6,402 21,909 6,233 5,206 3,781 13,790 4,140 5,881 3,450 13,113 3,426

All

19,930 24,143 17,653 13,994 12,202 7,971 14,335 6,814

PPP adjusted 2002 Region Large municipalities Eastern Central Western

Urban 9,577 6,836 5,535 6,129

Rural Migrant 3,477 4,617 4,076 4,986 2,640 4,230 2,039 4,853

2007 All 7,930 5,153 3,552 3,162

Urban 16,876 15,278 11,063 10,707

Rural Migrant All 8,103 12,161 14,867 6,418 11,701 10,742 4,380 9,824 7,031 3,630 11,648 6,106

Note: In this table long-term stable migrants are shown separately, and urban excludes migrants. CHIP income definition; calculated using weights (three-level weights for all, provincial and regional weights for urban, rural, and migrant); current-year prices. See notes to Table 2.3 regarding PPP adjustments.

82

Li Shi, Luo Chuliang, and Terry Sicular References

Brandt, L. and C.A. Holz (2006), “Spatial Price Differences in China: Estimates and Implications,” Economic Development and Cultural Change, 55(1), 43–86. Chan, K.W. (2010), “Fundamentals of China’s Urbanization and Policy,” China Review, 10(1), 63–94. Chen, S. and M. Ravallion (2008), “China Is Poorer Than We Thought, But No Less Successful in the Fight against Poverty,” World Bank Policy Research Working Paper No. 4621, The World Bank, Washington, DC. Chung, J.H., H. Lai, and J.H. Joo (2009), “Assessing the ‘Revive the Northeast’ (zhenxing dongbei) Programme: Origins, Policies and Implementation,” China Quarterly, no. 197, 108–125. Cowell, F. (2011), Measuring Inequality, New York: Oxford University Press. D´emurger, S., J.D. Sachs, W.T. Woo, S. Bao, G. Chang, and A. Mellinger (2002), “Geography, Economic Policy, and Regional Development in China,” Asian Economic Papers, 1(1), 146–197. Fan S., R. Kanbur, and X. Zhang (2010), “China’s Regional Disparities: Experience and Policy,” Working Paper No. 2010-03, Department of Applied Economics and Management, Cornell University, Ithaca, NY. Fang, X., Y. Zhang, and J. Li (2007), “Xibu da kaifazhong de zhongyang caizheng zhuanyi zhifu zhengce yanjiu” (Study on the Fiscal Transfer Policy of the Central Government in Western Development), Xi’nan nongye daxue xuebao, no. 5, 34–37. Gustafsson, B., S. Li, and T. Sicular (2008a), “Inequality and Public Policy in China: Issues and Trends,” in B. Gustafsson, S. Li, and T. Sicular, eds., Income Inequality and Public Policy in China, 1–34, New York: Cambridge University Press. Gustafsson, B., S. Li, and T. Sicular, eds. (2008b), Income Inequality and Public Policy in China, New York: Cambridge University Press. Kanbur, R. and X. Zhang (2009), “Fifty Years of Regional Inequality in China: A Journey through Central Planning, Reform, and Openness,” in S. Fan, R. Kanbur, and X. Zhang, eds., Regional Inequality in China: Trends, Explanations and Policy Responses, 45–63, London: Routledge. Khan, A.R. (1993), “The Determinants of Household Income in Rural China,” in K. Griffin and R. Zhao, eds., The Distribution of Income in China, 95–115, Basingstoke: Macmillan. Khan, A.R., K. Griffin, C. Riskin, and R. Zhao (1992), “Household Income and Its Distribution in China,” China Quarterly, no. 132, 1029–1061. Khan, A.R. and C. Riskin (1998), “Income and Inequality in China: Composition, Distribution and Growth of Household Income, 1988 to 1995,” China Quarterly, no. 154, 221–253. Khan, A.R. and C. Riskin (2008), “Growth and Distribution of Household Income in China between 1995 and 2002,” in B. Gustafsson, S. Li, and T. Sicular, eds., Inequality and Public Policy in China, 61–87, New York: Cambridge University Press. Knight, J. and L. Song (1999), The Rural-Urban Divide: Economic Disparities and Interactions in China, Oxford: Oxford University Press. Li, B. (2004), “Urban Social Exclusion in Transitional China,” CASEpaper 082, LSESTICERD Centre for the Analysis of Social Exclusion, London. Li, S. and C. Luo (2010), “Reestimating the Income Gap between Urban and Rural Households in China,” in M.K. Whyte, ed., One Country, Two Societies: Rural-Urban

Overview

83

Inequality in Contemporary China, 105–121, Cambridge, MA: Harvard University Press. Li, S. and C. Luo (2011), “Zhongguo shouru chaju yanjing you duoda? Dui xiuzheng yangben jiegou piancha de changshi” (How Unequal Is China? Attempting to Correct the Bias in the Sample), Jingji yanjiu, no. 4, 68–78. Li, S., C. Luo, Z. Wei, and X. Yue (2008), “Appendix: The 1995 and 2002 Household Surveys: Sampling Methods and Data Description,” in B. Gustafsson, S. Li, and T. Sicular, eds., Income Inequality and Public Policy in China, 337–353, New York: Cambridge University Press. Li, S. and S. Yang (2009), “Zhongguo chengshi dibao zhengce dui shouru fenpei he pinkun de yingxiang zuoyong” (The Impact of the Dibao Program on Income Distribution and Poverty in Urban China), Zhongguo renkou kexue, no. 5, 19–27. Lin, L. and C. Wong (2012), “Are Beijing’s Equalization Policies Reaching the Poor? An Analysis of Direct Subsidies under the ‘Three Rurals’ (Sannong),” China Journal, no. 67, 23–45. Ministry of Agriculture (2007), “Guojia jiang baochi zhinong huinong zhengce de wendingxing lianxuxing” (The State Will Maintain Stable and Continuous Policy Support for Agriculture), http://www.china.com.cn/news/2007–09/13/content 8869413.htm. Accessed August 22, 2011. Ministry of Civil Affairs (2007), “2007 nian minzheng shiye fazhan tongji baogao” (Statistical Report on Chinese Civil Affairs Development in 2007), http://cws.mca. gov.cn/article/tjbg/200805/20080500015411.shtml. Accessed August 22, 2011. Minoiu, C. and S. G. Reddy (2008), “Chinese Poverty: Assessing the Impact of Alternative Assumptions,” Review of Income and Wealth, 54(4), 572–596. National Bureau of Statistics (NBS) (2003), Zhongguo tongji nianjian 2003 (Chinese Statistical Yearbook 2003), Beijing: Zhongguo tongji chubanshe. National Bureau of Statistics (NBS) (2008a), Zhongguo tongji nianjian 2008 (Chinese Statistical Yearbook 2008), Beijing: Zhongguo tongji chubanshe. National Bureau of Statistics (NBS) (2008b), Zhongguo tongji zhaiyao 2008 (Chinese Statistical Abstract 2008), Beijing: Zhongguo tongji chubanshe. Osberg, L. (2000), “Poverty in Canada and the United States: Measurement, Trends, and Implications,” Canadian Journal of Economics, 33(4), 847–877. Ravallion, M. (1992), “Poverty Comparisons: A Guide to Concepts and Methods,” Living Standards Measurement Study Working Paper No. 88, The World Bank, Washington, DC. Ravallion, M. and S. Chen (2007), “China’s (Uneven) Progress against Poverty,” Journal of Development Economics, 82(1), 1–42. Ravallion, M., S. Chen, and Y. Wang (2006), “Dibao: A Guaranteed Minimum Income in China’s Cities?” World Bank Policy Research Working Paper No. 3805, The World Bank, Washington, DC. Riskin, C., R. Zhao, and S. Li, eds. (2001), China’s Retreat from Equality: Income Distribution and Economic Transition, Armonk, NY: M.E. Sharpe. Shorrocks, A.F. (1980), “The Class of Additively Decomposable Inequality Measures,” Econometrica, 48(3), 613–625. Shorrocks, A.F. (1982), “Inequality Decomposition by Factor Components,” Econometrica, 50(1), 193–212. Sicular, T., X. Yue, B. Gustafsson, and S. Li (2010), “How Large is China’s RuralUrban Income Gap?” in M. K. Whyte, ed., One Country, Two Societies: Rural-Urban

84

Li Shi, Luo Chuliang, and Terry Sicular

Inequality in Contemporary China, 85–104, Cambridge, MA: Harvard University Press. State Council (2003), “Chengshi shenghuo wuzhuode liulang qitao renyuan jiuzhu guanli banfa” (Administrative Measures on Helping and Managing Poor Urban Vagrants and Beggars), State Council Decree No. 381, http://news.xinhuanet.com/zhengfu/ 2003–06/22/content 931160.htm. Accessed April 13, 2012. Wan, G. (2007), “Understanding Regional Poverty and Inequality Trends in China: Methodological Issues and Empirical Findings,” Review of Income and Wealth, 53(1), 25–34. World Bank (2009a), “From Poor Areas to Poor People: China’s Evolving Poverty Reduction Agenda: An Assessment of Poverty and Inequality in China,” Poverty Reduction and Economic Management Department, East Asia and Pacific Region, Report No. 47349-CN, The World Bank, Washington, DC. World Bank (2009b), The World Development Report 2009: Reshaping Economic Geography, Washington, DC: World Bank Publications. Yao, Y. (2009), “The Political Economy of Government Policies toward Regional Inequality in China,” in Y. Huang and A. M. Bocchi, eds., Reshaping Economic Geography in East Asia, 218–240, Washington, DC: World Bank Publications.

THREE

Housing Ownership, Incomes, and Inequality in China, 2002–2007 Hiroshi Sato, Terry Sicular, and Yue Ximing

I. Introduction An important feature of the post-Mao period has been the resurrection of private property rights. A variety of interrelated policies, including the lifting of prohibitions on private enterprise, ownership reforms in industry, the development of stock markets, and real estate and housing reforms, have paved the way for the expansion of private property and household wealth, with implications for incomes and inequality. Estimates by Li, Luo, and Sicular in Chapter 2 of this volume, calculated using the 2002 and 2007 China Household Income Project (CHIP) surveys, show that the share of household income derived from financial assets and housing and their contribution to income inequality has increased, especially in urban China. In this chapter we examine changes in private ownership of housing and the implications for the distribution of housing wealth and income. We focus on housing wealth rather than on total wealth mainly because the CHIP 2007 data do not contain sufficient information to permit the estimation of total wealth. Housing wealth, however, can provide insights into the role of total wealth, because housing is the single most important household asset in China. Past studies of wealth in China have found that privately owned housing constitutes nearly 60 percent of household wealth and accounts for two-thirds of inequality in wealth among households (Li and Zhao 2008; Zhao and Ding 2008).1 1

These sources use the 2002 CHIP data.

We thank Xu Jing, Zhang Yichuan, and Jerry Lao for their help in carrying out background research and calculations for this chapter. Financial support from the Ontario Research Foundation, the JSPS Grant-in-Aid for Scientific Research (No. 18203018 and No. 21330065), the Research Unit for Statistical and Empirical Analysis in Social Sciences (JSPS Global COE Program), and Hitotsubashi University is gratefully acknowledged.

85

86

Hiroshi Sato, Terry Sicular, and Yue Ximing

Analysis of housing wealth is of interest not only because it influences the distribution of total wealth, but also because it can reveal the distributional implications of China’s urban housing reforms. In the late 1990s and early 2000s China carried out privatization of urban housing. As noted by Yemstov (2008) in his study of housing privatization in Eastern Europe and the former Soviet Union, housing reforms in transition economies are important because of the “sheer size” of the wealth transfer. Several studies have examined the effects of China’s urban housing reforms on urban wealth, incomes, and poverty (e.g., Meng 2007; Sato 2006; Zax 2003). Our work extends these analyses in two regards. First, we include not only urban but also rural and migrant households to understand the broader structure of homeownership and its implications for nationwide patterns of wealth and income. Second, we use the more recent 2007 data that reveal the longer-term consequences of the urban housing reforms. As discussed elsewhere in this volume, measurement of income should include imputed rental income from owner-occupied housing. Indeed, a distinctive feature of the CHIP studies of Chinese income inequality is that, unlike most other studies of incomes and inequality in China, they include imputed rental income. This chapter provides a reexamination and careful calculation of estimates of housing wealth and imputed rental income. Our estimates of imputed rental income are used to construct estimates of household income used elsewhere in this volume. Our view is that in the future close attention to these variables will be needed, because homeownership and personal wealth have become significant, long-term features of the Chinese economy. We begin the chapter with an overview of the policy reforms regarding housing ownership in urban and rural China. We then discuss estimates of housing wealth and imputed rental income for 2002 and 2007.2 The CHIP data contain some but not all of the variables needed to estimate housing rent and imputed rental income, so we must negotiate around the data constraints. In the following sections of this chapter, we highlight key aspects of the data and the estimation methods. Where possible, we have used information in the data sets to cross-check and identify possible biases in our estimates. Using these estimates, we measure inequality in the distribution of housing wealth and of income. We present estimates of inequality in housing wealth for China as a whole and for the urban and rural sectors separately. 2

We examine only owner-occupied housing, as information on other real estate holdings of households is incomplete in the CHIP data.

Housing Ownership, Incomes, and Inequality in China, 2002–2007

87

We also present estimates of inequality in the distribution of imputed rental income from owner-occupied housing and its contribution to overall income inequality. Finally, we analyze the factors associated with housing tenure and with levels of housing wealth. We find significant differences in patterns of housing tenure and wealth between urban and rural areas, as well as changes in these patterns between 2002 and 2007.

II. Institutional and Policy Background of Chinese Housing Reform Housing ownership was decisively different in the urban and rural areas during the Mao era, and the housing-reform policies in the urban and rural areas have developed independently. Although recent new policy experiments have started to bridge the land systems in the rural and urban areas, for the period covered in this chapter it is appropriate to describe them separately. This we do below, with reference to Table 3.1, which summarizes housing-related regulations and policies in China during the post-Mao period.

A. Urban Housing Policy Urban housing reform in the post-Mao era can be divided into three periods: (1) from the late 1970s to 1998: the period of the dual-track policy with coexisting public and private housing; (2) from 1999 to 2004: the period of privatization, and (3) 2005 and thereafter: the period of private housing but with an emphasis on social welfare housing policies (see Chen, Chen, and Liu 2008; Cheng 1999; Jia and Liu 2007; Sato 2006; Wu, Gyourko, and Deng 2010). During the 1980s government announcements about urban housing emphasized two basic policies. One was rent reform (zujin gaige), which involved raising the rent of publicly owned housing (although it is referred to as public housing, in fact housing was owned mainly by urban work units, with some ownership also by local governments), while simultaneously adding housing allowances to salaries. The other was the “commercialization of housing” (zhufang shangpinhua), that is, selling publicly owned housing to urban residents. The first official statement advocating the commercialization of housing was the State Council’s “Agenda of the National Work Conference on Capital Construction” in June 1980. After carrying out some limited experiments in selling publicly owned housing throughout the 1980s, in 1988 the State Council issued an agenda for housing reform that

88

Hiroshi Sato, Terry Sicular, and Yue Ximing Table 3.1. Chronology of housing reform A: Urban

June 1980 The State Council officially refers to “commercialization of housing” (zhufang shangpinhua) for the first time. April 1988 A constitutional amendment gives legal foundation to the transfer of the right-of-use of land. October 1991 The State Council’s “Directive on the Promotion of Urban Housing Reform” refers to privatization of housing, increases in rent of publicly owned housing, and the establishment of a housing construction fund as the main policy arrangements. July 1994 The State Council’s “Decision on Deepening Urban Housing Reform” advocates a transition from in-kind allocation of publicly owned housing to “commercialization” (shangpinhua) and “socialization” (shehuihua) of urban housing in the direction of a “socialist market economy.” As a core policy for the transition, the housing provident fund (zhufang gongjijin) for urban employees is adopted nationally at the end of 1990s. July 1998 The State Council’s “Directive on the Further Deepening of Urban Housing Reform and Accelerating Housing Construction” (Document No. 23 of 1998) announces the official termination of in-kind allocations of publicly owned housing as of the latter half of 1998. August 2003 The State Council’s “Directive to Promote Continuous Development of the Real Estate Market” emphasizes the role of markets in guaranteeing an adequate supply of housing for the urban population. April 2005 The State Council’s “Comments on Policies for the Stabilization of Housing Prices” prohibits “speculative” trade in housing and increases the supply of economically affordable housing, low-rent rental housing, and medium-quality commodity housing. August 2007 The State Council’s “Several Comments on How to Solve Housing Poverty among Low-income Urban Residents” focuses on the development of a “subsidized rental housing” (lianzu fang) program to alleviate housing poverty. October 2007 Hu Jintao refers to promotion of a low-rent housing policy program at the Seventeenth National Congress of the CCP. December 2008 The Central Work Conference on Economic Policy of the CCP emphasizes the critical importance of alleviating housing poverty and developing the real estate market. B: Rural April 1981 The State Council issues an urgent instruction to prohibit the diversion of farmland to housing use. February 1982 The State Council issues the “Regulations on the Administration of Land for Housing in Villages and Rural Townships.” October 1985 The Ministry of Construction and the Environment issues the “Regulations on the Administration of Construction in Villages and Rural Townships.” June 1986 Enactment of the Land Administration Law, based on the principle of “only one house-building plot per rural household.”

Housing Ownership, Incomes, and Inequality in China, 2002–2007

89

May 1997 The CCP Central Committee and State Council circulate an official notice strengthening land management and protection of farmland. May 1999 The State Council issues an instruction to prohibit transactions between rural and urban residents on the right of use of rural land for housing. October 2004 The State Council issues the “Decision on Strengthening Land Management,” emphasizing again the principle of “only a one-house building plot per rural household” and prohibiting the purchase by urban residents/work units of the right of use of rural land for housing. March 2007 Enactment of the Real Rights Law, ensuring that rural households given the right of use of land for housing are allowed to possess the land and to build their own houses on it. January 2008 The Tenth “Document Number One” emphasizes that urban residents are not allowed to purchase the right of use of rural land for housing or to purchase rural residents’ housing. October 2008 The Third Plenum of the Seventeenth CCP Central Committee emphasizes strengthening rural land management to protect the peasants’ right of use of farmland and land for housing. Sources: Chen, Chen, and Liu (2008); Jia and Liu (2007); Luo (2009); Sato (2006); Xu and Kong (2009); the official website of the Central People’s Government of the PRC, at http://www.gov.cn/. Accessed July 28, 2011; and the China Real Estate Law and Regulation Data Base, at http://www. law110.com/law. Accessed July 28, 2011.

stressed rent reform. The purpose of rent reform was to create a foundation for the commercialization of housing by making the maintenance costs of publicly owned housing visible. Despite these announcements, because of the high inflation of the late 1980s it was difficult to implement rent reform. Therefore, rent increases were modest. The 1988 and 1995 CHIP urban data show that rents from publicly owned housing were still quite low. The ratio of annual rent actually paid by renters to annual household food expenditures was only 0.05 in 1988 and 0.07 in 1995, indicating that the share of rent in the urban household budget was very small in both years.3 In accordance with the doctrine of the “socialist market economy” adopted in 1993, in July 1994 the State Council issued the “Decision on 3

Comparison of the unweighted averages for the ten provincial-level administrative units covered in both the 1988 and 1995 CHIP surveys (Beijing, Shanxi, Liaoning, Jiangsu, Henan, Anhui, Hubei, Guangdong, Yunnan, and Gansu). Food expenditures include both in-kind and cash expenditures. Because data on household expenditures are not complete in the 1988 data, we utilize this ratio as an indicator of the weight of rent in the budget of urban households. We use unweighted figures for the comparisons of 1988 and 1995 because we do not have appropriate population weights for the 1988 and 1995 data.

90

Hiroshi Sato, Terry Sicular, and Yue Ximing

Deepening Urban Housing Reform.” This decision called for a transition from the subsidized rental housing by work units to the commercialization and socialization (shehuihua) of housing. Socialization here meant promoting housing-related industries, such as construction, repair, and maintenance. A core policy arrangement for this transition was the housing provident fund (zhufang gongjijin) for urban employees, which was mentioned in the 1994 decision and adopted nationally at the end of the 1990s. The housing provident fund is an employer-subsidized savings program for the purchase of housing. In principle, the program covers not only employees of publicly owned work units but also those who are employed in the nonpublic sector, including foreign-owned enterprises. Although the standard contribution rate for employees has varied over time, by ownership status of the work units, and across regions (between approximately 2 to 10 percent of salary), the general requirement of one-for-one matching contributions by the employer has not changed. The funds are deposited into the employee’s own account in a state-owned commercial bank. Employees own the account but must retain it until they retire or resign from their work units. Those who have housing provident fund accounts also benefit from low-interest bank loans for housing (zhufang gongjijin dixi daikuan) when they purchase housing (Buttimer, Gu, and Yang 2004). During the 1990s, commercialization of housing coexisted with the continuation of the Mao-era system of allocation of subsidized rental housing (fuli fenfang) by work units. The sale of “housing-reform housing” (fanggai fang), that is, the sale of publicly owned housing to tenants (employees) at prices below market value, was the dominant channel for commercialization during this period. Purchase of “commodity housing” (shangpin fang) accounted for a small proportion of home ownership. According to the 1988 and 1995 CHIP urban samples, the proportion of homeowner households to total households increased from approximately 14 percent to 40 percent. The great majority of homeowners in 1988 were owners of previously owned or inherited old private housing, and those households who became homeowners through the housing reform accounted for less than 1 percent of the total households. By 1995, the share of households that owned housing-reform housing in the total had jumped to approximately 27 percent, whereas the share of households that had purchased commodity housing was still very small (approximately 1.3 percent).4 4

Comparison of the unweighted averages for the ten provincial-level administrative units covered in both the 1988 and 1995 CHIP surveys.

Housing Ownership, Incomes, and Inequality in China, 2002–2007

91

750 700 Newly constructed housing 650 Total sales of commodity housing

600 550

Sales of economically affordable housing

Million square meters

500 450 400 350 300 250 200 150 100 50 0 1990

1991

1992

1993

1994

1995

1996

1997

1998

1999

2000

2001

2002

2003

2004

2005

2006

2007

Year

Figure 3.1. Floor Area of Urban Housing, 1990–2007. Source: NBS, Zhongguo tongji nianjian, various years. These numbers represent annual flows.

As shown in Figure 3.1, National Bureau of Statistics (NBS) data show housing construction in urban areas increased rapidly in the late 1990s. The sale of commodity housing, however, remained small relative to new housing construction in terms of total square meters, suggesting that most of the new housing was still constructed by work units and distributed (either sold or rented) to employees. Indeed, the data on sales of commodity housing displayed in Figure 3.1 include housing purchased by work units and distributed to employees. In July 1998, urban housing reform entered a new phase with the State Council’s “Directive on the Further Deepening of Urban Housing Reform and Accelerating Housing Construction” (hereafter referred to as the “1998 Directive”). The 1998 Directive terminated the dual-track system in urban housing. It declared that administrative allocations of rental housing through work units or local governments would be terminated in the latter half of 1998 and that the privatization of housing would be implemented gradually. The privatization of urban housing was to occur throughout urban areas nationwide, but with the time frame differing among provinces. Complementary policy arrangements regarding housing financing (through

92

Hiroshi Sato, Terry Sicular, and Yue Ximing

Table 3.2. Housing tenure for rural, urban, and migrant households, 2002 and 2007 (% of households) 2002

Renters Owners Of which: housing-reform housing commodity housing inherited, self-built, and other Other/missing

2007

Rural

Urban

Migrant

Rural

Urban

Migrant

0.8 98.8

18.2 77.8

58.1 7.2

n.a. n.a.

9.8 88.7

74.5 3.9

60.7 7.4 9.7 0.3

4.0

54.9 27.0 6.8 34.7

n.a.

1.5

21.6

Note: Calculated using data from the CHIP surveys, with weights. Urban refers to households in the CHIP urban subsample, and migrant refers to long-term, stable rural-urban migrant households in the CHIP migrant subsample. For migrant households, these statistics refer to housing tenure in their urban place of residence. For rural households, information on housing tenure is not available for 2007 (see text). “Other/missing” includes collective housing arrangements, such as shared housing and dormitories. n.a. = not applicable.

the housing provident fund and mortgage loans to households) and support of the real estate industry were also implemented. The 1998 Directive set the stage for the expansion of private homeownership in urban China, and in its wake inequality of housing wealth began to emerge. Several features of the urban housing privatization have influenced the distribution of housing wealth. First, as noted earlier, the major channel of privatization was the sale of housing-reform housing. The 2002 urban data show that the proportion of owners of housing-reform housing to total households had increased to approximately 61 percent in 2002 (Table 3.2). These were nonmarket transactions between work units and renteroccupants, and the pricing and property rights arrangements varied considerably across work units. In addition, the quality of housing purchased through this channel was closely related to the work unit’s place in, or relationship to, the bureaucratic hierarchy as well as its economic performance. Naturally, the difference between the purchase price and the market price tended to be larger for employees of powerful work units (Ren and Kang 2003; Sato 2006). Although there were constraints on property rights attached to housing-reform housing according to the pricing method (Sato 2006), such housing became a source of inequality. Second, rapid development of the housing market and a surge in housing prices after privatization exacerbated housing inequality. As shown in Figure 3.1, sales of commodity housing began to increase rapidly from the

Housing Ownership, Incomes, and Inequality in China, 2002–2007

93

10.0

9.0

Commodity housing

8.0

Economically affordable housing

7.0

General CPI (urban household)

Price change from previous year (%)

6.0

5.0

4.0

3.0

2.0

1.0

0.0

-1.0

-2.0

Year

Figure 3.2. Changes in Urban Housing Prices, 1998–2007. Source: NBS, Zhongguo tongji nianjian, various years.

end of the 1990s and, as shown in Figure 3.2, housing prices increased markedly, driven by strong demand and the inflow of speculative money to the immature urban real estate market. As shown in Table 3.2, in 2002 owners of commodity housing reached approximately 7 percent of total households, and by 2007 the share had increased further to 27 percent. Third, during this period the main emphasis was on the marketization of housing, and social welfare-oriented housing policies were relatively weak (Chen et al. 2008).5 The 1998 Directive advocated two types of welfare-oriented housing projects (anju gongcheng): first, “economically affordable housing” (jingji shiyong fang) for sale to low- and lower-middleincome households and second, “subsidized rental housing” for rental to low-income households. In fact, the supply of economically affordable and subsidized rental housing stagnated throughout the 2000s. Although annual sales of commodity housing increased from 340 million square meters in 5

For example, the directive of the State Council on the development of the real estate market in August 2003 emphasized the role of the market in guaranteeing an adequate supply of housing for the urban populace.

94

Hiroshi Sato, Terry Sicular, and Yue Ximing

2004 to 700 million square meters in 2007, the amount of economically affordable housing sold annually remained around 32 to 35 million square meters (Figure 3.1). The fiscal incentives of local governments explain this tendency. In the 2000s the sale of the right of use of urban land became an increasingly important revenue source for local governments, the so-called land-dependent local public budget (tudi caizheng). Local governments therefore welcomed price increases in real estate markets and had an incentive to sell land-use rights at high market prices rather than to allocate it for subsidized housing for low-income families. Fourth, housing reform was aimed specifically at urban households with local urban household registrations (hukou). Residents not having a local urban household registration, especially rural-urban migrants, were systematically excluded from the reform. Many rural-urban migrants have consequently faced high rental costs and lived in substandard housing. Indeed, in some urban suburbs renting rooms to migrants has become a profitable sideline for local rural households. To cope with the rising inequality in access to housing, in 2005 national housing policy began to emphasize affordable urban housing as a social welfare policy. The State Council’s “Comments on Policies for the Stabilization of Housing Prices,” issued in April 2005, states that speculative transactions in housing should be strictly regulated and that housing construction should focus on affordable, medium-quality housing. In August 2007, the State Council issued “Several Comments on How to Solve Housing Poverty among Low-income Urban Residents,” which emphasized the importance of supplying rent-subsidized housing. According to the Ministry of Housing and Urban–Rural Construction, by the end of 2006 the rent-subsidized housing system covered approximately 80 percent of the country (512 of 657 cities).6 The Central Work Conference on Economic Policy held in December 2008 reiterated that increasing the supply of reasonably priced housing for low- and middle-income households was critical to stimulate domestic consumer spending. These policy documents indicate a redirection in urban housing policy, but because of their timing these initiatives may not be captured in the 2007 CHIP data.

B. Rural Housing Reform In strong contrast to the urban case, throughout the Maoist era in most parts of rural China households were allowed to preserve land for housing 6

Official report of the Ministry of Housing and Urban–Rural Development, February 14, 2007 (Zhongguo jianshebao, February 16, 2007).

Housing Ownership, Incomes, and Inequality in China, 2002–2007

95

use and were able to own, build, and inherit housing. Regulations for the administration of the People’s Communes issued in the early 1960s prescribed that rural households have ownership of their housing whereas ownership of land, including land for housing use (zhaijidi), belonged to the collective. In some places, collective farms built new, collectively owned, apartment-style housing for their member households, but this was generally limited to a few “model” communes and brigades. This ownership structure remained in place even after the breakup of the commune system in the early 1980s; since that time, rural housing has continued to be privately owned, built, and inherited by rural households on collective land. Rural housing policy in the reform era has mostly been aimed at controlling the conversion of farmland into housing. Following Xu and Kong (2009), here we distinguish three periods of rural housing policy since 1980. In the first period (1980–1985) efforts were made to reorganize the management systems of land for housing use in light of the institutional changes associated with the expansion of the household contract responsibility system and the subsequent collapse of the commune system. Rapid increases in peasant income in the early 1980s stimulated a boom in housing construction in rural areas and caused the diversion of farmland to land for housing use. This became a policy concern, and the government repeatedly issued orders prohibiting the diversion of farmland to housing use (e.g., the State Council’s urgent instruction of April 1981). To strengthen control over rural housing construction, in February 1982 the State Council issued the “Regulations on the Administration of Land for Housing in Villages and Rural Townships,” followed in October 1985 by a related regulation issued by the Ministry of Construction and the Environment. These regulations required that housing construction in rural areas be reviewed by the villages (the collective owners of the land) and then approved by the township authority. The second stage began in 1986 with enactment of the Land Administration Law. This law established a hierarchical land management system from the national down to the township level. With respect to the management of rural land for housing use, the Land Administration Law allowed each rural household to hold only one house-building plot, the size of which was to be limited to within the provincial standards. Ongoing concern about the preservation of farmland also led to an experimental policy introduced in the latter half of the 1980s – the introduction of a fee for use of rural land as housing. The fee experiment, however, was canceled before being adopted more broadly because it was incompatible with the overall policy of reducing the burden of taxes and fees in the rural areas.

96

Hiroshi Sato, Terry Sicular, and Yue Ximing

The third period, starting from 1997, was benchmarked by several policy documents addressing the rapid development of rural-urban migration and accelerating urbanization. These documents include an official May 1997 notice issued by the Chinese Communist Party (CCP) Central Committee and the State Council on strengthening land management and protecting farmland, an October 2004 decision issued by the State Council on strengthening land management, enactment in March 2007 of the Real Rights Law, and the CCP Central Committee’s decision on rural policies issued at the October 2008 Third Plenum of the Seventeenth Central Committee. This series of policy documents provided measures to address the growing pressure from urban areas to expand suburban housing into rural land and policies to maintain the farmers’ land rights. All these documents repeatedly prohibited the purchase of use rights of rural land for housing by urban residents/work units. Enforcing the prohibition, however, has been difficult, and the problem of commodity housing built on rural land without a formal deed to use the land (xiaochanquan zhufang) has grown. Contemporaneously, rural-urban migration led to the abandonment of rural land in some areas, pointing to the need to coordinate rural and urban housing policies in step with the reforms of the rural household registration (hukou) system. Governments at different levels adopted policy experiments in some rural areas to address migration and urbanization, under the general policy framework of “integrated and balanced urban-rural development” (tongchou chengxiang). Examples of such experiments include: the authorization of mortgages on rural housing land for which households have use rights, the exchange of rural housing–land use rights for urban commodity housing (zhaijidi zhihuan), and the reallocation and redevelopment of housebuilding plots through a land-shareholding system (tudi gufen hezuozhi) at the village level.7 Despite such policy experiments, the rural housing system remained at the stage of “individually built, individually owned, individually used, and individually abandoned” (zijian ziguan ziyong zimie), and rural housing markets were suppressed and underdeveloped. But rural households expanded and improved their housing; indeed, housing area and quality increased greatly, although with regional differences (He and Deng 2009). Moreover, despite government prohibitions, in some areas, especially near the cities, the rental and sale of housing continued (Xin and Zhou 2009; Zhao 2006). 7

See, for example, the cases in Chongqing and Zhejiang described by Chongqing Fuling Municipal Bureau of Land and Resources (2009); Qin and Zhong (2009); Ruo (2009); and Sun and Hua (2009).

Housing Ownership, Incomes, and Inequality in China, 2002–2007

97

Nevertheless, at least during the time period covered here, institutional and policy factors continued to constrain the development of rural housing markets (Qin and Zhong 2009). First, under the Land Administration Law, each rural household was still allowed to hold only one house-building plot, and in principle, the transfer of house-building plots and housing property was limited to transfers within the village. Second, a registration system for rural housing property (fangwu chanquan dengji) had not yet been established. Under these conditions, transfers of rural housing occurred mainly due to expropriations of rural land by local governments, reallocations of house-building plots by village authorities, and private underground transfers of housing property to nonvillagers (Qin and Zhong 2009).

III. Estimation of Housing Wealth and Imputed Rental Income: Methodology and Data Issues Housing wealth is equal to owned equity in housing. Housing wealth H is usually calculated as the difference between the market value of owned housing V and the amount of any debt or mortgage on the property M: H = V − M.

(1)

Calculation of the imputed rent on owner-occupied housing usually takes one of two approaches, the “rate of return” (or “opportunity cost”) approach or the “market rent” approach (see Saunders and Siminski 2005; Short, O’Hara, and Susin 2007; Smeeding and Weinberg 2001). The rate of return approach considers imputed rent to be the income the household would earn if its equity in the dwelling were invested in an equivalent financial investment. In this case, imputed rent is calculated as R = i(V − M) − C − D − I ;

(2)

that is, imputed rental income R equals a rate of return i times the household’s equity in the dwelling, minus the costs of ownership C (maintenance and repairs, property taxes, insurance on the property, and so forth), depreciation D, and interest costs I associated with any mortgage or loans on the property. The market rent approach considers imputed rent to be the net income that would have been earned if the dwelling had been rented out on the rental market. In this case, imputed rental income is calculated as R = R m − C − I,

(3)

98

Hiroshi Sato, Terry Sicular, and Yue Ximing

with imputed rental income R equal to estimated market rent on the dwelling Rm , minus costs of ownership C and interest costs I associated with any mortgage or debt on the property. Typically, depreciation is not subtracted in the market rent approach. In our analysis of the rural sample, we use the rate of return approach to estimate imputed rents on owner-occupied housing. The rate of return approach makes sense for rural households because active rental markets are absent in many parts of rural China. Nevertheless, rural households are able to estimate the value of their housing, either from the costs of construction or, in areas with more active real estate markets, from comparable sales in their locality. For the urban sample we use the market rent approach to calculate a set of “base” estimates; we also calculate alternative estimates using the rate of return approach. We prefer the market rent approach for the urban sample because of the rapid appreciation in housing prices in urban areas between 2002 and 2007. Due to this price appreciation, estimates of imputed rents based on the rate of return approach increased substantially between the two years. In general, during periods of rapid appreciation of housing prices, rents are more stable than housing prices. Such has been the case in China. Consequently, estimates of imputed rents calculated using the market rent approach increased more moderately. For migrants, our preference is to use the same approach as that for the urban sample, but we must work around the data constraints (as discussed later). Application of these formulae requires household-level information on the market value of the dwelling, housing debt, estimated market rent, ownership costs, interest paid on any mortgage or housing debt, and depreciation. Typically, complete data are not available and researchers must adapt their calculations accordingly. Such is the case here. We estimate housing wealth and imputed rents in rural China using information on housing in the CHIP rural, urban, and migrant survey data sets. Housing information in these data sets is self-reported by the respondents.8 The responses of urban and migrant households to questions about market value and market rent of their dwellings are likely to be reasonably accurate, as information about real estate markets, housing prices, and rents is readily available in cities. Where a household respondent was unclear about the market value of the dwelling, the answer could be based on the market price of similar housing in the neighborhood. The development of 8

We note that estimates of housing wealth and imputed rents for other countries also often rely on values reported by household respondents. Where household survey data are not available, researchers rely on information from other sources, such as national accounts or housing surveys (Saunders and Siminski 2005; Short et al. 2007).

Housing Ownership, Incomes, and Inequality in China, 2002–2007

99

housing markets in rural areas has been uneven, so rural households may have less information about market housing values and market rents. For this reason, rural households were only asked to report the market value of their dwellings and were not asked to provide market rents. With respect to housing market values, rural respondents could provide either an estimate of the local market price of the dwelling or an estimate of the costs to newly construct the dwelling (including both labor and materials, and adjusted to reflect the age and condition of the structure). Thus, rural housing values in locations without active housing markets are most likely based on construction costs. The 2002 and 2007 CHIP data sets do not contain all of the variables needed to calculate housing wealth and imputed rents. The Appendix to this chapter contains an in-depth discussion of data issues. Here we limit our discussion to four important issues: (a) identification of homeowners, (b) lack of information on the value of additional owned residential properties, (c) incomplete information on mortgages, and (d) incomplete information on housing costs. In order to calculate imputed rents, we first must identify which households are homeowners and which are renters. The CHIP data sets contain information on housing tenure for the urban and migrant subsamples for 2002 and 2007 and for the rural subsample for 2002, but not for the rural subsample for 2007. In 2002 only 0.8 percent of the households in the rural subsample reported that they did not own their housing (Table 3.2). These nonowners were distributed fairly evenly across the income distribution. More recent data on rural housing tenure in other sources similarly indicate that rural households overwhelmingly have continued to be homeowners. Using the NBS household survey data, He and Deng (2009: 67) report that at year-end 2006 only 0.7 percent of rural households did not own their dwellings. In view of the very low share of rural renters in 2002 and 2006, we assume that in 2007 all rural households were homeowners. The CHIP data contain information on the market value of the dwelling in which the household resides but not on any additional properties owned by the household. We therefore can only estimate the housing wealth associated with the primary dwelling. The value of any additional owned housing is not included in our estimates. Excluding additional properties will cause an underestimation of the level of housing wealth and its inequality. Based on available data, we believe the bias is more significant in 2007 than in 2002 and in the urban than in the rural sample. The CHIP urban data set contains information on whether the households own additional properties. In 2002 only 1.5 percent of urban homeowners owned additional housing; by 2007

100

Hiroshi Sato, Terry Sicular, and Yue Ximing

the share had increased to 7.5 percent. We note also that some rural-urban migrants living in rental housing in the city may have owned housing in their places of origin, but the migrant survey did not collect any information about housing owned in rural areas.9 Our estimates of housing wealth for migrants therefore include only the value of owned housing in their urban place of residence and so may understate migrant housing wealth. In principle, estimation of housing wealth requires data on mortgage debt, as housing wealth is equal to the household’s equity in the house, that is, the market value of the housing minus the outstanding principal on housing debt. Similarly, mortgage interest costs should be subtracted from imputed rental income. Unfortunately, the CHIP data sets contain information on mortgage debt only for 2002, and only for the rural and urban (not the migrant) samples. Past CHIP studies simply used the reported market value of housing as a proxy for housing wealth. In other words, past CHIP studies essentially assumed that households in China have zero housing debt. Also, they did not subtract mortgage interest costs when calculating imputed rents; that is, they assumed that mortgage interest costs were zero. Because of the lack of mortgage data for 2007, here we must follow the same approach in our analysis; however, we use the 2002 data to calculate alternative estimates of housing equity and imputed rents in that year, which we use to identify biases that may arise from these assumptions. In 2002 mortgages were more important for urban than for rural households (Table 3.3). Among urban homeowner households, 9 percent had mortgages. Of the households with mortgages, the mortgage on average was equal to 47 percent of the value of the dwelling. Fewer than 4 percent of rural homeowner households had mortgages, and among these households, on average, the mortgage was equal to 27 percent of the value of the dwelling. In both urban and rural areas, the per capita incomes of households with mortgages were similar to or higher than were those of households without mortgages. Also, households with mortgages owned more expensive housing than did households without mortgages. Thus, housing debt was not associated with low incomes. Using data from the 2002 survey, we calculated estimates of housing equity. Table 3.4 provides comparisons between market values and equity 9

Huang and Yi (2010) report that in 2005 6 percent of urban households, including both formal urban residents and migrants, who lived in owned housing also owned additional homes, and 5 percent of urban households who rented their dwellings owned other homes (this latter group included rural-urban migrants who rented in the city and owned a home in their hometowns).

Housing Ownership, Incomes, and Inequality in China, 2002–2007

101

Table 3.3. Mortgage debt among homeowner households, 2002 Urban (excluding migrants) No mortgage Mortgage Percentage of households Average size of mortgage (yuan) Average market value of dwelling (yuan) Average equity in dwelling (yuan) Average household income (NBS income definition) (yuan)

All

Rural No mortgage Mortgage

All

91.0% 0

9.0% 51,643

100% 4,634

96.2% 0

3.8% 10,055

100% 385

101,950

110,099

102,681

23,114

36,932

23,644

101,950

58,456

98,048

23,114

26,877

23,245

8,516

8,859

8,547

2,772

2,595

2,757

Note: Calculated using data from the CHIP urban and rural samples, with weights; migrant households from the CHIP migrant survey are not included in this table. Only homeowner households are included.

Table 3.4. Comparisons of housing market value and equity per capita, 2002 Mean (yuan)

Gini coefficient

Homeowners

All

Homeowners

All

5,759 5,665 0.984

0.528 0.551 1.044

0.534 0.538 1.007

Rural A. Market value B. Equity C. Ratio (B/A)

5,824 5,730 0.984

Urban (excluding rural-urban migrants) A. Market value B. Equity C. Ratio (B/A)

33,418 31,895 0.954

26,172 24,980 0.954

0.430 0.464 1.079

0.553 0.581 1.051

National (excluding rural-urban migrants) A. Market value B. Equity C. Ratio (B/A)

13,872 13,361 0.963

12,740 12,271 0.963

0.629 0.664 1.056

0.660 0.677 1.026

Note: Calculated using data from the CHIP urban and rural samples, with weights; households from the CHIP migrant sample are not included in the urban or national estimates because mortgage data are not available for the migrant sample.

102

Hiroshi Sato, Terry Sicular, and Yue Ximing

Table 3.5. Alternative estimates of imputed rents and income per capita based on market value versus equity value of owner-occupied housing, 2002 Urban (excluding rural-urban migrants)

Rural Imputed rent per capita (yuan)

Income per capita (yuan)

Gini of income per capita

Imputed rent per capita (yuan)

Income per capita (yuan)

Gini of income per capita

National (excluding rural-urban migrants) Imputed rent per capita (yuan)

Income per capita (yuan)

Gini of income per capita

2,815

A. Calculated using Market Value 0.3648 838 8,717 0.3221 408

4,833

0.4559

181

2,812

B. Calculated using Equity Value 0.3650 800 8,678 0.3223 393

4,818

0.4558

0.984

0.999

1.0005

0.997

1.0000

184

0.955

C. Ratio: B/A 0.996 1.0006

0.963

Note: Calculated from data in the CHIP urban and rural samples, with individual-level weights. For ease of comparison, in this table we use the rate of return approach for both rural and urban households. The rate of return is set equal to the interest rate on long-term (thirty-year) Chinese government bonds in 2002 (3.2028%). Ownership costs (e.g., depreciation) are not subtracted. Urban data in this table do not include migrants from the CHIP migrant sample. Income per capita is CHIP income which equals NBS income plus subsidies on low-cost rental housing in urban areas plus imputed rents (calculated using the rate of return approach for both urban and rural households).

values in that year. Mean equity values are about 4 percent lower than market values, with a larger difference for urban than for rural households. Inequality as measured by the Gini coefficient is higher for housing equity value than for housing market value, but the difference is not large, especially when inequality is measured over all households, not just homeowner households. We acknowledge, then, that using market value as a proxy for equity will lead to some understatement of the inequality of housing wealth, especially in urban China and among homeowners. Nevertheless, the understatement of inequality nationwide appears to be fairly small. We also use the 2002 mortgage data to calculate alternative estimates of imputed rental income based on both the market values and equity values of owned housing. As shown in Table 3.5, imputed rental income per capita is lower when the calculation is based on equity values, but mean per capita incomes (including imputed rents) are very close and income inequality is virtually identical for the two sets of estimates. This is true nationwide as well as for the urban and rural sectors separately. We conclude that we can reasonably use housing values as a proxy for housing equity in our analysis of the impact of imputed rents on incomes and income inequality.

Housing Ownership, Incomes, and Inequality in China, 2002–2007

103

Our fourth data issue is incomplete information about costs of homeownership. In 2002 and 2007 homeowners in China did not pay property taxes or purchase property insurance, so we do not need to consider these costs. Mortgage interest payments, maintenance and repairs, and depreciation, however, were relevant. We carried out several alternative calculations to investigate the sensitivity of our findings to different assumptions regarding the costs of homeownership (see the Appendix to this chapter). For 2002, we estimated mortgage interest payments by applying an interest rate to the reported household mortgage debt. For depreciation and for maintenance and repairs, we followed the literature and multiplied the market value of the dwelling by an appropriate rate of depreciation. We then compared the levels of income and of income inequality with and without subtracting these costs. The results were very similar. We conclude that although, in principle, the costs of homeownership should be subtracted from imputed rents, ignoring them in our analysis does not substantially affect our results. Ultimately, then, we follow past CHIP practice and simply use the market value of housing as a proxy for housing wealth, that is, H = V.

(4)

With respect to imputed rental income, for rural households we calculate the imputed rental income on owner-occupied housing as equal to a rate of return i times the market value of housing: R = iV.

(5)

For urban and migrant households, our base estimates of imputed rental incomes follow the market rent approach and are simply equal to the reported market rents: R = R m.

(6)

We also present some alternative estimates of imputed rents for urban and migrant households calculated using the rate of return approach shown in Equation (5). For the rate of return, we use the annual average interest rate on thirtyyear Chinese government bonds, which was 3.2028 percent in 2002 and 4.3615 percent in 2007. In this regard, we follow common practice in the literature, which typically applies the rate of return on a long-term, safe investment, such as government treasury bonds or municipal bonds, usually in the 4 to 5 percent interest range. Note that for migrants, in 2002 we have data on the market values of owner-occupied housing but no information on market rents, whereas in

104

Hiroshi Sato, Terry Sicular, and Yue Ximing

2007 we have data on market rents but not on market values. Therefore, for migrants we estimate 2002 market rents as the rate of return times the market value of housing, and we estimate 2007 market values of housing as market rents divided by the rate of return, as implied by Equation (5). In fact, these estimates do not affect income and wealth estimates for the majority of migrants, as few migrants are homeowners (see Table 3.2). In the following sections of this chapter, we use our estimates of housing wealth and imputed rents to calculate mean values and inequality, as well as to run regressions. In these calculations and regressions, in order to ensure that our results are representative, we apply weights to the rural, urban, and long-term stable migrant subgroups based on the rural, urban, and long-term stable migrant populations, by province and by region. Population weights are calculated using information in the 2000 census and the 2005 national 1 percent population sample survey. For householdlevel analyses (e.g., of household housing wealth), we use household-level weights; for individual-level analyses (e.g., of per capita housing wealth or per capita income), we use individual weights. Further discussion of the sample weights can be found in Appendix II to this volume. Some calculations and regressions use estimates of household income. Unless noted otherwise, we use the CHIP income definition, which is equal to NBS income plus urban rental subsidies on low-cost rental housing plus imputed rental income on owner-occupied housing.10 (For further discussion of income definitions, see Chapter 1.) Income is calculated using our base estimates of imputed rents unless stated otherwise.

IV. Housing Tenure and Levels of Housing Wealth Table 3.2 shows housing tenure patterns among rural, urban, and migrant households. As discussed earlier, homeownership is nearly universal among rural households. Ownership is also high among nonmigrant urban households, rising from nearly 80 percent in 2002 to nearly 90 percent in 2007. More than half of these urban households obtained their housing through housing reform, but housing obtained through market purchases of commodity housing increased substantially, rising from 8 percent of the urban households in 2002 to 27 percent in 2007. Inherited or self-built housing accounted for a small and declining proportion of urban households. 10

The number of individuals receiving subsidies on rental housing was small, and all were formal urban residents. In 2002 only 13 individuals and in 2007 only 549 individuals received rental-housing subsidies. For these individuals, the average subsidy per capita was 675 yuan in 2002 and 7,502 yuan in 2007.

Housing Ownership, Incomes, and Inequality in China, 2002–2007

105

Table 3.6. Mean housing wealth per capita, 2002 and 2007 (in yuan and as a percentage of income per capita) Homeowner households

Rural Urban w/out migrants Migrants Urban w/ migrants

All households

2002

2007

2002

2007

5,824 (220%) 33,418 (400%) 44,285 (488%) 33,510 (400%)

9,456 (191%) 76,258 (469%) 130,521 (490%) 76,453 (470%)

5,759 (217%) 26,172 (313%) 4,017 (45%) 24,646 (295%)

9,456 (191%) 68,391 (421%) 5,494 (21%) 63,907 (393%)

Note: The market value of housing is used as a proxy for housing wealth (see text).

This category is largely made up of households that regained ownership of properties that had historically belonged to their families before the nationalization of housing during the Maoist era. Later in this chapter we present the results of a multinomial logit analysis that identifies factors associated with urban housing tenure, and we discuss in more detail the pattern of housing tenure among nonmigrant urban households. Homeownership in the city of residence is rare among migrant households. Even for long-term, stable migrant households, the only category of migrants included in our analysis, fewer than 10 percent owned dwellings in the city where they lived, and the share of homeowners actually declined between 2002 and 2007. The majority of migrants lives in rented housing in cities; a substantial minority lives in collective housing, which includes housing shared with other migrants and dormitories provided by employers. From 2002 to 2007 the importance of collective housing declined somewhat, whereas that of rented housing increased. Levels of housing wealth in China appear to be fairly high (Table 3.6). Not surprisingly, housing wealth is substantially higher for urban and migrant than for rural households, both in absolute terms and relative to their (higher) incomes. For formal urban residents and migrants, the price-toincome ratio is about four to five, which is relatively high by international standards. For rural households, the price-to-income ratio is substantially lower, at about two. Housing wealth for all groups increased rapidly between 2002 and 2007 (Table 3.7). In per capita terms, rural housing wealth grew about 7 percent

106

Hiroshi Sato, Terry Sicular, and Yue Ximing Table 3.7. Average annual increases in per capita housing wealth, 2002 to 2007 (percentage, constant prices)

Rural Urban w/out migrants Migrants Urban w/ migrants

Homeowner households

All households

6.9 15.2 21.3 15.2

7.1 18.4 4.0 18.2

Note: Weighted. Calculated using constant 2002 prices; urban and migrant values are deflated using the NBS urban consumer price index and rural values are deflated using the NBS rural consumer price index.

annually, a growth rate similar to that of rural per capita incomes. Urban and migrant housing wealth grew 15 to 21 percent annually for homeowners, outpacing income growth. This growth reflected in part the rapid increases in urban housing prices (Figure 3.2) and in part the expansion of homeownership among formal urban residents (Table 3.2). The growth was also likely due to improvements in housing quality. Faster growth in urban housing values than in rural housing values led to a widening gap in housing wealth between urban and rural areas (Table 3.8). In 2002 per capita housing wealth for formal urban residents was 4.5 times that for rural residents. By 2007, this ratio had increased to 7.2. These urban-rural gaps in housing wealth exceed China’s high urban-rural gap in per capita incomes.

V. Inequality of Housing Wealth Table 3.9 shows the inequality of housing wealth nationally and for the urban and rural areas separately, as measured by the Gini coefficient. Among homeowners (excluding nonowners), inequality of housing wealth Table 3.8. Ratios of per capita housing wealth between urban, rural, and migrant households, 2002 and 2007

Urban/rural Urban/migrant Migrant/rural

2002

2007

4.5 6.5 0.7

7.2 12.5 0.6

Note: Calculated over all households, including nonowners; weighted. Urban does not include rural-urban migrants.

Housing Ownership, Incomes, and Inequality in China, 2002–2007

107

Table 3.9. Inequality of housing wealth, 2002 and 2007 (Gini coefficients) 2002

Homeowners, per household All, per household All, per capita

2007

All

Urban, w/out Rural migrants

0.59

0.51

0.42

0.55

0.63

0.56

0.45

0.52

0.63 0.67

0.52 0.53

0.55 0.55

0.58 0.58

0.67 0.69

0.56 0.55

0.52 0.52

0.56 0.56

Urban, w/ migrants

All

Urban, w/out Rural migrants

Urban, w/ migrants

Note: Calculated with weights.

per household in China is relatively high, at about 0.60. This compares to Gini coefficients of housing wealth for homeowners of about 0.40 to 0.45 in the Organisation for Economic Co-operation and Development (OECD) countries as well as in Russia and Serbia (Sierminska and Garner 2005; Yemstov 2008). Including nonowners, the Gini coefficient for housing wealth increased from 0.63 in 2002 to 0.67 in 2007. The relatively small difference between the Gini for home-owning households and the Gini for all households reflects the high level of homeownership in China. In this regard, China differs from many other countries. In the OECD countries as well as in Russia and Serbia, the rate of homeownership is lower, so that including nonowners increases the Gini coefficient substantially to between 0.6 and 0.8 (Yemstov 2008). Including nonowners, the inequality of housing wealth in China is no higher than that in these other countries. Inequality of housing wealth in per capita terms is higher than in perhousehold terms, reflecting the larger size of rural households. Urbanrural differences in per capita housing wealth contribute substantially to national inequality in housing wealth per capita. Using standard inequality decomposition methods, we find that in 2007 the urban-rural gap in per capita housing wealth contributed roughly 40 to 50 percent of national inequality in per capita housing wealth, up by about 10 percentage points from 2002.11 Nationwide, inequality of housing wealth both per household and per capita increased between 2002 and 2007. The increases in inequality nationwide reflect widening differences between urban and rural housing wealth 11

We calculate the contribution of the urban-rural gap to national inequality of per capita housing wealth using inequality decomposition by group of the Theil (GE 1) and mean logarithmic deviation (GE 0) inequality indices.

108

Hiroshi Sato, Terry Sicular, and Yue Ximing

Table 3.10. Distribution of housing wealth across income quintiles, 2002 and 2007 2002

2007

Per capita income quintiles

Housing value per capita (yuan)

% of nonowners

Housing value per capita (yuan)

% of nonowners

1 (lowest) 2 3 4 5 (highest) Ratio of top to bottom quintile

2,727 4,480 7,120 14,126 34,178 12.5

2.3 4.3 9.5 17.3 17.7

4,716 8,306 17,984 38,898 96,364 20.4

0.7 2.6 6.6 13.3 12.7

Note: Calculated with weights; in current prices. Includes rural, urban, and long-term, stable rural-urban migrant households.

and increases in rural inequality of housing wealth. In urban China, inequality of housing wealth declined.

VI. Income Inequality and Housing As shown in Table 3.10, households with higher income per capita have more housing wealth per capita. In 2002 households in the top quintile of the income distribution held housing wealth per capita that was, on average, 13 times that of households in the bottom quintile. By 2007, this ratio had risen to 20. The widening gap in housing wealth between low- and high-income households reflects in large part the widening gap between urban and rural housing values. Within sectors, inequality in housing wealth between poor and rich households remained relatively constant between 2002 and 2007. Because the pattern of urban housing wealth has resulted from China’s urban housing privatization and the related real estate market reforms, one can conclude that China’s housing and real estate market reforms have had a dis-equalizing effect. Within urban areas, higher-income households are more likely to be homeowners, and on average, higher-income households own more valuable housing. In addition, urban residents who have worked for profit-making work units or work units with higher bureaucratic status have been more likely to enjoy both higher housing wealth and higher incomes. Such households have been able to purchase highquality housing in good locations, often at heavily subsidized prices (Sato 2006; Tomba 2004). Nationwide, in the wake of the housing reforms higherincome urban households obtained increasingly valuable urban real estate

Housing Ownership, Incomes, and Inequality in China, 2002–2007

109

Table 3.11. Estimates of per capita imputed rental income from owner-occupied housing, 2002 and 2007 (in yuan and as a percentage of income per capita) Base estimates 2002

2007

Alternate estimates 2002

Homeowner households 187 412 187 (7.0%) (8.4%) (7.0%) Urban, incl. migrants 709 2,046 1,073 (7.6%) (11.8%) (11.8%) National average 340 1,055 447 (7.2%) (9.7%) (8.5%) Rural

Rural Urban, incl. migrants National average

All households 184 412 (7.0%) (8.4%) 522 1710 (5.6%) (9.9%) 305 979 (6.5%) (9.0%)

184 (7.0%) 789 (8.7%) 401 (7.6%)

2007 412 (8.4%) 3,334 (17.9%) 1,562 (12.1%) 412 (8.4%) 2,787 (14.9%) 1450 (11.2%)

Note: Weighted; in current prices. Base estimates of imputed rents are calculated using the rate of return approach for rural households and the market rent approach for urban households; alternate estimates are calculated using the rate of return approach for both urban and rural households. See Section III of the text for discussion of the estimates for migrant households.

assets. Lower-income rural households were already homeowners, but their housing was of lower value and did not appreciate as rapidly as urban housing. Table 3.11 shows estimates of per capita imputed rental income from owner-occupied housing. As discussed earlier, these estimates do not deduct the costs of ownership and mortgage interest payments and thereby overstate the level of imputed rental income, but probably do not bias measured income inequality. The level of imputed rental income per capita and its share in household per capita income have increased over time in both rural and urban areas, but especially in urban areas. Our base estimates of imputed rents constituted on average 6.5 percent of household per capita income for all households nationwide in 2002, rising to 9.0 percent in 2007. Our alternative estimates calculated using the rate of return for both urban and rural households show imputed rents at 7.6 percent of household income per capita in 2002, rising to 11.2 percent in 2007.

110

Hiroshi Sato, Terry Sicular, and Yue Ximing Table 3.12. Imputed rents and income inequality, 2002 and 2007 2002 Own inequality (Gini)

2007

Contribution to income inequality (%)

Own inequality (Gini)

Contribution to income inequality (%)

Base estimates Income per capita Income per capita excluding imputed rents Imputed rents per capita

0.450 0.451

100.0 93.5

0.476 0.474

100.0 89.3

0.651

6.5

0.673

10.7

Alternate estimates Income per capita Income per capita excluding imputed rents Imputed rents per capita

0.454 0.451

100.0 90.7

0.485 0.474

100.0 83.3

0.668

9.3

0.689

16.7

Note: Weighted; includes rural, urban, and long-term migrant households. Contributions to income inequality are calculated using income decomposition of the Gini coefficient by source of income. Note that income per capita excluding imputed rents equals NBS income per capita plus estimated subsidies on below-market rental housing. Base estimates calculate imputed rents using the rate of return approach for rural households and the market rent approach for urban households; alternate estimates use the rate of return approach for both rural and urban households. See Section III for a discussion of estimates for migrants.

Imputed rents were distributed more unequally than other income, as shown by their relatively high Gini coefficient (Table 3.12). Decomposition of income by source reveals that the contribution of imputed rental income to overall income inequality has been increasing: In 2002 imputed rents contributed 6.5 percent and in 2007 10.7 percent of inequality in per capita incomes. Our alternative estimates show the contribution increasing from 9.3 percent to 16.7 percent of inequality. Although these contributions to inequality are not exceedingly high, the upward trend is noteworthy, and by 2007 the contribution of imputed rents to national inequality was substantial.

VII. Determinants of Housing Tenure and Housing Wealth In this section we examine the factors that influence housing tenure in urban areas and the determinants of housing wealth in urban and rural areas. Our focus is on changes in the impacts of institutional factors and individual/ family characteristics between 2002 and 2007. In view of the regional differences in the adoption of urban housing reforms and in order to insure

Housing Ownership, Incomes, and Inequality in China, 2002–2007

111

comparability over time, in our analysis we utilize sample households in the cities that are covered in both the 2002 and 2007 CHIP urban surveys (forty cities in twelve provinces). Similarly, for the rural analysis we utilize sample households in the fifteen provinces that are covered by both the 2002 and 2007 CHIP rural surveys.12 The analysis does not incorporate rural-urban migrant households from the CHIP migrant survey, so the findings for the urban households only reflect the situation for formal urban residents. In the economics literature, household housing choices reflect both consumption and investment demand.13 Households consume housing, and their consumption of housing will reflect factors such as prices, income, and family size. In principle, consumption demand can be satisfied by either renting or owning, although the two are not perfect substitutes. Housing as an investment involves ownership. Households invest in housing as a form of wealth, and housing is often the largest component of households’ wealth portfolios. The demand for housing as a form of wealth is influenced by factors that affect wealth accumulation more generally, such as the stage in the life cycle, risk, risk preferences (which may be a function of education), inheritances, and the ability to borrow. Some authors point out that in developing countries, special considerations may arise due to the presence of multigenerational families and the need for precautionary savings (Burger et al. 2008; Deaton 1990). Until the end of the 1970s, housing consumption in urban areas was met through administrative allocations. In rural areas, although households had property rights to their housing, housing consumption was suppressed by low incomes and by egalitarian social and political pressures under the commune system. The role of household demand in housing allocations began to surface with the 1980s’ market reforms, and especially with the reforms in housing ownership and real estate markets in urban areas in the mid-1990s. With these reforms, the standard sorts of variables related to consumption and investment demands for housing began to influence housing tenure choice and housing wealth. At the same time, institutional factors such as the household registration system, ownership of the work unit, and the sociopolitical hierarchy, which influenced the distribution of 12

13

Provinces (provincial-level administrative units) included in the analysis in this section are Beijing, Shanxi, Liaoning, Jiangsu, Anhui, Henan, Hubei, Guangdong, Chongqing, Sichuan, Yunnan, and Gansu for the urban areas, and Beijing, Hebei, Shanxi, Liaoning, Jiangsu, Zhejiang, Anhui, Henan, Hubei, Hunan, Guangdong, Chongqing, Sichuan, Yunnan, and Gansu for the rural areas. The discussion here draws from Arrondel and Lefebvre (2001); Cagetti and De Nardi (2008); Campbell (2006); Davies and Shorrocks (2000); Ioannides and Rosenthal (1994); and Quadrini and R´ıos-Rull (1997).

112

Hiroshi Sato, Terry Sicular, and Yue Ximing

housing during the process of urban housing privatization, continued to shape the observed patterns of housing. Our working hypothesis is that between 2002 and 2007, the influence of institutional factors on housing tenure and wealth persisted, but the impact of individual and family characteristics associated with household consumption and investment demands, such as age, education, and income, became more important.

A. Housing Tenure Choice of Urban Households Our analysis of housing tenure for urban households distinguishes three categories: renters (households that do not own housing), owners of housingreform housing (obtained through the housing reforms), and owners of commodity housing (purchased on the market).14 Using these three housing-ownership types as the categorical dependent variables, we conduct a multinominal logit estimation to analyze the factors that affected housing tenure choice in 2002 and 2007. Our explanatory variables include variables related to household consumption and investment demands, as well as institutional factors relevant to China’s urban housing system. It should be noted that we treat the head of the household as the renter/owner and utilize the household head’s attributes for certain variables in the regression equation.15 Descriptive statistics of the key variables used in these regressions appear in Table 3.13. Age and age squared of the household head are included as indicators of the stage in the family life cycle, and as a measure of seniority that likely affected the administrative allocation of publicly owned housing in the Mao era as well as housing obtained during the privatization of housing (Sato 2006). To allow for the possibility that young families that were formed after the housing reforms may have moved into housing with their parents who obtained their housing during the housing reforms, we include a dummy for young household heads (under thirty years old) who live with a parent who has a local urban hukou. This variable captures another aspect of the household life cycle. The proportion of these households is small but increasing (Table 3.13). 14

15

A fourth category identified in the CHIP urban data set is self-built/inherited older housing. We exclude this group partly because it is largely the result of historical legacy rather than active choice. Because of the small number of households in this group, including it as a category in the analysis will cause the multinominal logit estimation not to converge. We can identify which family member is the housing owner in 2002 but no such information is available in 2007. The 2002 CHIP data show that approximately 80 percent of homeowners were heads of household.

Housing Ownership, Incomes, and Inequality in China, 2002–2007

113

Table 3.13. Characteristics of urban households used in the analysis of urban housing tenure choice, 2002 and 2007 2002 Mean

SD

2007 Min

Dependent variables: Housing tenure categories Renter 0.198 0.399 0 Housing-reform 0.699 0.459 0 housing owner Commodity 0.103 0.304 0 housing owner Characteristics of household head Age 48.430 10.838 19 Local urban hukou 0.985 0.122 0 Employed in 0.549 0.498 0 state-owned or urban collective work units Employed in 0.098 0.298 0 nonpublicly owned work units Self-employment/ 0.047 0.213 0 private business owner Retired 0.239 0.426 0 Others 0.067 0.249 0 Characteristics of household Young household 0.004 head (age 50%) Agricultural income 0.479 Local wage income 0.173 Nonagricultural 0.072 self-employment Migrant wage income 0.123 Multiple income 0.152 sources Provinces (number)

2007 Max

Mean

200 360,000 38,428

10.269 0.472

16 0

88 1

0.499 0.385

0 0

1 1

0.494 0.154

0.010

0

1

1.202 0.353

1 0

2.308

SD

Min

Max

67,062 150 2,500,000

17 0

99 1

0.500 0.361

0 0

1 1

0.013

0.114

0

1

11 1

3.998 0.129

1.368 0.335

1 0

18 1

0

34.865

4.470

3.663

0

74.729

0.249

0

15.103

0.152

1.004

0

75.100

0.104

0

1

0.026

0.158

0

1

0.500 0.379 0.259

0 0 0

1 1 1

0.419 0.168 0.067

0.493 0.374 0.251

0 0 0

1 1 1

0.329 0.360

0 0

1 1

0.161 0.185

0.368 0.388

0 0

1 1

15

48.615 10.209 0.340 0.474

15

Note: Weighted. The sample size used in the 2002 analysis is 6,076, and in the 2007 analysis, 12,176. Income in this table and in the regression analyses is based on the NBS definition.

126

Hiroshi Sato, Terry Sicular, and Yue Ximing

In view of the importance of consumption demand for housing in the rural sector, the family life-cycle stage and family structure may be correlated with the housing value. Unless offset by multigenerational interdependence, we expect an inverted-U-shaped curve for the relationship between the age of the household head and housing wealth. We expect a positive relationship between housing wealth and the variables that measure family size.19 As measures of financial ability and risk preferences, we introduce the educational attainment of the household head, household nonasset income (current per capita disposable nonasset income, based on the NBS definition), and current per capita asset income (a proxy for nonhousing family wealth). As in the case of the urban households, we expect a positive relationship between educational attainment and housing wealth. Because rural China does not have either an official housing financing system or commercial housing loans, we expect positive and significant effects of both asset and nonasset income on housing wealth. Traditional attitudes that regard housing as an important indicator of socioeconomic status may reinforce the relationship between income and housing wealth. With respect to borrowing constraints, available information on social assistance is not consistent between the 2002 and 2007 data sets. The 2007 rural data identify households that receive assistance from the five-guarantee (wubao) program and the rural minimum living standard guarantee (dibao) program, but they do not contain information on the amount of transfer income received from these programs. The 2002 data include information on the amount of transfer income from social welfare and relief programs, but do not identify wubao and dibao households. To address this inconsistency, for 2007 we employ a dummy variable for wubao or dibao households, and for 2002 we employ a dummy variable for households that receive any relief funds (jiuji kuan), subsidies from the collective welfare fund (jiti gongyijin), subsidies for the elderly (laonianren butie), or other public transfers from the state or collective. We expect that these indicators are associated with borrowing constraints and thus will negatively correlate with housing value. To capture the roles of entrepreneurship and out-migration, we include dummy variables for the main source of household income (exceeding 50 percent of the total household income), classified as follows: agricultural 19

For rural families, housing is an important traditional item to prepare for their sons’ marriages. A village survey in northern Zhejiang illustrates that the housing construction cycle in the village was attuned to the villagers’ age structure, and a housing construction boom occurred in the early 1970s when baby boomers born in the 1950s reached marriageable age (Zhang 1998).

Housing Ownership, Incomes, and Inequality in China, 2002–2007

127

income (including income from animal husbandry, forestry, and fishery), local wage income (earned within the township), revenue from nonagricultural self-employment/family business, wage income from migrant jobs (earned outside the township), and multiple sources of income (no single income source exceeding 50 percent of the total household income). Households that depend mainly on agricultural income are used as the reference group. We expect a positive relationship between entrepreneurship and housing wealth, because operating a family business can be associated with risk preferences and because rural housing often serves both residential and productive purposes. With respect to migration and housing, de Brauw and Giles (2008), using panel data of eighty-eight villages in eight provinces from 1986 to 2002, find a causal relationship between out-migration and the building of new housing. Although we are unable to capture causality because we have cross-sectional data, we expect households with income mainly from migration to have more housing wealth. Table 3.18 reports the estimation results. As expected, educational attainment was significantly and positively associated with housing wealth. Notably, the coefficients of educational attainment increased between 2002 and 2007. This finding is consistent with the urban case and suggests an increasing role of education in household risk-related decision making. Household income was also positively and significantly related to housing wealth. The coefficient of asset income was positive and significant in 2007, but not significant in 2002. This result may reflect the fact that in 2002 rural households held very little nonhousing wealth, but since then, the importance of nonhousing wealth has increased (see Table 3.17). Consistent with consumption demand, family structure and the family life-cycle stage had large and significant coefficients. The relationship between the age of the household head and housing wealth followed an inverted-U-shaped curve, with a peak slightly lower than forty-nine years old in 2002 and forty-seven years old in 2007. Family size had a positive coefficient. Contrary to our expectations, after controlling for family size and other factors the coefficient on the dummy variable for a three-generation family was not significant. The dummy variable for social assistance was negative in both years but significant only in 2007. These results imply that borrowing constraints became increasingly important for rural households between 2002 and 2007. It may also, however, reflect differences in the construction of the two variables, in which case they may be an indication that wubao households and dibao households have less housing wealth due to credit constraints,

Table 3.18. Determinants of housing wealth in the rural areas, 2002 and 2007

Characteristics of household head Age Age-squared Junior middle-school education Senior middle-school/vocational middle-school education College education or above Characteristics of household Number of household members “Three-generation family” Main income source Local wage income Revenue from nonagricultural self-employment/family business Wage income from out-migration Multiple income sources Current household nonasset income (per capita, 1,000 yuan) Current household asset income (per capita, 1,000 yuan) Receiving social assistance Province dummies Constant Number of observations Adjusted R-squared

(1) 2002

(2) 2007

0.035*** (0.008) −0.0004*** (0.00008) 0.127*** (0.027) 0.142*** (0.035) 0.117 (0.115)

0.034*** (0.006) −0.0004*** (0.00006) 0.138*** (0.020) 0.184*** (0.028) 0.417*** (0.077)

0.161*** (0.011) 0.024 (0.036)

0.127*** (0.007) 0.009 (0.028)

0.271*** (0.034) 0.113** (0.047) 0.014 (0.037) 0.102*** (0.034) 0.110*** (0.006) 0.105** (0.046) −0.006 (0.108) Yes 8.333*** (0.221) 6,076 0.271

0.362*** (0.027) 0.387*** (0.037) 0.119*** (0.026) 0.186*** (0.024) 0.050*** (0.003) 0.091*** (0.009) −0.429*** (0.055) Yes 9.159*** (0.171) 12,176 0.273

Notes: These ordinary least squares (OLS) regressions are carried out over the sample of households summarized in Table 3.17. The dependent variable is the log of the current value of owneroccupied housing (yuan). The omitted categories are “primary school or below” for education and “agriculture” for the main source of income. The main source of income is defined as the income source that exceeds 50 percent of the total household income by the NBS definition. Standard errors in parentheses. *** denotes statistically significant at the 1 percent level, ** at the 5 percent level, and * at the 10 percent level.

Housing Ownership, Incomes, and Inequality in China, 2002–2007

129

but this does not affect other households receiving subsidies (which are included in the 2002 dummy variable). Finally, we find that the income structure of the household significantly correlates with the housing value. Compared to households with agriculture as the main source of income, households that engage in nonagricultural activities have more housing wealth in both years. Moreover, the coefficients of the income structure dummies increased between 2002 and 2007, suggesting a growing influence of nonagricultural activities on household income risks and on attitudes toward housing consumption. Notably, the coefficient on the dummy for out-migration income was positive but insignificant in 2002 and became larger and significant in 2007. This finding is consistent with the causal influence of outmigration on housing consumption, as reported by de Brauw and Giles (2008).

VIII. Concluding Comments In this chapter we discuss the estimation of housing wealth and imputed rental income from owner-occupied housing using the CHIP data, and we examine the distribution of household housing wealth and the implications for income inequality. Because of incomplete information, we must rely on estimates of housing wealth and imputed rents that are based on the market value of, rather than the household equity in, housing. Nevertheless, sensitivity analyses using more complete information from the 2002 CHIP survey and checks against the published NBS data indicate that our estimates are informative. Our analysis reveals that the distribution of housing wealth in China has some unusual characteristics. When measured among homeowners, inequality of housing wealth is relatively high by international standards. When measured among all households including nonowners, however, inequality in the distribution of housing wealth is not high by international standards. This difference reflects the high level of homeownership in China. Indeed, with a rate of homeownership at or above 80 percent, China has one of the highest rates of homeownership in the world.20 Although property rights associated with homeownership in China may be weaker than elsewhere, this high rate of homeownership potentially has important 20

A study using 2001–2010 data from twenty-six developed countries finds only three countries with homeownership rates above 80 percent: Singapore, Spain, and Iceland (Pollock 2010).

130

Hiroshi Sato, Terry Sicular, and Yue Ximing

implications, both economically and politically. Homeownership affects microeconomic behavior. It also influences the distributional impact of economic policies and macroeconomic fluctuations. We find that the inequality of housing wealth nationwide among all households, including migrants and nonowners, increased between 2002 and 2007. This increase in inequality reflected increased inequality within the rural areas and the widening gap in housing wealth between the urban and rural areas. Within the urban areas, inequality of housing wealth declined due to a rise in the rate of homeownership. Similarly, imputed rental income was unequally distributed and its contribution to the inequality of household per capita income showed a marked increase between 2002 and 2007. These trends reflect the fact that China’s recent urban housing and real estate market reforms have disproportionately benefited higherincome residents, and that, as in the case of income, urban-rural differences are a key feature of the inequality in housing wealth. Using multinomial logit and regression analyses, we examine the factors associated with homeownership and housing wealth for urban (nonmigrant) and rural households. As expected, in urban areas institutional factors such as hukou and type of employer affect the likelihood of owning housingreform housing versus having other forms of housing as well as the value of owned housing. This reflects the legacy of the urban housing reforms. We also find that variables commonly associated with consumption demand for housing – income and family size – are significant. Some variables associated with investment demand for housing are also significant. Borrowing constraints, captured by the proxy for social welfare assistance, are negatively correlated with housing wealth, and the variables associated with risk preferences, such as education and entrepreneurship, are positively correlated with housing wealth. Life-cycle effects do not follow the usual pattern of increasing housing wealth through middle age and then declining in old age. Among urban homeowners, we find little relationship between housing wealth and age; among rural homeowners, housing wealth, on balance, declines with age. Because we employ cross-sectional data, these results may reflect differences across cohorts because of the relatively recent housing privatization and also because housing choices and investments have taken place in a rapidly changing institutional and economic environment. It remains unclear how patterns of housing wealth will play out in the future. Without policy interventions, inequality related to housing is likely to increase because of the strong urban-rural division and because younger and migrant households may be unable to afford to buy into urban housing. Recent measures to expand the supply of low-cost urban housing may help

Housing Ownership, Incomes, and Inequality in China, 2002–2007

131

these groups, but they do not address the underlying distortions in land management and real estate markets. Moreover, beneficiaries of low-cost urban housing are basically residents with local urban hukou. Only very recently have rural migrants been entitled to apply for low-cost urban housing, and only in a few cities.21 As housing is a form of investment as well as of consumption, the distribution of housing is also affected by China’s underdeveloped financial system and the lack of investment vehicles available to households. Regardless, ownership of housing will remain an important factor in personal welfare and inequality in China. We therefore recommend that future surveys place a priority on collecting good-quality housing-related statistics and that future studies of inequality pay close attention to the role of housing wealth. APPENDIX: ADDITIONAL DISCUSSION OF HOUSING DATA IN THE CHIP 2002 AND 2007 SURVEYS

As discussed in the text, calculation of housing equity and imputed rental income on owner-occupied housing requires information on ownership status, the market value of housing, mortgage amounts, and the costs of ownership. Table 3A.1 shows the relevant variables that are present in the 2002 and 2007 CHIP data sets. Because different information is available for the rural, urban, and migrant subsamples, the table shows each separately. In the table, “CHIP” refers to variables collected through interviews of households using the independent CHIP questionnaires. “NBS” refers to variables collected by the NBS in its annual household surveys that were provided to the CHIP and are available in the CHIP data sets. A few variables are available from both sources. All these variables are self-reported by the households. Because some relevant variables are not available for all subsamples in all years, estimates of housing wealth and imputed rental income presented in the body of this chapter are based on several simplifying assumptions (see Section III). For some subsamples in some years, however, most or all of the relevant variables are available, and thus, we can calculate alternate estimates that are based on fuller information. In the main text and tables we have reported some comparisons of alternate estimates. In this 21

Inclusive housing welfare policy open to rural migrants was one of the issues debated in the lead-up to the National People’s Congress in March 2012. See for example, a discussion meeting sponsored by People’s Daily Online, New China News Agency Online, and Guangming Online on March 9, 2012. Guangming Online, http://big5.gmw.cn/g2b/ politics.gmw.cn/2012-03/09/content 3737377.htm. Accessed March 10, 2012.

132

Hiroshi Sato, Terry Sicular, and Yue Ximing Table 3A.1. Relevant housing variables in the 2002 and 2007 CHIP data sets Rural

Urban

Migrant

2002 2007 2002 2007 2002 2007 (CHIP) (NBS) (CHIP & NBS) (NBS) (CHIP) (CHIP)

Variable Ownership status of the dwelling Market rent

87

b24(NBS)

b24 b210

Market value

704

b28

209

Outstanding mortgage Maintenance costs Interest payments on mortgage Depreciation Year when house was built or bought

708a 610b

503(CHIP); b210(NBS)a 503a(CHIP); b28(NBS)b 417(CHIP)

b211(NBS)

b211

410

x134

401

i114 i119

Notes: The table gives the question number/code for the variable in the questionnaires and the data set. “CHIP” refers to data collected using the independent CHIP survey questionnaires; “NBS” refers to data provided to the CHIP by the NBS from its household survey. All variables are self-reported by the households. a The CHIP and NBS data give very different market rents, on average. b The CHIP and NBS data give very similar market values, on average.

Appendix we discuss some issues regarding the mortgage data and urban rental values in the CHIP data sets. We also present some additional estimates of imputed rental income that incorporate the costs of ownership under different assumptions in order to gauge the sensitivity of our results to treatment of the costs of ownership.

A. Mortgage Data and Treatment of Negative Equity Mortgage data are available only for 2002. We carried out a variety of checks on the 2002 mortgage data; for example, we compared the size of the mortgage to the market value of the housing, checked whether households with mortgages have any particular characteristics, and so on. Based on these checks and examination of the data, we conclude that the 2002 housing data appear to be of good quality. One issue with the 2002 mortgage data is that a small number of homeowner households reported mortgage debt that exceeded the market value

Housing Ownership, Incomes, and Inequality in China, 2002–2007

133

of their housing, implying negative equity (less than 1 percent of the households in the rural sample and less than 2 percent of the households in the urban sample). We checked the market value per square meter of housing for these households and found it reasonable and similar to that for the households with positive equity. This suggests that the negative equity was not due to unusually low market values of housing, but instead to unusually high mortgage levels. Our view is that it is unlikely that households in fact had negative housing equity. Chinese households do not have easy access to credit, and so they typically pay substantial portions of the purchase price in cash. Moreover, negative equity is usually associated with falling housing prices, which did not occur in 2002. It is possible that the housing debt reported by households includes borrowing for purposes other than the purchase of their dwellings. Data errors may also be present. In view of these considerations, in analyses that use mortgage data from the 2002 CHIP survey, we assume that the true mortgage debt does not exceed the market value of the housing, that is, we set a minimum equity value of zero.

B. Inconsistent 2002 NBS and CHIP Data on Urban Rental Values of Housing For the 2002 urban sample, we have two sets of data on market values and rents for owner-occupied housing, one from responses to the independent CHIP questionnaire and the other provided by the NBS from its household surveys. The information on the market values of housing from these two sources is fairly consistent, but the information on rental values is not. Information from the two sources on housing market values (unweighted) is summarized in Table 3A.2. Note that this table includes information only for those urban households that own their dwellings and report a nonzero market value of housing. The lower panel of Table 3A.2 shows information for homeowner households that report nonzero housing value in both data sources, so the statistics are calculated over the same subsample of households.22 For this common sample, the average market value from the CHIP data is 91,763 yuan, and from the NBS data, 90,105 yuan. These numbers differ by less than 2 percent. 22

There are more missing values in the NBS data (5,343 − 5,112 = 231) than in the CHIP data (5,343 − 5,290 = 53).

134

Hiroshi Sato, Terry Sicular, and Yue Ximing

Table 3A.2. Comparison of urban housing market values from the CHIP and the NBS, 2002 Source of data

Number of households

Mean

Minimum

Maximum

CHIP NBS

Households with nonmissing market values in either the CHIP or the NBS 5,290 96,701 100 1,010,000 5,112 90,104 1,500 1,020,000

CHIP NBS

Households with nonmissing market values in both the CHIP and the NBS 5,062 91,763 100 980,000 90,105 200 1,020,000

Note: Unweighted.

Table 3A.3 gives the same comparison for market rents from the two data sources (again, unweighted). The mean rent from the CHIP is markedly higher than that from the NBS. For the same subsample of households (lower panel), the CHIP rent is 3.5 times that of the NBS rent. The difference in the reported rental value of housing between the two data sources is so large that one must make a judgment about which source is more reliable. We carried out several checks to identify which source provides more reasonable values of housing rents. Useful here was an analysis of the rentprice ratio, that is, the ratio of the rent to the market value of the dwelling, usually multiplied by 100. The rent-price ratio is a crude measure of the rate of return to housing assets. As shown in Table 3A.4, the average rent-price ratio (unweighted) is much higher for the CHIP data than for the NBS data. The CHIP data yield a ratio of 15, that is, the average market rent is 15 percent of the housing value; the NBS data yield a ratio of only 2.25. Available data for other countries typically reveal national average rent-price ratios for private housing in the Table 3A.3. Comparison of urban market rental values of housing from the CHIP and the NBS, 2002 Source of data

Number of households

Mean

Minimum

Maximum

CHIP NBS

Households with nonmissing values in either the CHIP or the NBS 5,266 5,344 240 60,000 4,985 1,396 120 48,000

CHIP NBS

Households with nonmissing values in both the CHIP and the NBS 4,909 4,864 240 60,000 1,402 120 48,000

Note: Unweighted. Rents are for twelve months.

Housing Ownership, Incomes, and Inequality in China, 2002–2007

135

Table 3A.4. Comparison of the urban rent-price ratio from the CHIP versus that from the NBS, 2002 Source of Data

Number of households

Mean

Minimum

Maximum

CHIP NBS

Households with nonmissing values in either the CHIP or the NBS 5,251 14.89 0.45 14,423.08 4,906 2.24 0.0975 240

CHIP NBS

Households with nonmissing values in both the CHIP and the NBS 4,820 15.41 0.45 14,423.08 2.25 0.0975 240

Note: Unweighted. The rent-price ratio is equal to the annual rent divided by the market value of the dwelling, times 100.

range of 3–10, although within countries the ratio in particular cities or local markets may be higher.23 Reports on the rent-price ratio for urban China generally give numbers below six.24 Such evidence suggests that the rent data from the NBS may be more reasonable than the rent data from the CHIP. Also, the CHIP rental data appear to be noisier than the NBS rental data. The maximum value of the rent-price ratio in the CHIP data is 14,423, too high to be regarded as believable. The 99th percentile for the CHIP rent-price ratio is 171, a much lower number. Still, it seems unlikely that 1 percent of the urban households truly have rent-price ratios exceeding 171. Calculated using the NBS data, the maximum rent-price ratio is 240, still high but not so stratospheric, and the 99th percentile is 12, a more believable number. We compared the CHIP and NBS market rent and housing value data for all households with rent-price ratios greater than 50. We found that for most of these households, the market value from the CHIP is one digit less than that from the NBS, and in most cases the missing digit is a zero. Moreover, for some of these observations the reported mortgage exceeded the value 23

24

For a recent study of British ratios, see http://www.dataspring.org.uk/Downloads/2009– 16%20HA%20&%20private%20RoR%20FINAL.pdf. Accessed July 28, 2011. For U.S. data, see http://www.sciencedirect.com/science? ob=ArticleURL& udi=B6WMG-4WM 74XR-1& user=940030& coverDate=09%2F30%2F2009& rdoc=1& fmt=high& orig= search& sort=d& docanchor=&view=c& searchStrId=1286891102& rerunOrigin= scholar.google& acct=C000048763& version=1& urlVersion=0& userid=940030&md 5=904aa18ac2e6317610a280323124e141. Accessed July 28, 2011. Leonhardt (2011) gives data for the United States that imply a rent-price ratio of about 0.10 from 1989 to 2000, then falling to 0.05 in the mid-2000s and recovering slightly to 0.067 in 2010. A report on rent-price ratios for China is provided in http://www.globalpropertyguide. com/Asia/China/Rental-Yields. Accessed July 28, 2011.

136

Hiroshi Sato, Terry Sicular, and Yue Ximing

of the house. This leads us to believe that the CHIP interviewers copied the NBS data onto the CHIP questionnaire, but with some transcription errors. Based on the preceding, we conclude that the 2002 NBS data on housing rents and housing values are more reliable than are those from the CHIP. Consequently, we use the NBS data on market rents and housing values for our analyses. This choice is also advantageous because in 2007 we only have NBS data on housing rents and housing values, and use of the NBS data for both years allows for consistent comparisons over time. A minor drawback of the NBS data is that they contain more missing values than the CHIP data. For 2002 housing market values, which are fairly consistent between the NBS and CHIP sources, we use the CHIP value when the NBS value is missing. However, even after these replacements, market values are still missing for a few households. For these households, we estimate the market value by multiplying the area of the dwelling by the average NBS market value per square meter for all households in the same urban district.

C. Costs of Ownership and Alternative Estimates of Imputed Rents We carried out alternative estimates of imputed rental income from owneroccupied housing. Table 3A.5 summarizes the alternative calculations. The base estimates (A) are simply equal to the market value of the housing times the rate of return. We follow standard practice in the housing literature and set the rate of return equal to the interest rate on long-term government bonds. The base estimates can be calculated for all subsamples in all years, except for the 2007 migrant subsample, for which we use the reported market rents (housing values are not available for the migrant subsample in 2002). Where possible, given data availability, we have calculated alternative estimates of imputed rents that deduct the costs of ownership, such as depreciation and interest payments on housing debt. These alternative estimates allow us to evaluate possible biases in our base estimates. Studies of imputed rents on owner-occupied housing typically estimate depreciation by multiplying the housing value by a depreciation rate. It is also possible to incorporate costs of maintenance and repairs as part of the depreciation costs. Household spending on repairs and maintenance affects the rate of economic depreciation – the rate of depreciation is higher for housing that is not maintained or repaired – so in fact these two types of

Housing Ownership, Incomes, and Inequality in China, 2002–2007

137

Table 3A.5. Formulae for alternative estimates of imputed rental income on owner-occupied housing incorporating costs of ownership, 2002 and 2007 2002

2007

A. Base estimate

Rural: R = .032028V Urban: R = Rm Migrant: R = .032028V

Rural: R = .043625V Urban: R = Rm Migrant: R = Rm

B. Base minus depreciation

Rural: R = .032028V – .01V Urban: R = Rm – .01V Migrant: R = .032028V – .01V

Rural: R = .043625V – .01V Urban: R = Rm – .01V Migrant: R = Rm – .01(Rm /.043615)

C. Base minus depreciation and mortgage interest

Rural: R = .032028V – .01V – .052028M Urban: R = Rm – .01V – .052028M Migrant: R = .032028V – .01V

na

Notes: For these estimates we have set the rate of return equal to the average annual interest rate on long-term (thirty-year) Chinese government bonds in 2002 and 2007, 0.032028 and .043615, respectively. Interest on mortgages is set equal to the interest on long-term Chinese government bonds plus two percentage points. Depreciation (inclusive of repairs and maintenance costs) is calculated using a depreciation rate of 1.0 percent. Note that for migrants we do not have mortgage data for either year; also, for migrants we only have information on housing values in 2002 and only on market rents in 2007. Estimates for migrants make use of the data available in each year (see Section III for additional discussion).

costs are closely related (Wilhelmsson 2008). For our calculations, we use a rate of depreciation that reflects the depreciation of housing that is not well maintained or repaired.25 For 2002, we have data on mortgages; therefore, we can estimate interest costs associated with housing mortgages. We follow common practice in the literature and assume that the mortgage interest rate is equal to the rate of interest on long-term government bonds plus 2 percentage points. 25

The economics literature on housing depreciation contains a range of estimates for depreciation rates. Wilhelmsson (2008) surveys recent work and reports that estimates that do not control for maintenance and repairs range from about 0.3 percent to 1.0 percent, with several studies finding rates of about 0.7 percent. He then estimates the effect of maintenance and repairs on housing depreciation rates. He finds that the difference between the depreciation rates for housing that is well maintained, as opposed to that for housing that is not well maintained, is 0.25–0.30 percentage points. We therefore use a depreciation rate of 0.7 percent plus 0.3 percent, or a total of 1.0 percent. Although this approach admittedly is crude, it approximately captures the effects of such costs so that we can examine their impact on our results.

138

Hiroshi Sato, Terry Sicular, and Yue Ximing Table 3A.6. Gini coefficients of household per capita income, calculated using different estimates of imputed rental income, 2002 and 2007 Formula used for estimation of imputed rents A B C A B C A B C A B C

Sector

2002

2007

rural rural rural urban excl. migrants urban excl. migrants urban excl. migrants urban with migrants urban with migrants urban with migrants national national national

0.3648 0.3656 0.3661 0.3255 0.3260 0.3272 0.3276 0.3279 0.3289 0.4501 0.4490 0.4493

0.3670 0.3675 n.a. 0.3386 0.3395 n.a. 0.3367 0.3376 n.a. 0.4758 0.4736 n.a.

Notes: Formulae for calculating the different estimates of imputed rents at the household level are given in Table 3A.5; these values are divided by household size to obtain the per capita values of imputed rents and are added to household per capita income. The Gini coefficients are calculated with weights and cover both homeowner and renter households, as well as long-term rural-urban migrants (included in the calculation of the estimates of the Gini for the urban with migrants and for the national). n.a. = not applicable.

Our main concern is how including costs of ownership in imputed rents affects the measured inequality of income. Table 3A.6 shows estimates of the Gini coefficient for household per capita incomes calculated using each of the three alternative estimates of imputed rental income. Estimates (A) use our base estimates of imputed rents, that is, the rate of return times the market value of housing. Estimates (B) subtract depreciation and maintenance/repair costs. Estimates (C) also subtract the mortgage interest costs. These last estimates can be calculated only for 2002. We find that inequality of household per capita income as measured by the Gini coefficient is little affected by the inclusion of the costs of ownership. This is true for both the urban and rural sectors as well as nationwide. In all cases, including the costs of ownership changes the Gini coefficient of household income per capita by less than 1 percent from its value calculated using the base estimate of imputed rents (A). We therefore conclude that for the purpose of analyzing income inequality, we can use estimates of imputed rents that do not subtract the costs of ownership.

Housing Ownership, Incomes, and Inequality in China, 2002–2007

139

References Arrondel, L. and B. Lefebvre (2001), “Consumption and Investment Motives in Housing Wealth Accumulation: A French Study,” Journal of Urban Economics, 50(1), 112–137. Burger, R., F. Booysen, S. van der Berg, and M. von Maltitz (2008), “Marketable Wealth in a Poor African Country: Wealth Accumulation by Households in Ghana,” in J. B. Davies, ed., Personal Wealth from a Global Perspective, 248–268, New York: Oxford University Press. Buttimer, R.J., A.Y. Gu, and T. Yang (2004), “The Chinese Housing Provident Fund,” International Real Estate Review, 7(1), 1–30. Cagetti, M. and M. De Nardi (2008), “Wealth Inequality: Data and Models,” Macroeconomic Dynamics, 12 (Supplement S2), 285–313. Campbell, J.Y. (2006), “Household Finance,” Journal of Finance, 6(4), 1553–1604. Chen, Z., J. Chen, and S. Liu (2008), “Ande guangsha qianwanjian: Zhongguo chengzhen zhufang tizhi shichanghua gaige de huigu yu zhanwang” (Retrospect and Prospects for the Marketization of the Chinese Urban Housing System), Shijie jingji wenhui, no. 1, 43–54. Cheng, S., ed. (1999), Zhongguo chengzhen zhufang zhidu gaige (Urban Housing Reform in China), Beijing: Minzhu yu jianshe chubanshe. Chongqing Fuling Municipal Bureau of Land and Resources (2009), “Chongqingshi tongchou chengxiang fazhanzhong cujin nongcun zhaijidi liuzhuan duice yanjiu” (Research on Measures to Promote Rural Housing Land Transfers in Chongqing City’s Overall Urban-Rural Development), December 15, at http://www.flgt.gov.cn/ html/1/tdgl/jsydsp/zjdgl/news 1150 4342.html. Accessed September 7, 2011. Davies, J.B. and A.F. Shorrocks (2000), “The Distribution of Wealth,” in A.B. Atkinson and F. Bourguignon, eds., Handbook of Income Distribution, Vol. 1, 605–675, New York: Elsevier. Deaton, A S. (1990), “Saving in Developing Countries: Theory and Review,” World Bank Economic Review, Proceedings of the 1989 World Bank Annual Conference on Development Economics, 61–96. de Brauw, A. and J. Giles (2008), “Migrant Labor Markets and the Welfare of Rural Households in the Developing World: Evidence from China,” World Bank Policy Research Working Paper, No. 4585, The World Bank, Washington, DC. Dong, J. and C. Yao (2011), “Ze’ou biaozhun: Ershiwunian de shanbian (1986–2010)” (Choice of Mate: Changes in 25 Years, 1986–2010), Zhongguo qingnian yanjiu, no. 2, 73–78. He, H. and N. Deng (2009), “Biange shidaide Zhongguo nongcun zhufang fazhan zhuangkuang: Chengjiu yu tiaojian” (Progress and Challenges of Rural Housing in Changing China), Gansu lianhe daxue xuebao (Shehui kexue ban), 25(3), 66–70. Huang, Y. and C. Yi (2010), “Consumption and Tenure Choice of Multiple Homes in Transitional Urban China,” International Journal of Housing Policy, 10(2), 105–131. Ioannides, Y.M. and S.S. Rosenthal (1994), “Estimating the Consumption and Investment Demands for Housing and Their Effect on Housing Tenure Status,” Review of Economics and Statistics, 76(1), 127–141. Jia, K. and J. Liu (2007), “Woguo zhufang gaige yu zhufang baozhang wenti yanjiu” (Study on Housing Reform and Housing Welfare), Caizheng yanjiu, no. 7, 8–23.

140

Hiroshi Sato, Terry Sicular, and Yue Ximing

Leonhardt, D. (2011), “Rent or Buy, A Matter of Lifestyle,” New York Times, May 11, B1. Li, S. and R. Zhao (2008), “Changes in the Distribution of Wealth in China, 1995–2002,” in J.B. Davies, ed., Personal Wealth from a Global Perspective, 93–111, New York: Oxford University Press. Luo, H. (2009), “Zhengfu zhongshi minsheng buneng hushi zhufang baozhang zeren: Dui woguo zhufang zhengce de huigu yu sikao” (The Government Should Assign Priority to Housing Welfare: A Retrospect and Reflection on Housing Policy), Renminwang (People’s Daily Online), February 1, http://politics.people.com.cn/GB/8198/ 44004/44005/8727122.html. Accessed September 7, 2011. Meng, X. (2007), “Wealth Accumulation and Distribution in Urban China,” Economic Development and Cultural Change, 55(4), 761–791. National Bureau of Statistics (NBS) (various years), Zhongguo tongji nianjian (China Statistical Yearbook), Beijing: Zhongguo tongji chubanshe. Pollock, A.J. (2010), “Housing Finance in International Perspective,” testimony at the hearing on “Comparison of International Housing Finance Systems” to the Subcommittee on Security and International Trade and Finance, Committee on Banking, Housing, and Urban Affairs, U.S. Senate, September 29, http://www.aei.org/docLib/ Testimony-Comparison-International-Housing-Finance-Systems-Pollock.pdf. Accessed June 4, 2011. Qin, H. and T. Zhong (2009), “Woguo nongcun zhufang zhidu gaige jiben xianzhuang yu zhengce jianyi” (Current Situation of the Reform of the Rural Housing System and Some Policy Implications), Jingji yaocan, no. 78, 16–23. Quadrini, V. and J.V. R´ıos-Rull (1997), “Understanding the U.S. Distribution of Wealth,” Federal Reserve Bank of Minneapolis Quarterly Review, 21(2), 22–36. Ren, B. and W. Kang (2003), “Fanggai miju” (Confusing Maze of Housing Reform), in Caijing zazhi bianjibu, ed., Zhuanxing Zhongguo (Transitional China), 40–52, Beijing: Shehui kexue wenxian chubanshe. Ruo, M. (2009), “Nongcun zhaijidi liuzhuan Jiaxing moshi diaocha” (An Investigation into the Jiaxing Model of Rural Housing Land Transfers), Ziyuan yu renju huanjing, no. 24, 49–50. Sato, H. (2006), “Housing Inequality and Housing Poverty in Urban China in the Late 1990s,” China Economic Review, 17(1), 37–50. Saunders, P. and P. Siminski (2005), “Home Ownership and Inequality: Imputed Rent and Income Distribution in Australia,” University of New South Wales Social Policy Research Centre Discussion Paper No. 144, Sydney. Short, K., A. O’Hara, and S. Susin (2007), “Taking Account of Housing in Measures of Household Income,” paper prepared for the Annual Meeting of the Allied Social Sciences Associations, Chicago. Sierminska, E. and T.I. Garner (2005), “A Comparison of Income, Expenditures and Home Market Value Distributions Using Luxembourg Income Study Data from the 1990’s,” Working Paper No. 380, U.S. Bureau of Labor Statistics, Washington, DC. Smeeding, T.M. and D.H. Weinberg (2001), “Toward a Uniform Definition of Household Income,” Review of Income and Wealth, 47(1), 1–24. Sun, W. and Q. Hua (2009), “Nongcun zhaijidi liuzhuan Jiaxing moshi diaocha” (An Investigation into the Jiaxing Model of Rural Housing Land Transfers), Diyi caijing ribao, November 25.

Housing Ownership, Incomes, and Inequality in China, 2002–2007

141

Tomba, L. (2004), “Creating an Urban Middle Class: Social Engineering in Beijing,” China Journal, no. 51, 1–26. Wilhelmsson, M. (2008), “House Price Depreciation Rates and Level of Maintenance,” Journal of Housing Economics, 17(1), 88–101. Wu, J., J. Gyourko, and Y. Deng (2010), “Evaluating Conditions in Major Chinese Housing Markets,” NBER Working Paper No. 16189, National Bureau of Economic Research, Cambridge, MA. Xin, C. and H. Zhou (2009), “Qianxi nongcun zhufang zhidu gaige: Yi Ningbo weilie” (A Simple Analysis of the Reform of the Rural Housing System: Using Ningbo as an Example), Nongcun jingji yu keji, 20(10), 31–33. Xu, C. and X. Kong (2009), “Gaige kaifang sanshi nian lai nongcun zhaijidi zhidu bianqian: Pingjia ji zhanwang” (Changes in the Management System of Rural Land for Housing Use: Evaluation and Prospects), Jiage yuekan, no. 8, 3–5, 15. Yemstov, R. (2008), “Housing Privatization and Household Wealth in Transition,” in J.B. Davies, ed., Personal Wealth from a Global Perspective, 312–333, New York: Oxford University Press. Zax, J.S. (2003), “Housing Reform in Urban China,” in N.C. Hope, D.T. Yang, and M.Y. Li, eds., How Far Across the River? Chinese Policy Reform at the Millennium, 313–350, Stanford, CA: Stanford University Press. Zhang, L. (1998), Gaobie lixiang: Renmin gongshe zhidu yanjiu (Farewell to the Ideal: A Study on the People’s Commune System), Shanghai: Dongfang chuban zhongxin. Zhao, R. and S. Ding (2008), “The Distribution of Wealth in China,” in B. A. Gustafsson, S. Li, and T. Sicular, eds., Inequality and Public Policy in China, 118–144, New York: Cambridge University Press. Zhao, S. (2006), “Shilun nongcun zhaijidi zhidu gaige” (Preliminary Discussion of the Reform of the System of Rural Housing Land), Nongcun gongzuo tongxun, no. 10, 32–34.

FOUR

Educational Inequality in China The Intergenerational Dimension John Knight, Terry Sicular, and Yue Ximing

I. Introduction The intergenerational distribution of education has received less attention from economists than has the intragenerational distribution. Yet, the degree of intergenerational transmission of education – the transfer of educational outcomes from parents to children – is an important determinant of the distribution of education among households at any point in time. This, in turn, influences the distribution of income among households. There are two concepts of intergenerational mobility. One focuses on aggregate mobility, that is, the extent to which the average education of one generation exceeds that of the previous generation. In the aggregate, economic growth, household incentives, and the policies of the state can all serve to promote mobility. A second concept focuses on mobility at the microeconomic level, that is, the extent to which the education of an individual depends on, or is related to, the education of her parents. In this case, state policies that equalize educational opportunities may be offset by the tendency for children of better-educated parents to receive more education than children of less well-educated parents. In this chapter we examine both concepts of mobility. The China Household Income Project (CHIP) 2007 survey contains information about the education of the parents of the household head and of the spouse of the household head, including parents who are not present in the household. Consequently, the data set contains matched information on one’s own and one’s parental education for a large and relatively complete sample. We use this information to analyze the intergenerational mobility of education in The authors gratefully acknowledge funding from the Social Sciences and Humanities Research Council of Canada and the Ontario Research Fund–CIGI–UWO China Project, and research assistance by Jerry Lao and Yichuan Zhang.

142

Educational Inequality in China

143

both the aggregate and at the microeconomic level. As our sample spans individuals born over a long period – from the 1930s through the 1980s – we can trace the evolution of this relationship in response to changes in policies and other factors over more than half a century since the founding of the People’s Republic of China (PRC).

II. Literature In many and diverse countries the parents’ education has been found to be a powerful positive determinant of their children’s education, thus reducing the extent of household intergenerational educational mobility (see, for instance, Bowles [1972] for the United States; Couch and Dunn [1997] for the United States and Germany; Lillard and Willis [1994] for Malaysia; Thomas [1996] for South Africa; Knight and Sabot [1990] for Kenya and Tanzania; Binder and Woodruff [2002] for Mexico; and Hertz et al. [2007] for an international summary). This can be the case even when education is heavily subsidized. For instance, in the United Kingdom in 1995, 80 percent of young people from households classified by the father’s occupation (likely to be closely correlated with education) as being in the highest social class (out of five classes) were enrolled in higher education. By contrast, only 12 percent of those in the lowest social class were enrolled in higher education. Yet, at that time, access to higher education was effectively free for poor students (National Committee of Inquiry into Higher Education 1997). In their comprehensive survey of the legacy of educational inequality, Hertz et al. (2007) provide comparable estimates based on an analysis of national household surveys in forty-two countries over a fifty-year period. Specifically, they report the estimated coefficients from simple regressions of a child’s education on the parents’ average years of education; they also calculate the correlation coefficients between these two variables. They report both the regression coefficients and the correlations for all ages pooled and for five-year birth cohorts. For most countries, the regression coefficient falls over time; that is, the cohort-specific effects of the number of years of the parents’ education on the number of years of their child’s education is higher for older cohorts and lower for younger cohorts. In contrast, the correlation coefficients display no significant time trend; that is, across cohorts, the variation in the parents’ education is associated with an unchanged proportion of the variation in the child’s education. Some of the plausible explanations for these patterns are examined in the following discussion of our results for China.

144

John Knight, Terry Sicular, and Yue Ximing

Is the same true for China? Existing studies suggest that this may be so. Knight and Li (1993) find that in 1988 (based on data from the first CHIP survey), spatial considerations – both regional and rural-urban – were important determinants of educational attainment in China. Other than age, the most important factor influencing a person’s years of schooling was whether he or she lived in a rural or an urban area. This is due to the separate administration and funding of rural and urban education and to the differences in opportunity costs and prospective economic returns. It was also found that the education of parents assisted the education of their children. In both urban and rural areas, the mother’s education was more important than the father’s, and in rural areas, the education of both parents had a greater effect on the education of daughters than on that of sons, suggesting that female education is more discretionary. The transmission of education from one generation to another was strengthened by the tendency of the educated to intermarry. Knight, Li, and Deng (2009), using the rural sample of the 2002 CHIP survey, examine the determinants of enrollment in junior and senior middle school. They find that dropping out from junior middle school was more likely if the child was from a household in the lowest quintile of income per capita and if the mother was poorly educated. Continuing to senior middle school was more likely when there was higher household income per capita, with more years of education of both the father and the mother, and if the household was not credit-constrained. The household income level and the parents’ education improved performance at school, thus increasing the chances of receiving more education. The authors argue that a vicious circle of both parental income poverty and parental education poverty held back the postprimary education of the next generation. Sato and Li (2007) use the rural sample of the CHIP 2002 survey to examine the influence of class background (chengfen) on the education of offspring. They find that the offspring of landlord or rich-peasant families (as officially classified) are likely to have more education than are the offspring of other families, even after controlling for parental education, family wealth, and other household characteristics. They attribute this effect in the postreform period to an education-oriented family culture, possibly a reaction to the class-based social discrimination of the prereform era. Moreover, they find important cohort effects, depending on government policies at the time that the child is of school age. For instance, in the pre-Mao and postMao periods, a lower proportion of children of landlord and rich-peasant families, as compared to children of other families, had six or fewer years of education, and a higher proportion had nine or more years of education.

Educational Inequality in China

145

By contrast, in the mid-Mao period (affecting children born in the 1945– 1959 period), 54 percent of the former had six or fewer years of education, compared with only 38 percent of the children of poor-peasant households, and 22 percent of the former had nine or more years, compared with 38 percent of the latter.

III. Education Policies and Trends in China China’s educational policies have passed through distinct phases over time, each with different implications for the relationship between the education of the parents and that of their children. In some periods the aim has been to pursue universal access to basic education. These periods are characterized by broad-based expansion of enrollments and rising levels of educational attainment. One would expect a weak relationship between parental education and child educational attainment during these periods. In other periods, educational policies have been shaped by the goal of training skilled labor to support economic growth. During these periods, enrollments and progression rates dipped, reflecting an emphasis on quality and selectivity rather than on universal access, and one might expect a strengthening of the relationship between parental and child education, depending on the criteria for selection and other relevant factors. Here we provide a brief survey of policy changes and trends most relevant to the intergenerational transmission of education, with a focus on primary and middle-school education.1 Our survey covers the period from 1950 until the mid-2000s, the time frame covered in our empirical analysis. The early years of the PRC (1949–1952) saw the recovery of the educational system and steps in the direction of nationalization of schools. At this time, the government articulated the goals of popularizing education and eliminating illiteracy (Hannum 1999). Both formal and alternative schooling expanded rapidly. From 1949 to 1952 enrollments in primary school rose from 24 to 50 million; in middle school, from 1.26 to 3.15 million; and in tertiary school, from 117,000 to 191,000 (Ministry of Education, Department of Planning 1984: 22–23; Hannum 1999: 196). These numbers include many older students, reflecting an effort to increase the levels of education of adults as well as of children. With the First Five-Year Plan (1953–1957), China embarked on its first comprehensive, Soviet-style economic plan. Education was an integral part 1

We do not discuss tertiary education because until very recently enrollments were low; nor do we discuss adult and nonformal education because policies in these areas are complex, and these forms of education are not well captured in the data.

146

John Knight, Terry Sicular, and Yue Ximing

of this plan (Ministry of Education, Department of Planning 1984: 9). Rapid industrialization was the central national goal, and training the skilled workers needed for rapid industrialization took priority (Hannum 1999; L¨ofstedt 1980: 79). Resources were directed to senior middle and higher levels of education and to specialized and technical training. At this time, government funding for schools was largely limited to urban areas; rural primary schools were funded by rural communities. As Hannum (1999: 197) writes, “In short, the priority placed on rapidly developing urban higher-level education limited the resources available for basic educational expansion; allocation of resources for basic education prioritized small numbers of urban ‘key-point’ schools likely to produce quick results.” Although the number of middle and tertiary educational facilities grew, expansion of enrollments, especially at the senior middle and university levels, was hampered by a shortage of individuals with sufficient prior schooling. The number of senior middle school graduates fell short of the enrollment targets for higher education (Niu 1992: 24–25). Students who had not completed senior middle school were recruited for university; as a result, the number of university entrants exceeded the number of senior middle school graduates (Thøgersen 1990: 22). During the First Five-Year Plan period, due in part to the shortage of skilled workers, members of the former elite social classes were not prevented from attending primary school or progressing on to senior middle school or university (Niu 1992: 19). Efforts were made, however, to expand access for those with worker and peasant class backgrounds, and preferential policies were adopted for children of party cadres (Niu 1992: 25–27). Figure 4.1 shows trends over time in net enrollment rates in primary school and in progression rates from primary to junior middle and from junior middle to senior middle school. The primary net enrollment rate increased from 49 percent in 1952 to 54 percent in 1955. In 1956–1957 the primary net enrollment rate jumped to more than 60 percent. At this time, there was no obvious change in education policies, but rural areas underwent dramatic institutional changes that affected the demand for education. Starting in 1955, China embarked on a campaign to raise the degree of collectivization in rural areas to a higher level. Rural households were organized into “advanced” agricultural producer cooperatives, typically encompassing more than 100 families, characterized by collective ownership of land, farm tools, and livestock, and in which the cooperative distributed income to households based on labor days or work points. The speed of institutional transformation was rapid. In 1955, less than 1 percent of rural households in China belonged to advanced agricultural

Educational Inequality in China

147

100.0

80.0

60.0

40.0

20.0

1952 1954 1956 1958 1960 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008

0.0

primary net enrolment ratio

jr middle progression rate

sr middle progression rate

jr middle progression rate (estimate)

sr middle progression rate (estimate)

Figure 4.1. Primary Net Enrollment and Middle-School Progression Rates, 1952–2008 Notes: The primary net enrollment rate is equal to the number of primary schoolage children enrolled in primary school divided by the number of primary school-age children in the population. Progression rates are calculated as the number of entrants to the given level of schooling divided by the number of graduates from the prior level of schooling. These data are from the same year; that is, entrants to school in August/September are divided by graduates who finished school several months earlier, that is, in June/July of the same year. The senior middle-school progression rate includes entrants to technical middle schools. Progression rates to junior and senior middle school for the years prior to 1978 are only published for selected years. For the 1950s to the 1970s, we have calculated estimates using published data on the numbers of graduates from and entrants to regular junior and senior middle schools. These estimated progression rates (dotted lines) in most cases are similar to the available published progression rates (squares and triangles). Sources: NBS (1996, 2001, 2009); Ministry of Education, Department of Planning (1984, 1991); Ministry of Education, Department of Development and Planning (2008).

producer cooperatives; the other 99 percent either engaged in standard household farming or participated in smaller-scale mutual aid teams and cooperatives where the land and other assets were still privately owned. By December 1956, 88 percent of rural households belonged to advanced

148

John Knight, Terry Sicular, and Yue Ximing

agricultural producer cooperatives. Most of the remaining 12 percent were in farms, but of a less “advanced” nature. Thus, in the space of eighteen months, household farming and private ownership had effectively disappeared (Riskin 1987: 86; Walker 1966: 35). In the new institutional context, the contribution of children to household income was substantially reduced, with implications for the demand for schooling. In addition, rural schools were funded by the rural communities, and the advanced cooperatives had the capacity to mobilize the resources needed to build and support schools. Thus, the supply of schools expanded. The result was a marked increase in both the number of primary schools and primary school enrollments (see Figure 4.1) (Ministry of Education, Department of Planning 1984: 20–21). In the 1950s there were few senior middle schools located in the rural areas (Thøgersen 1990: 22), so trends in senior middle-school enrollments at that time reflect the situation in urban schools, which were largely government funded. Enrollments in middle schools of all types, including regular junior and senior middle schools as well as specialized middle schools, rose annually during the First Five-Year Plan, reaching 7.1 million by 1957. Enrollment in regular junior middle schools, which accounted for more than 70 percent of middle-school enrollments, more than doubled from 2.2 million in 1952 to 5.4 million in 1957 (Ministry of Education, Department of Planning 1984: 22–23). Progression rates to junior middle school increased steadily from 30 percent in 1953 to 45 percent in 1957 (Figure 4.1). Progression rates to senior middle school also increased but were variable, probably reflecting changes from year to year in the intake of older students. With the launch of the Great Leap Forward in 1958, educational priorities shifted to the left. Universal access to primary education became a central goal, as did the extension of higher levels of education to rural townships and counties (L¨ofstedt 1980: 96). The expansion of rural schooling was facilitated by a reorganization of the advanced agricultural producer cooperatives into communes, a larger and even more “advanced” socialist form of collective organization that typically encompassed 5,000 households (Riskin 1987: 123). Alternative approaches to education were encouraged, including shortening and combining different levels of schooling and combining work with school (Hannum 1999). The educational agenda now became more politicized in terms of both curriculum and its emphasis on mass education for the proletariat rather than for the elites. Enrollments in all levels of school rose dramatically. Primary school enrollments increased from 64 million in 1957 to more than 90 million in 1959. Senior middle-school enrollments jumped from 7 million to over

Educational Inequality in China

149

12 million (Ministry of Education, Department of Planning 1984: 22–23). As shown in Figure 4.1, from 1957 to 1958 the primary school net enrollment rate rose from 62 percent to 80 percent, and the junior middle progression rate increased from 44 percent to 62 percent. The senior middle progression rate also rose substantially. With the failure of the Great Leap and the ensuing famine in 1960 and 1961, enrollment and progression rates plunged. The focus of China’s educational policies reverted to the training of skilled workers and a focus on quality rather than on quantity (L¨ofstedt 1980: 102). By 1962 the economy had stabilized and the education system began to recover. Emphasis was placed on the development of schools at the middle and tertiary levels, and the rates of progression to junior and senior middle schools rose. The effects of these policies were most evident in the urban areas, where progression rates to junior middle school were 90 percent or higher, and to senior middle school about 40 percent. In the rural areas the progression rates were lower, less than 30 percent for junior middle schools and less than 10 percent for senior middle schools.2 Rural students’ access to senior middle and higher levels of education was affected by the low quality of rural primary education; senior middle and tertiary schools were mainly located in the urban areas (Niu 1992: 56; Thøgersen 1990: 26). At this time, a two-track system of education was used to balance the objective of universal schooling with the need to train skilled workers. The government invested in a system of state-funded and high-quality key schools (Niu 1992: 45; Thøgersen 1990: 26). Entry to key schools and universities was based in part on political criteria, so that children of former capitalists and landlords were screened out, and in part on academic performance, thus benefiting the children of cadres, the intelligentsia, and the middle classes (Niu 1992: 50; Thøgersen 1990: 26). Educational policies again shifted to the left in the late 1960s with the launch of the Cultural Revolution. Political struggles during the peak years of the Cultural Revolution (1966–1969) brought chaos to the educational system. Universities were closed, as were many middle and primary schools, especially in the urban areas. Elitism was criticized, and the system of key schools was abolished (Thøgersen 1990: 28). In addition, egalitarian wage 2

Separate urban and rural progression rates provided in the text are calculated using the numbers of graduates from rural (or urban) primary and junior middle schools and the number of entrants into rural (or urban) junior middle and senior middle schools, as provided in Ministry of Education, Department of Planning (1984). Only students graduating from and entering regular senior middle schools are included in these calculations; specialized and vocational senior middle schools are excluded.

150

John Knight, Terry Sicular, and Yue Ximing

structures were adopted in both urban and rural areas, reducing the financial returns to investments in education. Data on educational trends during this period are incomplete, but available information indicates that progression rates declined markedly (Figure 4.1). In 1970 the government took steps to restore and reconfigure the educational system. Primary and middle schools reopened, and the government adopted policies to promote rural education, especially at the senior middle-school level. At this time, funding and administration of primary and middle schools were the responsibility of urban work units and rural collectives (Hannum et al. 2008: 217), although the government provided some subsidies to help pay teachers’ salaries (China Education Almanac Editorial Department 1984: 98–99). Schooling was largely free for households (Hannum et al. 2008: 217). The school curriculum emphasized political and ideological education in a uniform ten-year program (five years primary, three years junior middle, and two years senior middle). Academic achievement was downplayed, and class origin, political attitude, and education through labor were emphasized (Hannum 1999: 199; Niu 1992: 59; Thøgersen 1990: 27). Although the Cultural Revolution era has been criticized for the decline in the quality of education and the disruption of tertiary education, the data reveal that the 1970s were characterized by high primary enrollments and a remarkable expansion of senior middle-school education, especially in the rural areas. Primary school net enrollments reached 90 percent, and progression to junior and senior middle schools rose markedly (Figure 4.1). At their peak in 1976–1977, rural progression rates to junior and senior middle schools had risen as high as 90 percent and 70 percent, respectively. After the death of Mao, China once again changed course. Economic growth became the overriding goal. Educational policies emphasized quality and academic content rather than mass education and politics. In 1977–1978 the key schools and national universities were reopened, with admission based on academic achievement (Niu 1992: 75, 81). In 1981 senior middle school was lengthened to three years (Central Education and Scientific Research Institute 1983: 614–615). Concerns about the quality of education prompted the shutting down of many rural middle schools (Hannum et al. 2008: 219; Pepper 1990: 97). Barriers to schooling based on political criteria and class origin were removed (Niu 1992: 81–83). Trends in education at this time were affected not only by the new education policies but also indirectly by policy reforms in other areas. In the early 1980s China abandoned its experiment with collective farming.

Educational Inequality in China

151

Decollectivization took place rapidly: by 1983, household farming had returned to most of the country, with consequences for both the demand for and the financing of rural schooling. The costs of education that had been borne collectively were shifted to the rural households (Hannum, Park, and Cheng 2007), and the opportunity costs of schooling rose as children could now contribute to household farming. In the mid-1980s China carried out a fiscal decentralization, which had negative consequences for financing education. New measures clarified responsibilities for administration and financing, and encouraged governments at all levels to develop multiple sources of funding for education. In urban areas, district and city governments were responsible for primary and middle schools, respectively; in rural areas, county governments were responsible for senior middle schools, townships were responsible for junior middle schools, and villages were responsible for primary schools. Keypoint schools and universities were managed by the central and provincial governments. Following the fiscal decentralization, government budgetary revenues began a long decline, and local governments increasingly turned to extrabudgetary forms of financing to support public services (Fock and Wong 2008). School funding was more dependent on surtaxes, tuition and fees, profit-oriented school enterprises, and community fund raising (Fock and Wang 2008; Hannum et al. 2008: 220–224; Tsang 2001: 3–4). Urban areas and richer rural localities were better able to generate financial resources, whereas poor rural areas lagged. Educational funding became more unequal (Fock and Wong 2008; Tsang 2001), with implications for access to and quality of education. These developments contributed to changes in schooling patterns, especially at the middle level. In the 1980s national progression rates to junior middle school dropped from about 90 percent to below 70 percent, mainly reflecting changes in rural China, where the proportion of children continuing to junior middle school fell below 60 percent. Progression rates to senior middle school declined by half, from 70 percent to 35 percent. Again, the decline was most severe in the rural areas, where the senior middle-school progression rate fell from 65 percent to about 10 percent. Even in urban areas, progression to senior middle schools declined substantially, from 90 percent to about 50 percent. Concerns about rising educational inequality prompted the 1986 promulgation of the Compulsory Education Law, under which nine years of compulsory education (six years primary school plus three years junior middle school, or five years primary school plus four years junior middle

152

John Knight, Terry Sicular, and Yue Ximing

school) would become universal, but implementation occurred gradually and differentially depending on the level of local capacity (Hannum et al. 2008: 220; Wang 2003; Xing 2007). A gradual recovery in the rate of progression to junior middle school followed. In the rural areas, however, implementation of the law was hampered by ongoing fiscal constraints at county- and lower-level governments (Tsang 2001). In 1994 a tax reform and fiscal recentralization strengthened the central government’s fiscal capacity. Following this reform, government revenue began to recover, but insufficient central-local transfers exacerbated fiscal inequalities at the local level. In many rural areas local governments were unable to meet their expenditure obligations (Fock and Wong 2008; Wong and Bird 2008). In 1995 the government issued a new education law that clarified the responsibilities of the different levels of government, with local governments responsible for middle-school education and below, and implemented a local educational surtax to provide more funding for local education (Wang 2003). Nevertheless, regional inequality in the public financing of education remained high, and regional inequality in middleschool enrollment rates persisted (Dollar 2007: 11–12, 26–27; Li, Park, and Wang 2007; Wang 2003). China’s educational trends began a turnaround in the mid-1990s. At this time, the private returns to education, which had been low by international standards, began to increase (Cai, Park, and Zhao 2008: 185–187). In rural areas, rising returns to education were at first associated with the expansion of off-farm wage employment, initially in township and village enterprises and then later through migrant jobs. Studies have found a positive association between years of education and off-farm wage employment and earnings (de Brauw et al. 2002; de Brauw and Rozelle 2007, Knight, Li, and Deng 2010; Zhang, Huang, and Rozelle 2002; Zhao 1997). Some recent analyses have also found evidence of rising returns to education in agriculture, the result of the market reforms and the growing commercialization of agriculture, but even as late as 2002, the returns to a year of education in farming were only 4 percent (Knight et al. 2010). In urban areas “brain workers” received little if any more pay than did “hand workers” under the egalitarian central planning. Reforms in the employment system, wage structure, and urban enterprise management allowed wage differentials to emerge and expand, with the result that returns to education rose, especially after the early 1990s (Fleisher and Wang 2005; Zhang and Zhao 2007). The 1988 CHIP survey shows an earnings premium of college education over primary education in urban China of only 15 percent, whereas the 2002 CHIP survey shows a premium

Educational Inequality in China

153

of 82 percent (Knight and Song 1993, 2008) Zhang and Zhao (2007) find that the returns to education in urban China rose from 4 percent in 1988 to 11 percent in 2003, with most of the increase occurring in 1992– 1994 and 1997–1999. These developments likely affected the demand for education. In the late 1990s and 2000s the government adopted a series of new measures to strengthen education. In 1999 the government announced that it would expand nine-year compulsory education (targeting the poor areas) and increase senior middle and tertiary enrollments (Tsang 2000: 588). In the early 2000s the government increased central funding to support rural compulsory education and to reduce primary and junior middle education costs borne by rural households (Hannum et al. 2008; World Bank 2007). In 2001 payment of teachers’ salaries was shifted from the village to the county, and the central government implemented transfer payments to help local governments cover the costs of compulsory education (Fock and Wong 2008). In 2003 the central government announced the “Two Exemptions, One Subsidy” policy, under which the government would pay the costs of textbooks and school fees and would provide subsidies for boarding. This program was initially aimed at poor families in central and western China (Hannum et al. 2008: 244; World Bank 2007: 5). In 2006– 2007 the government announced central budgetary funding to finance the elimination all tuition and fees for nine years of compulsory rural education (Dollar 2007: 17; Hannum et al. 2008: 244). These changes in the returns to education and in government education policies were accompanied by substantial increases in the progression rates to junior and senior middle school. The progression rate from primary to junior middle school surpassed 90 percent in 1995, and increased further to 95 percent in 2000 and nearly 100 percent in 2005. The progression rate from junior to senior middle school rose from 45 percent in the early 1990s to 50 percent in the mid-1990s, 60 percent in 1993, 70 percent in 1995, and over 80 percent in 2008. The progression rate from senior middle school to tertiary schooling also rose substantially, from less than 30 percent in the early 1990s to over 70 percent in the mid-2000s (NBS 2009). This brief survey reveals how the substantial changes over time in Chinese government policies and goals affected educational outcomes, with implications for the intergenerational transmission of education. Based on this history, we identify several hypotheses regarding educational outcomes in rural and urban areas. For the rural areas, we propose that three key factors affected the intergenerational transmission of education. The first is government policies to popularize schooling. These policies occurred in several

154

John Knight, Terry Sicular, and Yue Ximing

waves – at the primary level in the 1950s and 1960s, the junior middle and to some degree senior middle levels in the 1970s, and again the junior middle and senior middle levels in the 1990s to the 2000s. The second factor is the private cost of, and returns to, education, which were affected by changes in the organization of farming (collective versus household), and by reforms that strengthened the link between earnings and education. The third is public financing of schools, which affected the supply and quality of rural schools, as well as the costs borne by households. In urban areas, public (or quasi-public) funding of schools was relatively generous, and levels of education were consistently higher than were those in rural areas. From the 1950s onward, primary and junior middle schooling were widespread, so that the point at which educational inequalities became apparent was senior middle school or later. For much of the time, access to middle and postmiddle-school education in urban China was rationed. Access to senior middle school was rationed from the late 1950s through the 1980s, and tertiary education remained rationed at least until the end of the 1990s. The key factor determining intergenerational transmission is the criteria used to select who continues into senior middle and tertiary schools. These criteria changed over time, at times emphasizing academic achievement, and at other times emphasizing politics, with predictable consequences for the role of parental education.

IV. Theory and Methodology The education of children is influenced by various factors, one of which is the education of their parents. We postulate that e = e(a, p , ap , y p , f ; X ),

(1)

where e is the years of one’s own education; a is one’s own unobserved genetic “ability”; p is the observed years of education of the parents; ap is the unobserved genetic “ability” of the parents; yp is the income of the parents at the time of potential educational investment in the child (unlikely to be observed); f is the unobserved nongenetic, noneducation family background, such as a socially acquired “ability”; and X is a vector of other observed and unobserved determinants, such as gender, educational policies and opportunities, and community influences. The education of the parents can causally influence the education of the child through several channels, ceteris paribus. One channel is the possible effect of the parents’ education on family attitudes toward education, on personal confidence, motivation, and ambition, and on knowledge about

Educational Inequality in China

155

the potential returns to education. Furthermore, more educated parents can provide out-of-school human capital and a stimulating home environment that will improve their children’s chances of success, especially in an educational system where continuation in school is rationed and based on school performance. Third, parental education can generate higher incomes: the higher incomes of more educated parents in turn help to overcome the credit constraints on investment in their children’s education. This implies that the full effects of the parents’ education on the child’s education can be measured only if income is omitted from the estimated equation. Of particular interest to policy-minded economists is the causal effect of p on e. Parental education p, however, is likely to be endogenous: it might be influenced by ap , f, and other unobservables. The econometric problem is to separate the causal effect of p from the noncausal association between e and p. Policy prescriptions require measurement of the causal effect. Otherwise, for instance, the consequences of a policy to raise or equalize the educational outcomes of the next generation cannot be predicted accurately. Various methodologies have been used in the literature to measure the causal effect of parents’ education in the likely presence of associated unobserved variables (Lochner 2008). One is to examine the educational differences between cousins whose mothers or fathers are identical twins, on the assumption that the educational differences will not be the result of differences in the parental abilities or environments (for instance, Behrman et al. [1999] for India and Behrman and Rosenzweig [2002] for the United States). However, we cannot use this approach with our data set. A second methodology is to study adopted children, on the assumption that parental genetic influences will be absent (Bj¨orkland, Lindahl, and Plug [2006] for Sweden). Again, this approach is ruled out by the nature of our data. A third methodology is to use instrumental variables, that is, to find a variable or set of variables that is closely associated with the parents’ education but does not have an independent influence on the child’s education; in that way, one can measure the effect of exogenous variations in the parents’ education on the child’s education. Examples of instruments used for this purpose include changes in the age of compulsory schooling (for instance, Black, Devereux, and Salvanes [2005] for Norway and Oreopoulos, Page, and Stevens [2006] for the United States). Here we confine our analysis to simpler methods that measure noncausal associations. The association – whether conditional on observed determinants or unconditional – between the education of one generation and the next is interesting in itself, regardless of the set of forces it might reflect. Moreover, the degree of association has implications for educational

156

John Knight, Terry Sicular, and Yue Ximing

inequality, and thus also for income inequality. It does not matter whether the association between the parents’ education and the child’s education is due to income or due to genetic or socially acquired abilities. The observed variable, education, serves as a proxy for the family endowments that contribute to educational and income inequalities. Here we use the terms “educational persistence” and “educational transmission” interchangeably to denote the association between the education of one generation of a household and the next, irrespective of whether that association indicates the causal effect of parental education. The weaker the degree of educational persistence or transmission, the greater will be the degree of intergenerational educational mobility within the household. Our empirical methodology follows that of Hertz et al. (2007). We estimate simple regressions of one’s own education on parental education, in some cases with additional explanatory variables. The regressions are estimated for the entire sample and separately for the rural and urban samples. In order to analyze changes over time, we also estimate the regressions separately for each five-year birth cohort. Our choice of five-year cohorts to some extent is arbitrary, but a five-year span is long enough to ensure that each cohort has a sufficient number of observations to support a regression, and short enough to allow us to observe the changes over time associated with the different policy periods in China. From the regressions we obtain an estimated coefficient on parental education, which we refer to as β, and the correlation between one’s own education and the parental education, which we refer to as ρ. The βs measure “grade persistence,” and the ρs measure “standardized persistence” (Hertz et al. 2007). These two measures are linked by the formula ρ = β∗ (σp /σo ),

(2)

where σ p and σ o are the standard deviations of parental and one’s own education, respectively. From this equation we can see that the correlation coefficient is “standardized” by the ratio of the standard deviations for the two generations. Thus, for instance, ρ will rise relative to β if a variation in the parents’ education can explain more of the variation in the child’s education, ceteris paribus; that is, if the standard deviation of the child’s education falls relative to that of her parents. Although the βs, and ρs do not identify causality, they quantify the persistence of interpersonal inequality in education from one generation to the next. With respect to educational mobility across generations, lower values of β and ρ, that is, less persistence, would be associated with greater mobility.

Educational Inequality in China

157

V. The Data Empirical analysis of the intergenerational transmission of education requires matched information on one’s own and the parental education. Household surveys are often not well suited to such analysis because they typically contain matched information only when both generations reside in the same household. The 2007 CHIP questionnaire, however, contained questions about the education of the parents of the household head and of the spouse of the household head, including parents who were not present in the household at the time of the survey. This makes possible an analysis of the intergenerational transmission of education with a large and relatively complete sample. The 2007 CHIP data set contains variables on years of completed education and level of education. Level of education measures whether the individual has ever attended that level of education. For example, if the stated level of education is primary school, then that individual has attended, but may or may not have completed primary school.3 Data on years of completed education and level of education are available for individuals who resided in the household at the time of the survey. For parents who were not resident members of the household, the CHIP data set only contains information on the level of education. For these parents, we must translate the levels of education into years of completed education. Categorical variables on education levels are common in the literature, and researchers typically translate them into a continuous variable on years of education by making some simple assumptions. Here we follow a standard approach, as explained in the Appendix to this chapter. In our analysis we confine our sample to individuals born before 1985. Our sample thus contains only individuals who completed school (in China the standard age for graduation from post senior-middle four-year institutions is twenty-two). We exclude younger individuals to avoid censored information on years of education for those who may still be in school. We use data from the 2007 CHIP rural and urban surveys but not from the separate migrant survey because relevant data for the migrant sample are incomplete and the migrant sample is difficult to incorporate into our analysis. Consequently, the urban component of our analysis includes only individuals with formal urban household registration (hukou). Migrants, however, are present in the analysis, because the rural survey contains 3

This approach to defining the level of education is consistent with that taken by the NBS in its rural and urban household surveys.

158

John Knight, Terry Sicular, and Yue Ximing

individuals engaged in short-term migrant work, and the urban survey contains individuals who originated in rural areas. In assembling the matched data for one’s own and parental education from the CHIP data sets, we encountered several data issues. Most of these are minor and discussed in the Appendix to this chapter, but two specific data issues deserve mention here. First, the urban sample contains individuals who originated in and received schooling in rural China. These individuals include the members of the rural population who were most successful in school. Indeed, education has been a path out of the countryside, because rural youth who gain entrance to university are eligible for nonagricultural hukou. In order to avoid the bias that would arise if we excluded this group from the rural sample, we reclassify as rural those urban residents who received their primary- and middle-school educations in rural areas. The specifics of the reclassification are explained in the Appendix to this chapter. Second, the distribution of individuals in the 2007 CHIP data set between urban and rural areas and by age (i.e., the proportions of people born in different years) is not representative. We correct for this in our analysis by using weights that reflect the shares of the urban and rural populations, and of individuals born in different years, from the NBS 1 percent population sample survey conducted in 2005. Again, details are provided in the Appendix to this chapter. Note that these weights differ from those used in other chapters in this volume. Table 4.1 shows the unweighted and weighted summary statistics for the matched sample of individuals and parents used in our analysis. The sample spans a long historical period. The oldest individuals were born in 1930, and the youngest in 1984 (as discussed earlier, the sample is restricted to individuals born in or before 1984). The oldest parents of individuals in the sample were born in the 1860s. The mean educational attainment in the sample is 8.7 years. After weighting to adjust the birth year and urban/rural shares to match those in the population, the mean falls to 7.3. Education levels are lower in rural than in urban areas. The urban-rural education gap (weighted) is 3.3 years. Education levels are also lower for girls than for boys, and more so in rural than in urban areas. Average parental education (weighted) is 4.2 years, lower than one’s own education and again lower in rural areas and for females. The difference in weighted mean years of education between parental and one’s own education is a crude measure of aggregate educational mobility in China. Overall, the weighted mean of years of education increased by 3.1 years, a 74 percent increase between the two generations. That is, on average individuals have 3.1 more years of education than their parents.

Educational Inequality in China

159

Table 4.1. Descriptive statistics for matched individuals and parents in the 2007 CHIP used in the analysis Years of education Number of observations Individuals Total Rural Male Female Urban Male Female Parents Total Fathers Mothers Parents’ avg Rural Fathers Mothers Parents’ avg Urban Fathers Mothers Parents’ avg

34,292 23,779 12,382 11,397 10,513 5,306 5,207

Earliest birth year

1930 1930 1930 1930

Mean

SD

Mean, weighted

SD, weighted

8.7 7.5 8.1 6.7 11.7 12.0 11.4

3.81 3.19 2.81 3.39 3.47 3.45 3.47

7.3 6.0 6.9 5.3 9.3 9.8 8.9

3.96 3.41 3.04 3.53 3.96 3.86 3.99

33,291 32,867 34,292

1873 1863

5.1 4.0 4.6

4.11 4.03 3.73

4.6 3.7 4.2

4.11 4.01 3.73

22,638 22,106 23,779

1873 1863

4.3 2.8 3.6

3.43 3.08 2.96

3.8 2.6 3.2

3.42 3.12 2.98

9,372 9,590 10,153

1875 1864

6.9 6.6 6.8

4.93 4.67 4.34

6.0 5.6 5.8

4.79 4.58 4.26

Note: In the columns with weighted means and standard deviations (SD), weights adjust the distribution of individuals across birth years in each of the urban and rural sectors to match their population shares given by the NBS 2005 1 percent population sample survey. In rows giving average parental education, if one parent’s years of education is missing, the average is set equal to the other parent’s years of education. In rows giving fathers’ and mothers’ education separately, the missing values are not replaced so that the descriptive statistics reflect the actual observations in the data set.

The intergenerational absolute gain in education applies in both the rural and the urban areas, but it is larger in the urban areas. In the rural areas the increase between generations is 2.8 years, and in the urban areas it is 3.5 years.

A. Aggregate Educational Mobility Aggregate mobility refers to average mobility, that is, changes in mean levels of education. Some studies, such as Hertz et al. (2007), examine the effect of the average years of the parents’ education on the child’s education. Others

160

John Knight, Terry Sicular, and Yue Ximing

distinguish the effects of the father’s and mother’s education and/or the effects on the son’s and daughter’s education. The choices depend partly on the hypotheses to be tested and partly on complications such as the possibility of positive or negative interactions between the spouses’ education levels or the possibility of “marriage sorting”; that is, the educated tend to marry the educated. We begin, like Hertz et al. (2007), with a description of the aggregate educational mobility for China as a whole, and then we explore the patterns for different sectors, cohorts, and genders. 1. Intergenerational Educational Mobility: Overall, and by Urban-Rural and Birth Year It is well known that average levels of education in China have increased over time. This reflects government measures to popularize primary and middle schooling, as well as the rising private returns to education in recent years. Rising average education is evident in the 2007 CHIP data. Table 4.2a provides an unweighted cross-tabulation of one’s own versus one’s father’s level of education; Table 4.2b provides the same for one’s own versus one’s mother’s level of education.4 Each cell of the table contains the number of individuals in the sample with the levels of one’s own and one’s father’s (or mother’s) education shown in the row and column headings. The second number in each cell is the percentage of all individuals in the sample whose fathers (mothers) have the level of education shown in that row’s heading. The bottom number in each cell is the percentage of all individuals in the sample whose own level of education is shown in that column heading. Thus, for example, the first cell tells us that there are 895 individuals in the sample with one’s own education and one’s father’s education both at level 1 (no schooling). This group constitutes 9.4 percent of the 9,503 individuals in the sample whose fathers had no schooling, and 65.7 percent of the 1,363 individuals in the sample who themselves had no schooling. From the far right columns in the tables, we can see that the most common educational level of fathers is primary school (39 percent) and of mothers no schooling (41 percent); from the bottom rows we can see that the most common level of one’s own education is junior middle school (40 percent). In only 8 percent of the cases, the child’s level of education is lower than the father’s; in 27 percent, it is the same; and in 66 percent, it is higher. For mothers, these proportions are 3 percent, 19 percent, and 78 percent, 4

The number of observations in the two tables is slightly different owing to some missing data for the mother’s level of education.

Table 4.2a. Cross-tabulation of one’s own educational level by the educational level of the father (number of observations) Own education level Educational level of father 1 No education

2 Primary

3 Junior middle

4 Senior middle

5 Post

Total

1

2

3

4

5

895 2,762 3,565 1,701 580 (9.42) (29.06) (37.51) (17.90) (6.10) (65.66) (39.76) (26.81) (23.45) (13.08) 363 3,390 5,922 2,474 901 (2.78) (28.98) (45.38) (18.96) (6.90) (26.63) (48.80) (44.54) (34.10) (20.32) 61 603 2,863 1,701 1,211 (0.95) (9.36) (44.46) (26.42) (18.81) (4.48) (8.68) (21.53) (23.45) (27.32) 39 167 830 1,083 1,004 (1.25) (5.35) (26.58) (34.68) (32.15) (2.86) (2.40) (6.24) (14.93) (22.65) 5 25 115 296 737 (0.42) (2.12) (9.76) (25.13) (62.56) (0.37) (0.36) (0.86) (4.08) (16.63) 1,363 6,947 13,295 7,255 4,433 (4.09) (20.87) (39.93) (21.79) (13.32) (100.00) (100.00) (100.00) (100.00) (100.00)

Total 9,503 (100.00) (28.54) 13,050 (100.00) (39.20) 6,439 (100.00) (19.34) 3,123 (100.00) (9.38) 1,178 (100.00) (3.54) 33,293 (100.00) (100.00)

Table 4.2b. Cross-tabulation of one’s own educational level by the educational level of the mother (number of observations) Own education level Educational level of mother 1 No education

2 Primary

3 Junior middle

4 Senior middle

5 Postsenior middle

Total

1

2

3

4

5

1,070 3,886 5,567 2,311 703 (7.90) (28.71) (41.12) (17.07) (5.19) (79.55) (57.79) (42.81) (31.78) (15.55) 219 2,596 5,680 2,632 1,115 (1.79) (21.21) (46.40) (21.50) (9.11) (16.28) (38.61) (43.68) (36.19) (24.66) 25 164 1,344 1,409 1,214 (0.60) (3.95) (32.34) (33.90) (29.21) (1.86) (2.44) (10.34) (19.38) (26.85) 28 64 337 736 1,030 (1.28) (2.92) (15.35) (33.53) (46.92) (2.08) (0.95) (2.59) (10.12) (22.78) 3 14 75 184 459 (0.41) (1.90) (10.20) (25.03) (62.45) (0.22) (0.21) (0.58) (2.53) (10.15) 1,345 6,724 13,003 7,272 4,521 (4.09) (20.46) (39.56) (22.13) (13.76) (100.00) (100.00) (100.00) (100.00) (100.00)

Total 13,537 (100.00) (41.19) 12,242 (100.00) (37.25) 4,156 (100.00) (12.65) 2,195 (100.00) (6.68) 735 (100.00) (2.24) 32,865 (100.00) (100.00)

Note: Percentages are shown in parentheses. The numbers in Tables 4.2a and 4.2b are calculated without weights because the tabulations are done with integer values and cannot be done with fractional weights.

John Knight, Terry Sicular, and Yue Ximing

Own Years of Education 6 7 8 9 10 11 12 13 14 15

162

8382 8081 79 84 77 78 76 75 74 73 72 71 63 6970 64 62 6668 61 6567 60

4

5

59 5758 56 55 54 53 552 1 50 48 45 49 42 46 47 44 43 41 30

1

2

3

35 32 3640 37 34 3938 33 31

1

2

3

4 5 6 7 8 9 10 11 12 13 14 15 Average Years of Parents' Education

own education years

45 degree line

Figure 4.2. One’s Own Years of Education and Average Years of Education of Parents, Total Sample. Note: All figures based on the CHIP survey data are calculated using urban, rural, and birth year population weights.

respectively. Thus, two-thirds or more of the individuals in the CHIP sample had higher levels of education than their parents’. Educational mobility in China has changed over time. This can be seen in Figures 4.2 through 4.4, which show the relationship between one’s own and parental education for individuals born from 1930 to 1984. Each dot plots the average own education for individuals in that birth year against the average education of their parents. Figure 4.2 is the national sample, Figure 4.3 is the rural sample, and Figure 4.4 the urban sample. All figures use weighted data. Note that the dots for birth years prior to 1941 form a lower cluster than do those for later birth years. This discontinuity occurs

Own Years of Education 6 7 8 9 10 11 12 13 14 15

Educational Inequality in China

163

3

4

5

84 79 808382 78 81 76 77 74 75 72 7173 70 66 69 6365 68 62 6064 67 61 59 58 57 56 55 54 53 52 548 1 50 49 46 4447 45 42 41 43

1

2

32 40 38 37 363539 34 33 31

1

2

3

4 5 6 7 8 9 10 11 12 13 14 15 Average Years of Parents' Education

own education years

45 degree line

Figure 4.3. One’s Own Years of Education and Average Years of Education of Parents, Rural Sample.

because we do not have separate population shares for individuals born before 1941; thus, the weight for these early birth years is aggregated (see also the Appendix to this chapter). The figures reveal a strong positive association between parental education and one’s own education. Moreover, the association is closely linked to the year of birth: younger individuals have more educated parents and are themselves more educated. The dots lie well above the 45-degree line, reflecting that on average one’s own education is higher than that of the parents. Ignoring the earliest cohorts (for which the number of observations is small and may not be representative), we see that one’s own education is generally three to four years higher than their parents’ education. These

John Knight, Terry Sicular, and Yue Ximing

Own Years of Education 6 7 8 9 10 11 12 13 14 15

164

80 77

59 56 57 58 55

83 82

81 78 75 7479 7673 72 71 84 63 64676669 62 70 65 61 68 60

48 4250 51 54 45 53 52 47 49 46 4044 41 38 36

43

30 3539

5

37

4

32

1

2

3

34 33 31

1

2

3

4 5 6 7 8 9 10 11 12 13 14 15 Average Years of Parents' Education

own education years

45 degree line

Figure 4.4. One’s Own Years of Education and Average Years of Education of Parents, Urban Sample.

patterns are not necessarily due to a causal relationship between parental and child education, however, because they are also affected by government policies that expanded the provision of education to both parents and children over time. A distinction can be made between urban and rural China. Whereas in both cases the relationships are above the 45-degree line, the urban dots (Figure 4.4) tend to be higher than the rural dots (Figure 4.3), reflecting that the extent to which one’s own education exceeds that of the parents is larger in urban areas than in rural areas. This pattern is likely due to government policies that have provided more, and better subsidized, education to urban children than to rural children, with funding from higher tiers of

Educational Inequality in China

165

government. Rural education has been more dependent on local funds provided by households, villages, and townships: demand-side factors would thus play a larger role for this group. Differences and changes over time in the prospective returns to education might also help to explain our contrasting results for the urban and rural areas. Interestingly, the relationship between parental and one’s own education in rural areas is distinctly steeper for those born between 1941 and 1960. That is, the extent to which one’s own education exceeds that of the parents rose very rapidly for the cohorts that reached school age from the late 1940s through the 1950s and 1960s. This pattern reflects, on the one hand, the very limited access to education in rural areas prior to 1950 and, on the other hand, the emphasis that the government placed on achieving universal primary education during the early planning period. Indeed, for the pre1960 rural birth cohorts, one’s own average schooling climbs rapidly to five or six years, the average length of primary school at the time. Progress in average education beyond primary levels continues for later rural birth cohorts, but at a much slower pace, as indicated by the flatter slope. For individuals born after 1959, the slope is flatter than the 45-degree line, indicating that an additional year in the average parental education is associated with slightly less than an additional year in the average of one’s own education. In urban areas, except for early cohorts born in the 1930s, most cohorts had, on average, six or more years of education. This pattern reveals that primary education was already widespread for urban residents in all but the oldest age groups, and consequently government policies promoting universal primary education in the 1950s had less impact in urban areas than in rural areas. For urban cohorts born after 1940, the relationship between one’s own and parental education shown in Figure 4.4 is a bit flatter than the 45-degree line, indicating that an additional year of parental education is associated with slightly less than an additional year of one’s own education. As for the rural sector, such a pattern may reflect many factors and is not necessarily due to a direct causal relationship between parental education and the child’s education. 2. Intergenerational Educational Mobility: Mothers, Fathers, Sons, and Daughters Table 4.3a shows the average years of the sons’ education by the educational levels of their fathers and mothers. The columns are sorted according to the level of the fathers’ education, and the rows according to that of the

166

John Knight, Terry Sicular, and Yue Ximing

Table 4.3a. Average years of education of son by levels of father’s and mother’s education Education level of mother 1 2 3 4 5 Total

Education level of father 1

2

3

4

5

Total

6.15 7.91 9.69 9.73 10.58 6.48

7.30 7.69 9.64 10.16 11.69 7.75

8.47 8.96 10.16 11.16 12.84 9.50

8.38 9.88 10.94 10.82 12.31 10.40

11.43 10.94 11.90 13.58 12.34 12.28

6.68 8.10 10.27 11.07 12.19 7.95

Table 4.3b. Average years of education of daughter by levels of father’s and mother’s education Education level of mother 1 2 3 4 5 Total

Education level of father 1

2

3

4

5

Total

4.39 7.12 8.95 8.59 10.63 4.84

5.69 6.52 8.62 9.57 8.67 6.51

7.20 7.94 9.43 10.41 11.62 8.57

7.42 9.34 10.62 9.99 10.79 9.68

8.81 9.44 11.42 12.53 12.98 11.44

4.94 6.97 9.51 10.20 11.32 6.53

Note: The education levels of mothers and fathers given in the row and column headings are: 1 = no schooling, 2 = primary school, 3 = junior middle school, 4 = senior middle school, and 5 = postsenior middle school. These indicate the level of education attained. The cells contain the average years of education for individuals whose parents have the levels of education given by that row and column. Weights are used.

mothers’. Each cell in the table gives the average years of education for sons whose parents have the levels of education shown in the row and column headings. The cells on the diagonal give the sons’ average years of education when both parents have the same level of education. For example, when both parents are uneducated (the upper-left-hand cell, father and mother both at education level 1), the average education of the son is 6.2 years; when both parents are educated beyond senior middle school (level 5), the son’s average is 12.3 years. The effect of increasing one parent’s education is similar regardless of which parent is considered. For instance, looking across the columns of row 1 (i.e., raising the father’s education while holding the mother’s education constant at level 1), a son’s education increases by 5.3 years, and looking

Educational Inequality in China

167

down column 1 (i.e., raising the mother’s education while holding the father’s education constant at level 1), the increase is 4.4 years. The results for a daughter (Table 4.3b) are similar, although the equivalent calculations show the daughter’s education to be less sensitive to that of the father (4.4 years) and to be more sensitive to that of the mother (6.2 years). The same pattern is found in the range of the father’s education irrespective of the mother’s education (the final row) and of the mother’s education irrespective of the father’s education (the final column). Although the trend is weak, it appears that “like father, like son; like mother, like daughter” (Thomas 1994). The extent of intergenerational transmission of education may be accentuated by the phenomenon of “marriage sorting”; that is, the educated tend to marry the educated. Indeed, correlations between the levels of the mother’s and father’s educations are relatively high in the CHIP sample. The (weighted) correlation for the total sample is 0.61, but slightly lower for the rural sample (0.56) than for the urban sample (0.61). Further evidence of marriage sorting is the high share (weighted) of individuals whose parents have the same level of education – 60 percent – and the much lower share whose parents’ education differs by more than one level – 11 percent. In other words, for 89 percent of individuals, the father’s and mother’s education levels are either the same or differ by no more than one level. In the presence of positive marriage sorting, including the mothers’ and fathers’ education separately in a regression equation can cause the estimated β coefficients to be a misleading indication of intergenerational educational persistence because people with well-educated fathers are also likely to have well-educated mothers. This point is discussed further in the following.

B. Educational Mobility: A Microeconomic Analysis The transmission of education from one generation of the household to the next is of particular interest in the case of China, where the government for much of its recent history has emphasized mass education, with policies that increased access to education for children of the poor and the less well educated and, during some periods, limited access to education for children of former elites. Such policies would, in theory, reduce the intergenerational transmission of education. Do the data reveal a relatively low degree of educational transmission? In this section, we discuss micro-level estimates of educational persistence between generations, that is, the βs and ρs discussed in Section IV,

168

John Knight, Terry Sicular, and Yue Ximing

with an eye to understanding the extent to which one’s own education in China is associated with the parental education. Owing to marriage sorting, which can bias the coefficients when the mothers’ education and the fathers’ education are included separately in the regression equation, in most specifications we use the average years of the education of the mother and father as a measure of parental education. In analyses that combine all ages, the sample includes individuals from all birth years. Owing to the small number of observations for individuals born before 1940, cohortspecific parameters are estimated only for individuals born in or after 1940. 1. All Cohorts Combined Table 4.4 reports estimates from ordinary least squares (OLS) regressions predicting one’s own education as a function of the average years of the individual’s mother’s and father’s education. Some specifications include dummy variables for the urban versus rural sector and for male versus female. The first three columns give estimates for the pooled urban and rural samples, the next two columns give estimates for the rural sample only, and the last two give estimates for the urban sample only. In all cases the βs are positive and significant at the 1 percent confidence level. The simplest base equation for the sample as a whole (column 1) gives an estimate of β equal to 0.51. This estimate can be compared with the estimates for other countries. Hertz et al. (2007: 15) provide estimates of β for 42 countries, ranging from 0.40 or less for Malaysia, New Zealand, and Ukraine to 1.00 or more for Egypt, Pakistan, and Brazil. If we rank the forty-two countries from highest to lowest β, our estimate of β for China puts it below the middle, between Estonia and Denmark, and higher than the United States (at 0.46). Therefore, despite the apparently egalitarian educational policies during the Maoist era, by international standards educational persistence in China was not exceptionally low. Educational policies and outcomes in China have been substantially different in the urban and rural areas. A dummy variable denoting rural residence is included in the regressions reported in column 2. The coefficient on this dummy variable is negative and significant (−2.2). Its inclusion raises the R2 and substantially reduces the coefficient on the parents’ education to 0.41. These results suggest that the intergenerational persistence in education is related to rural-urban differences. The further addition of a dummy variable denoting that the child is male (column 3) barely affects the other coefficients, but it is positive and significant at 1.3. This indicates

Table 4.4. Regressions of one’s own education as a function of the parents’ average education, all birth cohorts combined

(1) Base specification Parents’ avg. years of education

0.509*** (101.0)

169

Rural dummy

(2) Base with rural 0.412*** (80.6) −2.230*** (−56.6)

Male dummy Constant Adj R2 Degrees of freedom

5.124*** (181.3) 0.229 34,290

6.925*** (166.0) 0.295 34,289

(3) Base with rural and gender 0.407*** (81.1) −2.235*** (−57.9) 1.303*** (36.8) 6.349*** (144.9) 0.322 34,288

Note. Standard deviations are given in parentheses. The regressions are done with weights. * p < 0.05, ** p < 0.01, and *** p < 0.001.

(4) Rural only, base

(5) Rural only with gender

0.417*** (60.2)

0.416*** (62.2)

4.678*** (153.9) 0.132 23,777

1.632*** (40.8) 3.933*** (113.7) 0.189 23,776

(6) Urban only, base

(7) Urban only, with gender

0.408*** (50.0)

0.402*** (49.6)

6.949*** (118.5) 0.192 10,511

0.757*** (10.9) 6.627*** (101.4) 0.201 10,510

170

John Knight, Terry Sicular, and Yue Ximing

Table 4.5. Regressions of men’s and women’s own education as a function of their parents’ average education, all birth cohorts combined

(1) Men, base Parents’ avg. years of education Rural dummy Constant Adj R2 Degrees of freedom

0.450*** (69.3)

6.073*** (163.6) 0.213 17,686

(2) Men, base with rural 0.361*** (53.5) −1.893*** (−36.3) 7.636*** (136.3) 0.268 17,685

(3) Women, base 0.550*** (74.0)

4.358*** (106.1) 0.248 16,602

(4) Women, base with rural 0.448*** (60.8) −2.542*** (−44.9) 6.371*** (107.5) 0.330 16,601

Standard deviations are given in parentheses. The regressions are done with weights. * p < 0.05, ** p < 0.01, and *** p < 0.001.

that gender does not govern the degree of persistence, but on average, males have more education than females. In the separate urban and rural regressions (columns 4–7), the βs are all in the range of 0.40 to 0.42, lower than that for the pooled base case (column 1) and similar to the estimates from pooled regressions that include a rural dummy variable (columns 2 and 3). Thus, separate regressions for the rural and urban samples further reinforce the conclusion that the urban-rural divide contributes to the intergenerational persistence. For rural China, the coefficient on the parents’ education in the base regression (column 4) is 0.42. This is not altered when a gender dummy variable is included (column 5). The urban coefficient is only marginally lower, at 0.41 in the base case and 0.40 when a gender dummy is included (columns 6 and 7). The coefficient on the dummy variable denoting that the child is male is substantially smaller for the urban than for the rural sample, indicating that on average the education gap between men and women in urban China is smaller than that in rural China, after controlling for the parental education. Table 4.5 reports the results from regressions that differentiate between men and women. Columns 1 and 2 give the estimates for men and columns 3 and 4 give the estimates for women. The coefficient on the parents’ education in all cases is positive and significant at the 1 percent level. For men, it has a value of 0.45, but it is reduced to 0.36 when a rural dummy variable is included; for women, it has a higher value, 0.55, and again it is reduced by the addition of a rural dummy but it is still higher than that

171

0.10

Regression and Correlation Coefficient 0.30 0.60 0.40 0.20 0.50

Educational Inequality in China

40-4

45-9

50-4

55-9

60-4 Cohort

65-9

regression coefficient

70-4

75-9

80-4

correlation coefficient

Figure 4.5. Regression Coefficients and Correlation Coefficients by Cohort, Total Sample.

for men. These estimates reveal that educational persistence is greater for daughters than it is for sons in both urban and rural areas. Thus, the children of educated parents tend to have more years of education than do the children of uneducated parents, but the effect is stronger in the case of daughters. The education of girls may be more “discretionary” than that of boys. The social norms in rural society require that a daughter transfer her allegiance to her husband’s family when she marries, whereas a son remains in the village and takes responsibility for his parents in their old age (Hannum 2005). Expenditure on a daughter’s education is thus a form of consumption good, whereas expenditure on a son’s education is more like an investment good. 2. By Birth Cohort Figures 4.5 through 4.7 show estimates of the β and ρ coefficients for each of the five-year birth cohorts. Figure 4.5 shows the estimates for the total sample, Figure 4.6 for the rural sample, and Figure 4.7 for the urban sample. As the number of individuals born in the early years is small, we do not show the estimates for cohorts with birth years in or before 1940. The

John Knight, Terry Sicular, and Yue Ximing

Regression and Correlation Coefficient

172

regression coefficient

correlation coefficient

0.10

Regression and Correlation Coefficient 0.30 0.40 0.60 0.20 0.50

Figure 4.6. Regression Coefficients and Correlation Coefficients by Cohort, Rural Sample.

40-4

45-9

50-4

55-9

60-4 Cohort

regression coefficient

65-9

70-4

75-9

80-4

correlation coefficient

Figure 4.7. Regression Coefficients and Correlation Coefficients by Cohort, Urban Sample.

Educational Inequality in China

173

estimates are derived from a simple OLS regression in which the dependent variable is years of one’s own education and the explanatory variable is the parents’ average education. The βs measure the effect of an additional year of the parents’ education on one’s own education, whereas the ρs are the correlations between the two variables. Since educational policies and outcomes have differed between rural and urban areas, in Figures 4.6 and 4.7 we begin with a discussion of the separate sectors. For the rural subsample (Figure 4.6) both β and ρ are low for the first three cohorts (1940–1944, 1945–1949, and 1950–1954), that reached school age during periods when primary schooling was expanding rapidly. For the 1955–1959 cohort, the coefficients remain low. This group would have completed primary school in the late 1960s through the early 1970s, when primary education was widespread in rural China. The βs and ρs increase for the 1960–1964/1965–1969 cohorts, who would have completed primary school in the 1970s. This is when the first generation of rural children began to progress beyond primary school to junior and senior middle school. Parental education thus appears to have played a role influencing who were the first children to continue past primary school. The coefficients take another step upward for the 1970–1974 cohort. This is the generation that completed primary school in the early 1980s, a period when the average progression rates dropped sharply. During the 1980s the importance of parental education in rural China reached its peak, with a β of 0.3. Thereafter, as new policies supported the recovery of middle and higher levels of education, the coefficients decline to about 0.2, still higher, however, than those for the cohorts born in the 1950s. The correlation coefficients follow a similar pattern. For the urban subsample the estimated coefficients are volatile (Figure 4.7). The impact of the Cultural Revolution, however, is clear: the coefficients are relatively low and declining for the three cohorts that would have reached senior middle-school age during the Cultural Revolution (1950–1954, 1955– 1959, and 1960–1964). For these Cultural Revolution cohorts, parental education had a relatively small effect on one’s own education. For the initial post–Cultural Revolution cohort (born in 1965–1969, completing senior middle school in 1977–1982), the βs and ρs increase, but then they decrease as urban middle and tertiary education were reestablished in the 1980s. Thereafter, (for cohorts born from 1970 onward) the importance of parental education increases, probably reflecting the rising costs and performance-based selectivity of postmiddle urban education. For the pooled sample of rural and urban individuals (Figure 4.5), changes in the coefficients across cohorts reflect trends in the underlying

174

John Knight, Terry Sicular, and Yue Ximing

urban and rural samples, with more weight on the more populous rural populations. Higher coefficients for the combined sample than for the separate rural and urban subsamples reflect barriers between the two sectors that heighten educational persistence: the lack of mobility from the less-educated rural sector to the more-educated urban sector increases intergenerational persistence. 3. Educational Mobility of Uneducated Households In some households parents have little or no education. To what extent can their children break out of “educational poverty”? We explore this question by means of Table 4.6, which provides estimates for three categories of individuals from “education-poor” households: those in which both parents have no education, in which one parent has no education and the other has only primary education, and in which both parents have only primary education (columns 1–3, respectively). We compare individuals from such households to those from “education-rich” households (column 4), defined as households where both parents have at least a junior middle-school education. Because the sample is divided according to the education of the parents, the regression equations do not include the parental education as an explanatory variable. The differences between these groups are captured by the differences in the estimated coefficients of the constant term and the dummy variables. The first four columns do not differentiate among birth cohorts, and the constant terms measure the average education of urban women (the omitted category for the dummy variable). The last four columns include dummy variables for birth cohorts, and the constant terms measure the average education of urban women born in 1940–1944 (the omitted categories for the dummy variables). The estimated coefficients shown in column 1 of Table 4.6 indicate the following average education levels for individuals whose parents had no education: urban women 6.0 years, urban men 7.7 years, rural women 3.7 years, and rural men 5.5 years. If one or both parents attained primary schooling (moving to columns 2 and 3), the average education rises, the rural disadvantage increases, and the male advantage decreases. Overall, a shift from no education to primary schooling for one or both parents reduces the educational poverty of the children. Column 4 shows the results for education-rich households. Children from education-rich households, especially girls, have considerably more

Table 4.6. Regression equations: One’s own education as a function of location, gender, and birth cohort for education-poor and education-rich households

Variable Rural dummy Male dummy

(1) Both parents no education

(2) One parent no education, one primary

(3) Both parents primary education

(4) Both parents junior middle school or higher

−2.230*** (−26.0) 1.778*** (22.5)

−2.519*** (−24.0) 1.450*** (15.2)

−3.015*** (−37.1) 1.224*** (18.4)

−3.002*** (−28.5) 0.564*** (6.0)

5.956*** (75.0) 0.128 8011

7.809*** (79.1) 0.152 4389

8.871*** (114.2) 0.175 7954

11.127*** (154.6) 0.140 5265

1930–1934 cohort 1935–1939 cohort 1945–1949 cohort

175

1950–1954 cohort 1955–1959 cohort 1960–1964 cohort 1965–1969 cohort 1970–1974 cohort 1975–1979 cohort 1980–1984 cohort constant Adj R2 Degrees of freedom

Standard deviations are given in parentheses. The regressions are done with weights. *p < 0.05, **p < 0.01, and ***p < 0.001.

(5) Both parents no education

(6) One parent no education, one primary

(7) Both parents primary education

(8) Both parents middle school or higher

−2.805*** (−36.4) 1.991*** (28.5) −2.176*** (−12.8) −1.425*** (−9.7) 0.455** (3.2) 1.128*** (8.3) 1.980*** (14.2) 2.870*** (19.6) 2.718*** (18.1) 3.065*** (17.1) 3.916*** (15.2) 4.173*** (10.8) 5.192*** (43.0) 0.320 8001

−3.139*** (−32.6) 1.551*** (18.2) −3.654*** (−12.2) −0.920** (−3.3) 0.838*** (3.7) 1.170*** (5.6) 1.824*** (8.8) 2.728*** (13.4) 2.601*** (13.0) 2.983*** (14.1) 2.968*** (13.4) 3.516*** (13.1) 6.360*** (34.1) 0.324 4379

−2.969*** (−41.3) 1.350*** (22.8) −1.372*** (−4.6) −1.633*** (−8.5) 0.707*** (4.0) 0.891*** (5.3) 1.801*** (11.0) 2.605*** (16.2) 2.712*** (17.4) 2.957*** (18.9) 3.431*** (20.7) 4.034*** (22.5) 6.668*** (43.6) 0.356 7944

−3.142*** (−30.6) 0.520*** (5.8) −0.535 (−0.6) 3.412*** (4.6) 1.761*** (3.6) 1.618*** (3.7) 2.790*** (6.9) 2.936*** (7.7) 2.913*** (7.8) 3.572*** (9.7) 4.195*** (11.4) 4.355*** (11.9) 7.664*** (21.5) 0.199 5255

176

John Knight, Terry Sicular, and Yue Ximing

education than do those from education-poor households. For this group, the education of rural women averages 11.1 years, 5.1 more years than that for rural women with uneducated parents. Similarly, the education of rural men from education-rich households averages 11.7 years, 6.5 more years than that for rural men with uneducated parents. For urban women the gap between education-rich and education-poor households is 2.1 years, and for urban men it is 1.0 years. The difference between the education levels of women and men is larger for individuals from education-poor households; that is, educated parents invest more equally in the education of boys and girls. Education levels have increased over time, and this influences the differences between individuals from education-rich and education-poor households. We control for historical shifts by introducing dummy variables for the five-year birth cohorts. Columns 5 through 8 correspond to columns 1 through 4, respectively, but now the regressions include dummy variables for birth cohorts. The 1940–1944 birth cohort, the first cohort to receive an education after the founding of the PRC, is the omitted category. These estimates reveal that education increased as we move from older to younger cohorts from both education-poor and education-rich households. For the most recent cohort, 1980–1984, a person from an education-poor household has 3.5 to 4.2 more years of education than does a similar person born in 1940–1944. The corresponding figure for a child from an educationrich household is 4.4 years. In other words, compared to the 1940–1944 cohort, education levels among the 1980–1984 cohort rose for both children of education-poor and education-rich households, but somewhat more for the latter. Thus, the absolute gap in years of education between the two groups widened slightly. Table 4.7 shows the difference in years of education between the education-poor (both parents with no education) and the education-rich (both parents with a junior middle-school education or higher), by cohort. The differences are calculated using the estimated coefficients in columns 5 and 8 of Table 4.6. The numbers shown in Table 4.7 are for urban women, but the pattern is the same for men and for the rural sample (see the note to the Table 4.7). In all cohorts, individuals from education-poor households have fewer years of education than those from education-rich households. The education gap is largest for earlier cohorts. The gap narrows to about 3 years for the 1950–1954 cohort, where it remains, more or less, for all ensuing cohorts.

Educational Inequality in China

177

Table 4.7. Differences in one’s own education between individuals whose parents have no education and individuals whose parents have a junior middle-school or higher education, by cohort Difference in years of education

Cohorts 1930–1934 1935–1939 1940–1945 1945–1949 1950–1954 1955–1959 1960–1964 1965–1969 1970–1974 1975–1979 1980–1984

−4.1 −7.3 −2.5 −3.8 −3.0 −3.3 −2.5 −2.7 −3.0 −2.8 −2.7

Note: Calculated from the estimates in columns 5 and 8 of Table 4.6. The numbers shown are for urban women. The gap is smaller for all cohorts by 1.5 years for men and larger for all cohorts by 0.4 years for rural individuals.

In other words, the schooling gap between individuals from educationrich and education-poor households narrowed for cohorts educated in the 1950s and early 1960s, probably reflecting a combination of educational policies and household choices. Educational conditions in later periods apparently did not further reduce this gap. It is remarkable that even the Cultural Revolution, which affected those cohorts born from the mid1950s through the 1960s, did not reduce the average education gap between individuals from education-rich and education-poor households.

C. Educational Inequality Educational persistence has implications for educational inequality. If educational persistence is high, and if the distribution of education among parents is unequal, then educational inequality will be transmitted from generation to generation. Educational policies that expand educational opportunities for children of less-educated parents in principle can reduce inequality in educational attainment. To shed light on these concerns, we

John Knight, Terry Sicular, and Yue Ximing

0.00

0.20

Gini Coefficient 0.40 0.60

0.80

1.00

178

40-4

45-9

50-4

55-9

national

60-4 Cohort

65-9

70-4

rural

75-9

80-4

urban

Figure 4.8. Gini Coefficients of Years of Education by Cohort.

provide estimates of educational inequality in China across cohorts and measure the contribution of parental education to inequality in one’s own education. 1. The Extent of Educational Inequality Rising levels of education have been associated with declining inequality in the distribution of education, at least for some measures of inequality. This can be seen in Figures 4.8 through 4.10, which show changes in inequality of years of education for nine five-year birth cohorts from 1940–1944 through 1980–1984.5 As measured by the Gini coefficient (Figure 4.8), inequality for all three groups (total, rural, and urban) fell from older to younger cohorts. The decline is greatest and ongoing in rural China, although steeper through the 1960–1964 cohort and more gradual thereafter. For urban China the Gini declines gradually from older to younger cohorts, with a distinct dip below the trend for the 1950–1954, 1955–1959, and 1960–1964 birth cohorts. 5

We do not include those individuals born before 1940 in the analysis owing to the small number of observations for the oldest cohorts.

179

0.00

0.20

Coefficient of Variation 0.40 0.60 0.80

1.00

Educational Inequality in China

40-4

45-9

50-4

55-9

national

60-4 Cohort

65-9

rural

70-4

75-9

80-4

urban

Figure 4.9. Squared Coefficients of Variation of Years of Education by Cohort.

Inequality in rural education is generally higher than inequality in urban education but the difference narrows over time. For the youngest cohorts, urban and rural levels of inequality in education are about the same. The squared coefficient of variation (CV2 ) in Figure 4.9 shows trends that are very similar to those for the Gini coefficients. By construction, both the Gini coefficient and the CV2 fall as the mean value rises, ceteris paribus. The declines in these measures of educational inequality shown in Figures 4.8 and 4.9 are associated with increases in the mean years of education in China (as shown in Figures 4.2–4.4). The standard deviation (SD) is not mean-dependent. Insofar as absolute differences in the number of years of education are the criteria for evaluation (and not their relative distances from the mean), the SD is a useful measure of inequality. As discussed later, the standard deviation of education may be relevant if we are interested in the impact of educational inequality on income inequality. As shown in Figure 4.10, by this measure inequality of education overall and in rural areas fell slightly across cohorts that entered school before the late 1970s, but it remained fairly constant for more recent cohorts. For the most recent cohort, inequality appears to decline slightly. The SD for urban

John Knight, Terry Sicular, and Yue Ximing

1

2

Standard Deviation 3 4

5

6

180

40-4

45-9

50-4

55-9

national

60-4 Cohort

65-9

rural

70-4

75-9

80-4

urban

Figure 4.10. Standard Deviation of Education Years by Cohort.

areas closely follows the national trend, except for the 1950–1954, 1955– 1959, and 1960–1964 cohorts. For these three Cultural Revolution cohorts, the SD dips below the national trend. Together, Figures 4.8, 4.9, and 4.10 imply that the decline in educational inequality across cohorts largely reflects rising mean levels of education over time, rather than a narrowing in the absolute differences in years of education. The decline in educational inequality has been greatest in rural China, and especially for cohorts born before 1965 who benefited from the expansion of rural primary education in the 1950s and early 1960s, followed by the expansion of rural senior middle-school education in the 1970s. Later cohorts entered primary school in or after the mid-1970s, and decisions about their continuing to higher levels of education took place during the post-Mao period. Educational inequality for these cohorts remained unchanged, except for those born after 1980 who may have benefited from the educational reforms of the 1990s. For urban China, Figures 4.8 through 4.10 reveal clearly the impact of the Cultural Revolution. Cohorts born in 1950–1954, 1955–1959, and 1960– 1964 reached senior middle-school age in the 1965–1979 period, at which time urban middle and postmiddle schooling was disrupted and governed by political criteria. These cohorts were also affected by the “sent-down

Educational Inequality in China

181

youth” program, which sent middle-school-age urban youth to farms and factories for “real-world education.” 2. The Contribution of Parental Education to Educational Inequality: Methodology To what extent has inequality of parental education contributed to this observed inequality in education? We answer this question using a regression-based inequality decomposition. Several methods of regressionbased inequality decomposition are available. We use the straightforward method for the Gini coefficient as outlined in Morduch and Sicular (2002). The first step in the decomposition is estimation of a regression equation. Table 4.4 contains estimates of the determinants of one’s own education for China as a whole using the regression equation e i = α + βp i + δui + γg i + εi ,

(3)

where ei is years of own schooling, pi is the parents’ average years of schooling, ui is a dummy variable that equals 1 if the individual is an urban resident, gi is a dummy variable that equals 1 if the individual is female, and εi is the residual. Table 4.4 also contains estimates of alternative specifications without the urban and gender dummy variables and for urban and rural samples separately. The second step is to use the regression results to calculate how much of the dependent variable – years of own education – is contributed by the explanatory variable of interest – in this case, parental schooling. The amount of own education contributed by parental education can be calculated for each individual in the sample, and it is equal to the estimated regression coefficient on parental education times the level of parental education p ˆ i. eˆ i = βp

(4)

The third step is to calculate the share of inequality in own education contributed by parental education. The share of inequality contributed by p parental education can be written as the weighted sum of the eˆ i s given by Equation (4). For the Gini coefficient, the share of inequality contributed by parental education SGini is   n n+1 p eˆ i i=1 i − 2 2 (5) S Gini = 2 n μ G

182

John Knight, Terry Sicular, and Yue Ximing Table 4.8. Educational inequality and the contribution of parental education

National Rural Urban

Gini of education

Gini concentration ratio of one’s own education derived from parental p education (ˆe i )

Contribution of parental education to inequality in one’s own education (%) (SGini )

0.301 0.314 0.233

0.244 0.193 0.186

19.0 13.7 20.0

Note: Inequality is measured over completed years of education, using the Gini coefficient for one’s own education and the Gini concentration ratio (pseudo Gini) for one’s own education derived from parental education. The contribution of parental education to inequality in one’s own education is calculated for the Gini using all cohorts and with weights. See the text for further explanation.

where G is the Gini coefficient for years of own schooling, n is the number of individuals in the population, μ is the mean years of own schooling, and i is each individual’s rank in the distribution of years of schooling, with individuals arranged in ascending order of years of schooling such that e 1 ≤ e 2 ≤ e 3 ≤ · · · ≤ e n .6 Using Equation (5), we calculate the contribution of parental education p to inequality of own education. In this calculation, we use estimates of eˆ i based on the regression results for Equation (3) reported in Table 4.4 that include dummy variables for urban versus rural residence and for gender. We also calculate the contribution of parental education separately for the urban and rural samples, using the regression equation that includes a dummy variable for gender. 3. The Contribution of Parental Education to Educational Inequality: Findings Table 4.8 presents the estimates of inequality in education, inequality in education associated with parental education, and the share of inequality 6

Morduch and Sicular (2002) point out that decomposition of the Gini coefficient does not satisfy the property of uniform additions, which states that if a variable that determines one’s own education is equal for all individuals, then that variable will be inequalityreducing and will have a negative contribution to overall inequality. The property of uniform additions will be relevant if parental education is relatively similar across individuals. In fact, levels of education in China have risen across the board over time, so we would expect that the uniform component of parental education will have increased across cohorts. Unfortunately, we cannot use alternative decompositions that satisfy this

Educational Inequality in China

183

in education contributed by parental education (SGini ).7 The first column shows that inequality in education is higher in rural China than it is in urban China, and national inequality in education lies between that in the two sectors. The second column reveals that the portion of one’s own education p associated with parental education (as estimated by eˆ i ) is distributed more equally than one’s own education. On balance, then, inequality in education arising from nonparental factors is more disequalizing than the transmission of parental education. This is especially true in rural China. The third column gives the share of inequality in education contributed by parental education. Parental education contributes 19 percent of educational inequality in China as a whole as well as 20 percent in urban China. The contribution of parental education in rural China is lower, 13.7 percent, reflecting the more equal distribution of parental education and the ongoing emphasis on universal access to schooling in the rural areas. These estimates indicate that although differences in parental education do play a role, in China most of the inequality in education is not associated with inequality in parental education. Figure 4.11 shows shares of inequality associated with parental education by cohort. The contribution of parental education to educational inequality increased from about 3 percent for cohorts born in the 1940s (educated in the 1950s) to 11 to 12 percent for cohorts born in the late 1970s and early 1980s (educated during the post-Mao era). The contribution increases consistently across cohorts except for the Cultural Revolution cohorts (born in 1950–1954 and 1955–1959). The overall level of inequality in education, however, declined, so this pattern indicates that parental education is contributing an increasing share of smaller values. Figure 4.12 disaggregates urban and rural China. The share of inequality contributed by parental education has generally been lower in rural China than in urban China. For rural residents born before 1960, the contribution of parental education to educational inequality is exceedingly low, only about 2 percent. This pattern is consistent with the rapid expansion of access to primary and then senior middle-school education for rural cohorts that completed their schooling in the 1950s, 1960s, and 1970s. The contribution

7

property as suggested by Morduch and Sicular (2002) because the formulae cannot be used with zero values, and some individuals in the CHIP sample have zero years of education. We carried out a decomposition for the squared coefficient of variation, which also did not satisfy the property of constant additions but nevertheless is a check on the Gini coefficient decomposition. The results were very similar to those for the Gini coefficient.

14

12

Percent

10

8

6

4

2

80 -4

-9 75

70 -4

65 -9

-4 60

-9 55

-4 50

45

40 -4

-9

0

Cohort

Figure 4.11. Contribution of Parental Education to Inequality in Years of Education by Cohort (%). 14

12

Percent

10

8

urban rural

6

4

2

-4 80

-9 75

70 -4

65 -9

60 -4

55 -9

50 -4

-9 45

40

-4

0

Cohort

Figure 4.12. Contribution of Parental Education to Inequality in Years of Education by Urban versus Rural and by Cohort (%).

Educational Inequality in China

185

of parental education rises for the next several cohorts, reaching a maximum of 10 percent for those born in 1970–1974, the group that completed education in the late 1970s through the 1980s, when progression to junior and senior middle school in the rural areas fell. For more-recent cohorts, the share of educational inequality contributed by parental education has declined markedly, reflecting new policies to increase government funding for education and to increase access to senior middle schooling. In urban China the share of inequality due to parental education is variable, reflecting political and policy shifts. For the 1940–1944 cohort that completed school in the 1950s, the share of educational inequality contributed by parental education is only 3 percent. It jumps to 12 percent for the 1945–1950 cohort that completed school in the early 1960s, and then falls to 6 to 7 percent for the Cultural Revolution generation. The contribution of parental education increases again for the 1965–1969 cohort, the first group affected by the educational reforms of the early post-Mao years, at which time selectivity was high and based on academic performance. As the reforms continued, the importance of parental education initially declined, but then increased to about 12 percent for the two most recent cohorts that completed their education in the 1990s.

VI. Conclusion The transmission of education across generations is a general phenomenon found in every society. China is no exception. Our estimates suggest that educational transmission in China is slightly below the average among a group of forty-two countries for which comparable estimates are available. The degree of educational persistence from one generation to the next has implications for the persistence of other outcomes, such as income and social status. It provides a guide to the inequality of opportunities as opposed to the inequality of income. In some countries – the United States as a possible example – high income inequality may be tolerated because of the perception that equality of opportunity is sufficient. Now that China has relatively high income inequality, there is a case for strengthening policies to reduce educational inequality. Our analyses of intergenerational educational persistence using the 2007 CHIP data yield several relevant findings. In the aggregate, levels of education have risen across generations from parents to children. This general result is not surprising. A closer analysis by cohort reveals that the gains are the largest for rural individuals born prior to 1960, reflecting the rapid

186

John Knight, Terry Sicular, and Yue Ximing

expansion of primary education in the rural areas in the first decades of the PRC. The educational policies during that period provided educational opportunities for rural children, many of whose parents had received little or no education. Accordingly, inequality in schooling declined markedly across the early birth cohorts born prior to 1965. Educational inequality for later cohorts remained relatively constant, whether measured by the Gini coefficient or by the education gap between individuals from education-poor and education-rich households. The gap in years of education for these two groups declined across the early cohorts, but remained between 2.5 and 3.0 years for those cohorts born after 1960. Decomposition of inequality reveals that for recent cohorts the contribution of parental education to inequality in education increased. But recent measures promoting universal nine years of compulsory education appear to have reversed this trend in rural China. We do not observe such a reversal for the urban sample, even though the latest cohorts in our analysis should have benefited from the rapid expansion of tertiary education since the late 1990s. Our regression analyses reveal that the intergenerational persistence of education in China is associated with the differences between the rural and urban sectors. Transmission of education from parents to children in China as a whole is not low by international standards, but when we examine the urban and rural sectors separately, the level of transmission differs notably. We conclude that a key contributor to intergenerational educational persistence in China is the urban-rural divide, which segregates the education-poor from the education-rich across generations. Finally, our analyses show that intergenerational educational transmission and mobility changed across policy periods, but not always in the ways expected. We find that in rural areas the expansion of primary schooling in the 1950s and 1960s, and of senior middle schooling in the 1970s, reduced educational persistence; that is, it increased educational mobility from one generation to the next. Collectivization in the late 1950s also contributed to this pattern. Policy changes after Mao’s death, including decollectivization, led to marked declines in progression to senior middle school. Thus, during the early reform period educational mobility declined and parental education became increasingly important to the educational inequality in rural China. It is possible that the decline in mobility was exacerbated by demographic trends, because at this time large numbers of children born during the post–Great Leap baby boom reached

Educational Inequality in China

187

school age, creating competition for available spaces. In the mid-1990s rising returns to education and new government policies supporting universal nine years of compulsory education appear to have reversed these trends. In urban areas we find that changes in politics and policies caused substantial variability in educational mobility over time. Not surprisingly, the Cultural Revolution is associated with reduced educational inequality and increased educational mobility. Such is also the case for the periods of the First Five-Year Plan and the Great Leap Forward. The early 1960s are characterized by notably less educational mobility. Educational persistence increased during the post-Mao era, especially for the generations educated since the late 1980s. Findings such as these open up new areas for exploration and hypothesis testing. Studies of schooling in China and elsewhere have found that the mother’s schooling is an important determinant of the educational investment in daughters. This pattern is present in our estimates of intergenerational educational persistence. An additional year of the mother’s education has a relatively large effect for women, and this is true in both the rural and urban sectors. The education of a son appears to be more sensitive to that of the father than to that of the mother. It is possible, therefore, that parental role models are gender specific. We have provided two measures of the degree of intergenerational educational transfer, the regression coefficient and the (partial) correlation coefficient in equations predicting the child’s education from the parents’ education. Which measure is preferable? This requires a normative judgment. Is our interest in educational differences (in years) or educational differences in relation to the mean level of education in society? Note that an additional year of education has a constant proportionate effect on income in the (conventional) semi-log income function. Thus, the regression coefficient may be the more relevant measure if our ultimate concern is the distribution of income, and the correlation coefficient may be the more relevant if we are focused on the distribution of education. For policy purposes, it is important to measure the extent to which the relationship from the parents’ education to the child’s education is causal. If it is merely an association, for instance, if the parents’ education serves as a proxy for genetic or socially acquired family “ability,” or for environmental factors such as location, the policy implications are likely to be different and may well be weak. If it is a causal relationship, it represents yet another positive externality that stems from educational expenditures:

188

John Knight, Terry Sicular, and Yue Ximing

this effect is unlikely to be taken into account by decision-making households. Given that our primary concern is inequality, we have not attempted to measure the causal effect of the parents’ education. Our data do not permit an analysis of first-generation twins or second-generation adoptees, each of whom might be capable of eliminating the effects of genetic transmission. The challenge therefore is to find good instruments and variables that are well correlated with the parents’ education but do not have a direct effect on the child’s education. This might be found in the timing and location of policies on compulsory education. Since the 1980s, income inequality in China has increased substantially. Income inequality is typically associated with unequal investments in schooling – poorer households invest less than richer households in the education of their children. Moreover, studies have found that rising returns to education have also contributed to recent trends in inequality (Gustafsson, Li, and Sicular 2008; Knight et al. 2009, 2010), such that, ceteris paribus, better-educated parents have more income and thus invest more in their children’s education. We would therefore expect to find an increase in the importance of parental education and reduced educational mobility for cohorts educated in recent years. Such is indeed the case for our urban sample. For the rural sample, however, after an initial decline, educational mobility appears to have recovered. Our findings for rural China suggest that recent policies supporting universal nine years of compulsory education have been effective. More generally, there is a case, based on equity, for policies to achieve greater equality of educational opportunity, irrespective of whether there is a causal effect of the parents’ education on the child’s education. Indeed, it may be that education policy can offer the best cure for the socially and economically acquired household characteristics that create inequality in its educational and other forms among the next generation. An obvious policy would be to ensure greater equality of educational opportunities across localities, in particular, across the rural-urban divide, but also across cities, counties, and villages, because different local income levels produce differences in the quantity and quality of the provision of education. It is also likely that many poor households suffer from credit constraints (Knight et al. 2009). Credit constraints can be addressed by ensuring that school funding relies less on fees paid by households and local communities, and more on central and provincial funding. This indeed is an important theme in recent policy reforms promoting universal nine-year compulsory education in rural China.

Educational Inequality in China

189

APPENDIX: DATA ISSUES In this Appendix we describe several data issues and our treatment of them. First, some problems arise in identifying the parents of the head of the household and of the spouse of the head of the household in urban households. In the 2007 CHIP data set, information on the education of the parents of the head of the household and of the head’s spouse is collected for parents who reside in the households and for parents who are not present in the households (including those who are deceased). For all current household members, including parents who reside in the household, the relationship to the head is asked as part of the basic information collected on current household members. In the urban questionnaire this information groups the parents of the head of the household and of the spouse of the head of the household together and does not distinguish whether they are parents of the head of the household or of the spouse of the head of the household. A separate section of the questionnaire asks questions about the education of parents who are not present in the household; this section does distinguish between the parents of the head of the household and the parents of the spouse of the head of the household, so the problem exists only for the parents of the head of the household and the parents of the spouse of the head of the household who reside in urban households. In some cases, additional information in the survey allows us to identify whether the parents are the parents of the head of the household or of the spouse of the head of the household. For the remaining cases, we do not have matched information for the head (or spouse) and his or her parents. We drop these observations from the analysis. The number of these observations is small because there are relatively few multi-generation families in urban China. Of the 5,000 urban households in the CHIP sample, only 401 (8.0 percent) report parents of the head of the household and of the spouse of the head of the household residing in the household. These households contain 397 heads and 335 spouses (some households report no head, and others report no spouse). After using other available information in the survey, we are left with 147 household heads for whom we have no information about their fathers’ education, and 145 for whom we have no information about their mothers’ education. The corresponding figures for the spouses are 109 and 119. Thus, information on one or both parents’ education is missing for fewer than 3 percent of the heads and spouses in the CHIP urban sample. Given these relatively small numbers, we believe that dropping these observations will not substantially affect our results.

190

John Knight, Terry Sicular, and Yue Ximing

Second, the CHIP urban survey contains individuals who originated in rural areas and thus who were educated in rural areas. These individuals can be identified using variables in the data set that identify individuals who have changed from agricultural to nonagricultural hukou and, if so, the year when the change took place. Using this information together with the year of birth and years of education, we reclassify as rural any individual who would have completed school prior to the year when he or she converted to a nonagricultural hukou. Individuals who attended university are reclassified as rural if they would have completed senior middle school prior to the year when they converted to a nonagricultural hukou.8 In this way, we reclassified as rural 1,455 individuals in the urban sample. This group of reclassified individuals makes up 11.84 percent of the urban sample and 6.12 percent of the rural sample (here the urban and rural samples sizes refer to the sample sizes after reclassification). Third, the 2007 CHIP sample does not reflect the composition of the underlying population in terms of its sectoral and age distributions. We reweight the CHIP sample using the shares of China’s population born in each year in each of the rural and urban sectors from the 2005 NBS 1 percent sample population survey. Unless otherwise stated, all estimates, tables, and figures are calculated using these weights. For birth years prior to 1940, the NBS only gives a single, aggregated population share. Our sample contains some individuals born before 1940. For these individuals we therefore use weights based on the aggregated population share of all birth years prior to 1940. Note that our weights do not reflect provincial or regional (eastern, central, western, and large municipality) population shares. In this regard, and in our use of age-based weights, our weights differ from those used in the other chapters in this volume. The reason for this is that we have reclassified as rural those urban residents who received schooling in rural areas, and we only know whether their place of schooling was urban or rural, not their region or province of origin. 8

The standard age for beginning primary school in rural China is seven. We add seven-plus years of schooling (up to a maximum of twelve years of schooling) to each individual’s birth year to calculate the year when the individual completed pre-university schooling. If this year is earlier than the year of the hukou conversion, we reclassify the individual as rural. We also tried a second, simpler approach, which was to reclassify all individuals who changed from agricultural to nonagricultural hukou, regardless of whether or not they completed school prior to the conversion. The two approaches provide very similar results.

Educational Inequality in China

191

Table 4A.1. Educational levels used in the analysis Code 1 2 3 4 5

Educational level Illiterate and semi-illiterate Elementary school Junior middle school Senior middle school, inclusive of middle-level professional, technical, or vocational school College and above

Note: Educational levels indicate attainment of that level of education.

Table 4A.2. Conversion of educational levels in the rural questionnaires to codes and years of education used in the analysis Converted to Code in rural questionnaire 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Educational level in rural questionnaire Never schooled Graduated from five-year primary school Attended but did not graduate from five-year primary school Graduated from six-year primary school Attended but did not graduate from six-year primary school Graduated from two-year junior middle school Attended but did not graduate from two-year junior middle school Graduated from three-year junior middle school Attended but did not graduate from three-year junior middle school Graduated from two-year senior middle school Attended but did not graduate from two-year senior middle school Graduated from three-year senior middle school Attended but did not graduate from three-year senior middle school Graduated from vocational senior middle school (zhiye gaozhong) Attended but did not graduate from vocational senior middle school Graduated from senior middle technical school/ junior middle technical school (gaozhong zhongji xiao zhongzhuan)

Level of Years of education education 1 2 2

0 5 3

2 2

6 4

3 3

7 6

3 3

9 8

4 4

10 9

4 4

12 11

4

12

4

11

4

12

(continued)

192

John Knight, Terry Sicular, and Yue Ximing Table 4A.2 (continued) Converted to

Code in rural questionnaire 16 17 18 19 20 21 22 23 24 25 26 27 28

Level of Years of education education

Educational level in rural questionnaire Attended but did not graduate from senior middle technical school Graduated from specialized senior middle school (zhongzhuan) Attended but did not graduate from specialized senior middle school Graduated from junior/specialized college (da zhuan) Attended but did not graduate from junior/specialized college Graduated from TV/correspondence/distance university (dianda/hanshou/yuancheng jiaoyu) Attended but did not graduate from TV/correspondence/distance university Graduated from university Attended but did not graduate from university Graduated from a master’s degree program Attended but did not graduate from a master’s degree program Graduated from a PhD program Attended but did not graduate from a PhD program

4

11

4

12

4

11

5 5

14 13

5

14

5

13

5 5 5 5

16 15 18 17

5 5

21 20

Table 4A.3. Conversion of educational levels in the urban questionnaire to codes and years of education used in the analysis Converted to Code in urban questionnaire 1 2 3 4 5 6 7 8 9

Educational level in urban questionnaire

Level of education

Years of education

Never attended school Literacy class (saomang ban) Primary school Junior middle school Senior middle school Specialized middle school (zhongzhuan) Junior/specialized college (daxue zhuanke) University Graduate school

1 1 2 3 4 4 5 5 5

0 2 6 9 12 12 14 16 19

Educational Inequality in China

193

References Behrman, J.R., A.D. Foster, M. Rosenzweig, and P. Vashihtha (1999), “Womens’ Schooling, Home Teaching and Economic Growth,” Journal of Political Economy, 107(4), 682–714. Behrman, J.R. and M. R. Rosenzweig (2002), “Does Increasing Women’s Schooling Raise the Schooling of the Next Generation?” American Economic Review, 92(1), 323– 334. Binder, M. and C. Woodruff (2002), “Inequality and Intergenerational Mobility in Schooling: The Case of Mexico,” Economic Development and Cultural Change, 50(2), 249–267. Bj¨orkland, A., M. Lindahl, and E. Plug (2006), “The Origins of Intergenerational Associations: Lessons from Swedish Adoption Data,” Quarterly Journal of Economics, 121(3), 999–1028. Black, S. E., P. J. Devereux, and K.G. Salvanes (2005), “Why the Apple Doesn’t Fall Far: Understanding Intergenerational Transmission of Human Capital,” American Economic Review, 95(1), 437–449. Bowles, S. (1972), “Schooling and Inequality from Generation to Generation,” Journal of Political Economy, 80(3), Part 2, S219–251. Cai, F., A. Park, and Y. Zhao (2008), “The Chinese Labor Market in the Reform Era,” in L. Brandt and T. G. Rawski, eds., China’s Great Economic Transformation, 167–214, New York: Cambridge University Press. Central Education and Scientific Research Institute (1983), Zhonghua renmin gongheguo jiaoyu da shiji, 1949–82 (Major Events in Education in the People’s Republic of China, 1949–82), Beijing: Jiaoyu kexue chubanshe. China Education Almanac Editorial Department (1984), Zhongguo jiaoyu nianjian, 1949–1981 (China Education Almanac, 1949–1981), Beijing: Zhongguo baike quanshu chubanshe. Couch, K.A. and T.A. Dunn (1997), “Intergenerational Correlations in Labor Market Status: A Comparison of the United States and Germany,” Journal of Human Resources, 32(1), 210–232. de Brauw, A., J. Huang, S. Rozelle, L. Zhang, and Y. Zhang (2002), “The Evolution of China’s Rural Labor Markets during the Reforms,” Journal of Comparative Economics, 30(2), 329–353. de Brauw, A. and S. Rozelle (2007), “Returns to Education in Rural China,” in E. Hannum and A. Park, eds., Education and Reform in China, 207–223, New York: Routledge. Dollar, D. (2007), “Poverty, Inequality and Social Disparities During China’s Economic Reform,” World Bank Policy Research Working Paper No. 4253, The World Bank, Washington, DC. Fleisher, B. M. and X. Wang (2005), “Returns to Schooling in China under Planning and Reform,” Journal of Comparative Economics, 33(2), 265–277. Fock, A. and C. Wong (2008), “Financing Rural Development for a Harmonious Society in China: Recent Reforms in Public Finance and Their Prospects,” World Bank Policy Research Working Paper No. 4693, The World Bank, Washington, DC. Gustafsson, B., S. Li, and T. Sicular (2008), “Inequality and Public Policy in China: Issues and Trends,” in B.A. Gustafsson, S. Li, and T. Sicular, eds., Inequality and Public Policy in China, 1–34, New York: Cambridge University Press.

194

John Knight, Terry Sicular, and Yue Ximing

Hannum, E. (1999), “Political Change and the Urban-Rural Gap in Basic Education in China, 1949–1990,” Comparative Education Review, 43(2), 193–211. Hannum, E. (2005), “Market Transition, Educational Disparities, and Family Strategies in Rural China: New Evidence on Gender Stratification and Development,” Demography, 42(2), 275–299. Hannum, E., J. Behrman, M. Wang, and J. Liu (2008), “Education in the Reform Era,” in L. Brandt and T. G. Rawski, eds., China’s Great Economic Transformation, 215–249, New York: Cambridge University Press. Hannum, E., A. Park, and K. M. Cheng (2007), “Introduction: Market Reforms and Educational Opportunity in China,” in E. Hannum and A. Park, eds., Education and Reform in China, 1–23, New York: Routledge. Hertz, T., T. Jayasundera, P. Paraino, S. Selcuk, N. Smith, and A. Verashchagina (2007), “The Inheritance of Educational Inequality: International Comparisons and Fifty-year Trends,” B.E. Journal of Economic Analysis and Policy, 7(2), 1–45. Knight, J. and S. Li (1993), “The Determinants of Educational Attainment in China,” in K. Griffin and R. Zhao, eds., The Distribution of Income in China, 285–330, Basingstoke: Macmillan. Knight, J., S. Li, and Q. Deng (2009), “Education and the Poverty Trap in Rural China: Setting the Trap,” Oxford Development Studies, 37(4), 311–332. Knight, J., S. Li, and Q. Deng (2010), “Education and the Poverty Trap in Rural China: Closing the Trap,” Oxford Development Studies, 38(1), 1–24. Knight, J.B. and R.H. Sabot (1990), Education, Productivity and Inequality: The East African Natural Experiment, New York: Oxford University Press for the World Bank. Knight, J. and L. Song (1993), “Why Urban Wages Differ in China,” in K. Griffin and R. Zhao, eds., The Distribution of Income in China, 216–284, Basingstoke: Macmillan. Knight, J. and L. Song (2008), “China’s Emerging Urban Wage Structure, 1995–2002,” in B.A. Gustafsson, S. Li, and T. Sicular, eds., Inequality and Public Policy in China, 221–242, New York: Cambridge University Press. Li, W., A. Park, and S. Wang (2007), “School Equity in Rural China,” in E. Hannum and A. Park, eds., Education and Reform in China, 27–43, New York: Routledge. Lillard, L.A. and R.J. Willis (1994), “Intergenerational Educational Mobility: Effects of Family and State in Malaysia,” Journal of Human Resources, 29(4), 1126– 1166. Lochner, L. (2008), “Intergenerational Transmission,” in S.N. Durlauf and L.E. Blume, eds., The New Palgrave Dictionary of Economics, 2nd ed., New York: PalgraveMacmillan. L¨ofstedt, J. I. (1980), Chinese Educational Policy: Changes and Contradictions, 1949–79, Stockholm: Almqvist & Wiksell International. Ministry of Education, Department of Development and Planning (2008), Zhongguo jiaoyu shiye tongji nianjian 2007 (Educational Statistics Yearbook of China 2007), Beijing: Renmin jiaoyu chubanshe. Ministry of Education, Department of Planning (1984), Zhongguo jiaoyu chengjiu tongji ziliao 1949–1983 (Achievement of Education in China: Statistics, 1949–1983), Beijing: Renmin jiaoyu chubanshe.

Educational Inequality in China

195

Ministry of Education, Department of Planning (1991), Zhongguo jiaoyu chengjiu: Tongji ziliao 1986–1990 (Achievement of Education in China: Statistics, 1986–1990), Beijing: Renmin jiaoyu chubanshe. Morduch, J. and T. Sicular (2002), “Rethinking Inequality Decomposition, with Evidence from Rural China,” Economic Journal, 112(476), 93–106. National Bureau of Statistics (NBS) (various years), Zhongguo tongji nianjian (China Statistical Yearbook), Beijing: Zhongguo tongji chubanshe. National Committee of Inquiry into Higher Education (1997), “Higher Education in the Learning Society (the Dearing Report),” London. Niu, X. (1992), Policy, Education, and Inequalities: In Communist China since 1949, Lanham, MD: University Press of America. Oreopoulos, P., M.E. Page, and A.H. Stevens (2006), “The Intergenerational Effects of Compulsory Schooling,” Journal of Labor Economics, 24(4), 729–760. Pepper, S. (1990), China’s Education Reform in the 1980s: Policies, Issues, and Historical Perspectives, Berkeley: Institute of East Asian Studies, University of California. Riskin, C. (1987), China’s Political Economy: The Quest for Development since 1949, New York: Oxford University Press. Sato, H. and S. Li (2007), “Class Origin, Family Culture, and Intergenerational Correlation of Education in Rural China,” IZA Discussion Paper No. 2642, Institute for the Study of Labor, Bonn, Germany. Thøgersen, S. (1990), Secondary Education in China after Mao: Reform and Social Conflict, Aarhus, Denmark: Aarhus University Press. Thomas, D. (1994), “Like Father, Like Son; Like Mother, Like Daughter,” Journal of Human Resources, 29(4), 950–988. Thomas, D. (1996), “Education across Generations in South Africa,” American Economic Review, 86(2), 330–334. Tsang, M. C. (2000), “Education and National Development in China since 1949: Oscillating Policies and Enduring Dilemmas,” in C.M. Lau and J. Shen, eds., China Review 2000, 579–618, Hong Kong: The Chinese University Press. Tsang, M. C. (2001), “Intergovernmental Grants and the Financing of Compulsory Education in China,” unpublished ms, Teachers College, Columbia University, New York, http://www.tc.columbia.edu/centers/coce/pdf files/a1.pdf/ Accessed July 11, 2011. Walker, K. R. (1966), “Collectivisation in Retrospect: The ‘Socialist High Tide’ of Autumn 1955–Spring 1956,” China Quarterly, no. 26, 1–43. Wang, D. (2003), “China’s Rural Compulsory Education: Current Situation, Problems and Policy Alternatives,” Institute of Population and Labor Economics, Chinese Academy of Social Sciences, Working Paper Series No. 36, Beijing. Wong, C. P.W. and R. M. Bird (2008), “China’s Fiscal System: A Work in Progress,” in L. Brandt and T.G. Rawski, eds., China’s Great Economic Transformation, 429–466, New York: Cambridge University Press. World Bank, Rural Development, Natural Resources and Environmental Sustainable Development Department, East Asia and Pacific Region (2007), “China: Improving Rural Public Finance for the Harmonious Society,” World Bank Report No. 41579-CN, The World Bank, Washington, DC.

196

John Knight, Terry Sicular, and Yue Ximing

Xing, C. (2007), “Xuezhi gaige yu jiaoyu huibaol¨u” (Reform of the Primary Schooling System and Returns to Education), unpublished manuscript, Beijing Normal University, Beijing. Zhang J. and Y. Zhao (2007), “Rising Returns to Schooling in Urban China,” in E. Hannum and A. Park, eds., Education and Reform in China, 248–259, New York: Routledge. Zhang, L., J. Huang, and S. Rozelle (2002), “Employment, Emerging Labor Markets, and the Role of Education in Rural China,” China Economic Review, 13(2–3), 313–328. Zhao, Y. (1997), “Labor Migration and Returns to Rural Education in China,” American Journal of Agricultural Economics, 79(4), 1278–1287.

FIVE

Inequality and Poverty in Rural China Luo Chuliang and Terry Sicular

I. Introduction The rural sector has featured prominently in China’s policy agenda since the change in leadership in the early 2000s. For each of the seven consecutive years from 2004 through 2010, the State Council’s No. 1 Central Document addressed rural policies. As the first policy communiqu´e of the year, these documents are indicative of the high priority placed on the rural sector (Xinhua News Agency 2008, 2010), and they have introduced an array of policy initiatives, such as the “New Socialist Countryside” program. Key rural policies during this period have included the elimination of agricultural taxes and fees, government subsidies for agricultural production, public investments in rural infrastructure, extension of the minimum living standard guarantee (dibao) program to rural areas, the rural cooperative medical system, and the expansion of universal, free nine-year public education (Chen 2009, 2010; Lin and Wong 2012). In addition, the government has implemented measures to ease restrictions on rural-urban mobility and to improve work and living conditions for migrants (Cai, Du, and Wang 2009). The recent emphasis on the rural sector reflects two national concerns: the widening gap between urban and rural incomes and the slow growth of agricultural production. The growing gap between urban and rural incomes has been noted in numerous studies and has been a major factor contributing to the secular increase in income inequality (Gustafsson, Li, and Sicular 2008; Sicular et al. 2007; Chapter 2 in this volume). The welfare of the

We thank Li Shi for comments and suggestions. Support from the Ontario Research Foundation is gratefully acknowledged.

197

198

Luo Chuliang and Terry Sicular

rural population has lagged behind that of the urban population, not only in terms of income but also in other areas, such as health, education, and social support (Whyte 2010). Agricultural production has experienced ups and downs, with implications for both the supply of food and rural incomes. Trends in grain output, of particular concern to the central government, are indicative. After reaching peak levels in 1998–1999, China’s grain production fell markedly and in 2003 was at its lowest level in more than a decade. This drop was associated with declining prices for key farm products, to some extent a byproduct of the trade liberalization leading up to and following China’s accession to the World Trade Organization (WTO) in 2001 (Huang et al. 2007). These price trends affected growth in rural household earnings from agriculture, a major source of income for rural households (Gale, Lohmar, and Tuan 2005; Khan and Riskin 2008). In this chapter we document changes in rural household incomes and inequality from 2002 to 2007, a period of renewed emphasis on rural policy. We use data from the 2002 and 2007 CHIP rural household surveys, and make comparisons to findings reported in studies based on previous rounds of the CHIP rural survey. We begin by examining changes in the level of household per capita income. As noted in other chapters in this volume, between 2002 and 2007 China’s urban-rural income gap widened. Was this expansion of the urbanrural gap the result of stagnation in rural household incomes? Our answer is no. We find that rural incomes grew substantially, and at a more rapid pace than during the preceding period. Moreover, this income growth was relatively balanced, reflecting increases in income from both agriculture and off-farm employment and other sources. Therefore, the widening of the urban-rural income gap between 2002 and 2007 was the result of even more rapid growth in urban incomes, rather than the result of stagnation in rural incomes. Second, we analyze changes in inequality within the rural areas. China’s countryside is large and diverse, characterized by differing economic conditions and opportunities. Some policies have targeted poorer rural areas and groups; others have not. We find that, on balance, rural inequality increased only slightly during this period. The lack of deterioration in inequality reflects the fact that rural income growth during this period was widely shared. Third, we analyze changes in rural poverty. As measured against an absolute poverty line, the poverty rate and poverty gap declined substantially. We find, however, that for the remaining poor, extreme poverty has increased.

Inequality and Poverty in Rural China

199

In addition, we find no improvement in relative poverty, as measured in relation to median income rather than an absolute poverty line. How do these trends in income and poverty relate to recent rural policies? Although a full analysis of this question is beyond the scope of this chapter, we use available information in the CHIP data sets to investigate the impact of several key policies. In the sections that follow, we examine the distribution of income from migrant employment, the effects of reductions in government taxes and fees, and the relationship between poverty and participation in the dibao program.

II. Data and Methods In our analysis we use the 2002 and 2007 CHIP rural household survey data. As discussed in Chapter 1 and the appendices to this volume, the rural survey samples include household members with rural hukou who are short-term migrants and longer-term migrants who maintain close ties with their rural households of origin. In 2007 the CHIP rural survey covered 16 provinces, 13,000 households, and 51,847 individuals. The 2002 CHIP rural survey covered fewer households and individuals, but more provinces – 9,200 households and 37,969 individuals from 22 provinces. Fifteen provinces (Beijing, Hebei, Shanxi, Liaoning, Jiangsu, Zhejiang, Anhui, Henan, Hubei, Hunan, Guangdong, Chongqing, Sichuan, Yunnan, and Gansu) were covered in both years, seven provinces (Jilin, Jiangxi, Shandong, Guangxi, Guizhou, Shaanxi, and Xinjiang) only in 2002, and one (Fujian) only in 2007. Some incomplete and missing data slightly reduce the number of observations used in our analyses. In our calculations we include all provinces for both years. Except where noted otherwise, all calculations are done using two-level regional and provincial weights; consequently, the results should be nationally representative for both years despite the coverage of different provinces.1 We note that the weighting approach used here improves upon that used in earlier analyses of the CHIP rural data. For growth across the two years, we report results calculated in constant prices using the national rural consumer price index compiled by the National Bureau of Statistics (NBS). In some calculations we also adjust for differences in the cost of living among the provinces, using the price indices from Brandt and Holz (2006) and extended to 2007 using the annual provincial rural consumer price indices from the NBS. We refer to estimates 1

See the Appendix to this volume for additional explanations of the weights.

200

Luo Chuliang and Terry Sicular Table 5.1. Rural household per capita income, 2002 and 2007

CHIP Rural Survey Data NBS income definition CHIP income definition Published NBS Statistics

2002 (yuan)

2007 (yuan)

Average annual growth (%, constant prices)

2,590 2,771 2,476

4,221 4,617 4,140

6.96 7.44 7.51

Notes: All mean incomes are in current prices. CHIP incomes are calculated with weights, and average annual growth is calculated using constant prices deflated using the NBS rural consumer price index. The published NBS income statistics and rural consumer price index are from NBS (2008).

adjusted for differences in provincial costs of living as purchasing power parity (PPP) estimates. As mentioned in other chapters, two income definitions are commonly used in analyses of China’s income distribution. One is the NBS measure of household per capita net income. The other is CHIP income, a broader measure of household per capita net income that is used in the earlier CHIP studies (Gustafsson et al. 2008; Khan and Riskin 1998; Khan et al. 1992). The main difference between these two measures is that the latter includes imputed rents on owner-occupied housing and, compared to the former, has a fuller accounting of income subsidies. In the context of the rural sector where households have received few subsidies, the major difference between these two income measures is imputed rent. Our measure of income in this chapter is CHIP income, which for rural households is equal to NBS income plus the imputed rent; below we refer to this as “CHIP income.” Our estimates of imputed rents are taken from Chapter 3 of this volume. For purposes of comparison, we present some results for both the NBS and CHIP measures of income.

III. Trends in Rural Incomes Table 5.1 shows the mean values of income per capita calculated using the CHIP rural survey data. Overall, these income levels are consistent with the published NBS statistics on rural incomes based on its annual rural household surveys. If we use the NBS definition of income, in both years the weighted mean incomes calculated using the CHIP rural survey data are higher than, but within 5 percent of, the published NBS figures. The CHIP data also yield growth rates in real per capita income, measured using the NBS definition of income, that are lower than but close to those

Inequality and Poverty in Rural China

201

Table 5.2. Rural household per capita income, by source 2002

2007

Average Share of Share of annual Increment Share of income income growth (constant increment Yuan (%) Yuan (%) rate (%) 2002 yuan) (%) Wage earnings from 314 migrant employment Wage earnings from 678 local employment Net income from 1,099 agriculture Net income from 363 nonagricultural businesses Net transfer income 117 Asset income 19 Imputed rent on 181 owner-occupied housing TOTAL 2,771

11.3

816

17.7

17.4

387

32.3

24.5

929

20.1

3.3

121

10.1

39.7

1,686

36.5

5.7

349

29.2

13.1

471

10.2

2.2

42

3.5

4.2 0.7 6.5

197 121 397

4.3 2.6 8.6

7.7 40.9 13.5

52 85 160

4.4 7.1 13.4

100.0

4618

100.0

7.4

1195

100.0

Note: Weighted. Mean income levels for 2002 and 2007 are in current prices; income growth and income increments are in constant 2002 prices. Numbers may not match exactly due to rounding.

published by the NBS. Including imputed rents increases the level of per capita income and the rate of income growth. Hereafter, we carry out our analysis using the CHIP income definition, except where otherwise noted. The estimates in Table 5.1 show that real growth in rural incomes between 2002 and 2007 was fairly rapid, averaging 7.4 percent annually. Much of this income growth was due to increased earnings from agriculture and migrant employment. Table 5.2 shows the composition of income during the two years. By 2007 wage income, including that from both migrant and local employment, accounted for 38 percent of per capita rural household income. Wage earnings from migrant work increased very rapidly – at 17 percent per year. Wage earnings from local employment increased more slowly at 3 percent a year. In 2007 agriculture contributed 37 percent of income. Although the share of agriculture to total income declined slightly from 2002, agricultural income nevertheless showed solid growth of 6 percent per year, rebounding from slow growth of only 1.2 percent per year between 1995 and 2002

202

Luo Chuliang and Terry Sicular

(Khan and Riskin 2008: 63). Moreover, in absolute terms agriculture contributed nearly one-third of the overall income increment between 2002 and 2007 (the last two columns of Table 5.2). These positive trends in agricultural income are consistent with the proagriculture policies adopted at the time. The CHIP data do not allow us to distinguish the effects of new agricultural support policies from other factors, such as improved farm prices and technical changes, but information from other sources allows us to make a rough calculation. Lin and Wong (2012) provide data on government agricultural support subsidies, from which we calculate that direct production subsidies were equal to 11 yuan of household agricultural income per capita in 2003 and 57 yuan in 2007. These numbers suggest that receipts of farm production subsidies contributed roughly 57 yuan, or 10 percent, of the 587 yuan increase in nominal farm income between 2002 and 2007. We conclude that although government agricultural production subsidies were not trivial, they explain only a small fraction of the increase in rural household agricultural income during this period. Income from household nonagricultural businesses, transfers, and property all grew to greater or lesser extents. As a share of total income, earnings from nonagricultural businesses declined slightly, whereas asset income and imputed rents on owner-occupied housing increased. By 2007, income from these two latter sources accounted for 11 percent of total income, signaling the emergence of assets as a significant component of income in rural China. Net transfer income, which includes public transfers such as dibao and wubao support, net of taxes, as well as private transfers such as gifts and migrant remittances, increased in absolute terms, as one might expect given the new subsidy programs and the reductions in taxes and fees at the time. Still, they remained a relatively small component of total income. We note that some government programs that were adopted in the 2000s operated indirectly by reducing household outlays on education, health, and production, or by increasing net income from farming, rather than explicitly through “transfer” income.

IV. Trends in Rural Inequality Tables 5.3 and 5.4 show our estimates of rural inequality in 2002 and 2007. Table 5.3 reports estimates of the Gini coefficient calculated with and without imputed rents from owner-occupied housing. Our estimates of

Inequality and Poverty in Rural China

203

Table 5.3. Estimates of the rural Gini coefficient, 2002 and 2007 Not PPP

CHIP Rural Survey Data NBS income definition CHIP income definition Published NBS Statistics

PPP

2002

2007

% change

2002

2007

% change

0.358 0.354 0.365

0.363 0.358 0.374

1.4 1.1 2.5

0.356 0.352

0.364 0.357

2.2 1.4

Notes: The CHIP data are weighted. The PPP estimates correct for provincial differences in cost of living using the Brandt and Holz (2006) price indices updated to 2007 using the NBS provinciallevel rural consumer price indices. The NBS published Gini coefficients are based on the NBS rural household surveys and can be found in Department of Rural Surveys (2010: 46, Table 2–26).

the Gini coefficients calculated using the NBS income definition (excluding imputed rents), shown in the last row of Table 5.3, are similar to those published by the NBS. For both years these two Gini estimates differ by less than 3 percent. In both cases the Gini coefficients increased between 2002 and 2007. The increase is larger for the official NBS statistics but is still modest, that is, less than 3 percent. Including imputed rents slightly reduces inequality and slightly reduces the change in inequality. The mildly equalizing effect of imputed rents reflects their relatively equal distribution due to almost universal homeownership in rural China (see Chapter 3). Spatial differences in the cost of living have led to an overstatement of measured inequality for China as a whole (Brandt and Holz 2006; Sicular et al. 2007). We therefore present estimates of the rural Gini coefficient after adjusting for the spatial price differences. Estimates of PPP inequality are shown in the last three columns of Table 5.3. We find that the PPP adjustment has a trivial effect on the measured levels of rural inequality and that the change in the Gini coefficients between 2002 and 2007 remains modest. We conclude that cost-of-living differences within the rural sector are not important to our analysis. Consequently, hereafter we do not adjust for spatial price differences. Our preferred estimates of the Gini coefficient, calculated using the CHIP income definition, show little change in inequality over the two years: 0.354 for 2002 and 0.358 for 2007. We conclude that inequality in rural China remained low and relatively stable throughout this period. Even the highest estimates in Table 5.3 are well below 0.4, and changes in the level of inequality for all estimates between 2002 and 2007 are modest.

204

Luo Chuliang and Terry Sicular Table 5.4. Alternate measures of inequality in rural China, 2002 and 2007

Coefficient of Variation Theil index (GE(1)) Mean Log Deviation (GE(0)) Income ratio of top 20% to bottom 20% Income ratio of top 10% to bottom 10% Income ratio of top 5% to bottom 5%

2002

2007

% change

0.8039 0.2258 0.2129 6.09 10.02 15.87

0.8134 0.2260 0.2165 6.39 11.11 19.89

1.18 0.09 1.69 4.93 10.88 25.33

Note: Calculated with weights and using the CHIP income definition.

Alternate inequality indices yield similar findings (Table 5.4). The Coefficient of Variation, Theil index, and Mean Log Deviation increase only slightly between 2002 and 2007. Table 5.4 also shows estimates of the range, calculated as the ratio of the mean incomes of the richest and poorest groups in the income distribution. The range shows more change between 2002 and 2007 than the other inequality indices, and the change is greater when the cutoffs for the top and bottom income groups are more extreme. The range for the top 20 percent versus the bottom 20 percent increased 4 percent, whereas that for the top 5 percent versus the bottom 5 percent increased a marked 25 percent. In 2002 the richest 5 percent of rural households enjoyed sixteen times, and in 2007 twenty times, the per capita income of the poorest 5 percent of rural households. Thus, although inequality overall was relatively stable, the gap between the very low and very high extremes widened. An examination of income growth for each decile group in the income distribution provides more detailed information about the changes in income distribution (Figure 5.1). Except for the poorest decile group, income growth between 2002 and 2007 was in the 7 to 8 percent range. Income growth lagged, however, for the poorest decile, for which income per capita grew at a slower rate of 5 percent. To explore the contribution of the different income sources to inequality, we decompose the Gini coefficient by its source components (Adams 1999; Stark, Taylor, and Yitzhaki 1986). If total income is composed of k  components, that is, Y = k Yk , then the Gini coefficient of total income G(Y) can be expressed as the sum of the contributions Sk of each income source:   Sk = uk G (Yk )R k . (1) G (Y) = k

k

Inequality and Poverty in Rural China 8.0%

7.0%

7.7%

7.6%

7.5%

7.5%

7.7%

7.5%

7.6%

7.3%

7.5%

205

7.3%

6.5% 6.0% 5.5% 5.0%

5.1%

4.5% 4.0% 1

2

3

4

5

6

7

8

9

10

decile

Figure 5.1. Average Annual Income Growth from 2002 to 2007 for Decile Groups in the Distribution of Income. Note: Growth rates for each decile are calculated as (y 2007,p /y 2002,p )1/5 − 1, where 2002 and 2007 denote the two years, and p denotes the decile group. Growth is calculated using constant 2002 prices, with weights, and using the CHIP income definition.

Here uk = YYk is the share of source k income in total income, G(Yk ) is the Gini coefficient measured over income from source k, and Rk is the rank correlation between income from source k and total income, that is, Rk =

cov(Yk , F (Y)) , cov(Yk , F (Yk ))

(2)

where F(.) is the cumulative distribution of total household income or income from source k in the sample.2 The share of income component k in total inequality can then be written as   G (Yk )R k = uk uk c k . (3) sk = k k G (Y) In Equation (3) ck , the relative concentration coefficient, is of particular interest, because it indicates whether an income source is inequality increasing or inequality decreasing. A value of ck greater than one indicates that income from this source is inequality increasing; a value of less than one indicates that it is inequality decreasing. 2

For example, F(Y) = (f(y1 ), . . . , f(yn )), where f(yi ) equals the rank of yi divided by the number of observations n.

206

Luo Chuliang and Terry Sicular Table 5.5. Gini coefficient decomposition, by income source

Percentage of income Income source Wage earnings from migrant employment Wage earnings from local employment Net income from agriculture Net income from nonagricultural business Net transfer income Asset income Imputed rent on owner-occupied housing TOTAL

Gini relative concentration coefficient (ck )

Percentage of Gini contributed (sk × 100)

2002

2007

2002

2007

2002

2007

11.3

17.7

0.81

0.82

9.2

14.5

24.5

20.1

1.43

1.29

34.9

25.9

39.7

36.5

0.58

0.71

22.9

26.0

13.1

10.2

1.58

1.66

20.7

16.9

0.7 4.2 6.5

2.6 4.3 8.6

2.04 1.16 0.91

1.69 1.00 0.93

1.4 4.9 6.0

4.4 4.3 8.0

100.0

100.0

1.00

1.00

100.0

100.0

Note: Calculated with weights and using the CHIP income definition.

Table 5.5 provides estimates of ck (in the middle two columns) and of sk (in the last two columns). These estimates reveal how different sources of income affected overall inequality in rural China. Agriculture, with the lowest relative concentration coefficient in both years, remained the most equalizing income component. The rise in agriculture’s ck between 2002 and 2007 implies that the extent to which agriculture was equalizing declined. Incomes from migrant wages and imputed rent on owner-occupied housing were also equalizing. Net transfer income was dis-equalizing in 2002, but by 2007 it had become less so, possibly reflecting the elimination of taxes and fees as well as government transfers to poorer households. Because we cannot separate public from private transfers, and because government subsidies for agriculture enter income through their influence on net income from agriculture, changes in the distribution of net transfer income do not fully capture the effects of such policies on inequality. The remaining sources of income – income from nonagricultural household businesses and asset income – were dis-equalizing in 2002. Between 2002 and 2007 household business income became more, and asset income less, dis-equalizing. In 2007, asset income had a neutral impact.

Inequality and Poverty in Rural China

207

The last two columns in Table 5.5 show the contributions of different sources of income to overall inequality. The size of the contribution depends on both the relative concentration coefficient ck and the share of income uk . In 2007 agriculture contributed about a quarter of total inequality, a slight increase from 2002. This large contribution reflects agriculture’s substantial share of total income. Wage earnings from local employment also contributed about a quarter of overall inequality in 2007. This was a substantial drop from 2002, when local wages contributed more than one-third of the inequality. The contribution of wages from migrant employment was relatively low in both years, reflecting its fairly equal distribution. Its contribution increased substantially between 2002 and 2007; however, this was due to its increased share of household income. The combined contributions of asset income and imputed rent on owneroccupied housing grew from 10.9 percent in 2002 to 12.3 percent in 2007. Thus, the importance of income from property to rural inequality showed a modest increase and constituted a nontrivial share of overall inequality in the rural sector.

V. Changes in Rural Poverty During China’s economic transition poverty in rural China declined dramatically. According to the NBS, in 2007 the rural poverty rate was only 1.6 percent, down from 30.7 percent in 1978 (Department of Rural Surveys 2008). These trends are measured using China’s official poverty lines, which many observers believe to be low (e.g., Poverty Reduction and Economic Management Division 2009, Ravallion and Chen 2007). Using a higher poverty line yields a higher poverty rate, but it does not change the conclusion that in recent decades, rural China has witnessed substantial poverty reduction (Poverty Reduction and Economic Management Division 2009; Ravallion and Chen 2007). In view of the various poverty lines used in the literature, we present several estimates, two using absolute poverty lines and two using relative poverty lines. In all cases we use the NBS measure of income, which does not include imputed rents on owner-occupied housing. Imputed rents are excluded because the official poverty lines are set without reference to imputed rents as a cost of living; therefore, including them would artificially reduce the poverty rates. Our first absolute poverty line is the widely used international purchasing power parity (PPP) poverty threshold of PPP$1.25 per day per person,

208

Luo Chuliang and Terry Sicular Table 5.6. Poverty lines 2002

PPP$1.25 per day per person Official poverty line 0.5 × median income 0.6 × median income

2007

Amount (yuan)

Share of mean income (%)

Amount (yuan)

Share of mean income (%)

1,451 964 1,051 1,261

56.0 37.2 40.6 48.7

1,689 1,123 1,714 2,057

40.0 26.6 40.6 48.7

Note: All poverty lines are expressed in terms of income per capita. Median and mean incomes are calculated using the weighted CHIP rural sample incomes and the NBS income definition, which does not include imputed rent from owner-occupied housing.

which we convert to yuan using the recently updated PPP exchange rate of 3.46 yuan to the U.S. dollar in 2005 (Chen and Ravallion 2008). The second is the Chinese government’s official poverty line. In view of past criticisms of the official poverty line, we use the more recent, higher 2008 official poverty line of 1,196 yuan. We adjust both of these poverty lines to their 2002 and 2007 levels using the NBS rural consumer price index. Relative poverty lines are commonly applied for measurements of poverty in higher-income countries, where few households experience absolute deprivation but where individuals at the lower end of the income distribution nevertheless may be disadvantaged (Osberg 2000; Ravallion 1992). In view of China’s rapid growth over the past decades, we believe the concept of relative poverty is increasingly relevant. Following common practice in the literature, we use a relative poverty line equal to 50 percent of the median income and a second, higher relative poverty line of 60 percent of the median income. Median income is calculated using the weighted rural CHIP sample incomes for each of the two years. Table 5.6 shows our four poverty lines. Due to growth in rural incomes between 2002 and 2007, the ratio of the absolute poverty lines to the mean sample income fell. For the relative poverty lines, the ratios remained constant. Using these poverty lines, we calculate the level of poverty. Consistent with the literature, we adopt the approach developed by Foster, Greer, and Thorbecke (1984), which yields the common poverty headcount as well as estimates of the poverty gap. The Foster-Greer-Thorbecke (FGT) index can be written as   z − Yi α 1 q , (4) F G T(α) = i=1 N z

Inequality and Poverty in Rural China

209

Table 5.7. Poverty estimates 2002

PPP$1.25 per day Official poverty line 0.5 × median income 0.6 × median income

2007

Poverty headcount (%)

Poverty gap (%)

Squared poverty gap

Poverty headcount (%)

Poverty gap (%)

Squared poverty gap

27.48 11.22 13.69 20.75

8.37 2.97 3.75 5.99

3.72 1.27 1.60 2.59

13.88 5.59 14.32 21.07

4.65 2.25 4.79 6.93

5.04 7.09 5.03 5.28

Note: The poverty headcount FGT(0) measures the incidence of poverty; the poverty gap FGT(1) measures the depth of poverty; the squared poverty gap FGT(2) measures the severity of poverty (Ravallion 1994). Calculated using the poverty lines shown in Table 5.6, the weighted CHIP rural sample incomes, and the NBS income definition, which excludes imputed rent from owner-occupied housing.

where N is the size of the total population, q is the size of the poor population, z is the poverty line, and Yi is the income of individual i. This index calculates the poverty gap g i = z − Yi for each individual under the poverty line, which is then divided by the level of the poverty line and raised to the power α. The parameter α can be interpreted as the degree of poverty aversion: the larger the α, the greater the degree of poverty aversion. Conveniently, when α = 0, FGT(0) is simply the headcount ratio (the proportion of the population that is poor). FGT(1) gives the average poverty gap, which measures the average percentage income shortfall below the poverty line of the poor. FGT(2) is the squared poverty gap, which places more weight on the income shortfall of the extreme poor than that of the near-poor who are close to the poverty line. These three poverty measures capture the incidence, depth, and severity of poverty, respectively (Ravallion 1994). Table 5.7 shows estimates of these three poverty measures calculated for each of the alternative poverty lines. The level of poverty and the change in poverty between 2002 and 2007 differ depending on the choice of the poverty line. For the absolute poverty lines, the poverty headcount FGT(0) declines substantially between 2002 and 2007: for the PPP$1.25 per day poverty line, the poverty headcount drops by more than half, from 27 percent to 14 percent, and for the official poverty line, the headcount declines from 11 percent to 6 percent. For the relative poverty lines, the poverty headcount remains almost unchanged between 2002 and 2007. For example, relative to 50 percent of the median income, the poverty headcount increased slightly from 13.7

210

Luo Chuliang and Terry Sicular

to 14.3 percent. This suggests that although the income of the poor grew enough between 2002 and 2007 to raise roughly half of the poor above absolute poverty, this income growth was not sufficient to catch up with the median income. Results for the poverty gap FGT(1) also differ between the absolute and relative poverty lines. For the former, the poverty gap decreased between 2002 and 2007, and for the latter, it increased. Results for the squared poverty gap are consistent for the four poverty lines: in all cases, the severity of poverty as measured by FGT(2) increased. These findings suggest that between 2002 and 2007 the near-poor – those near the absolute poverty lines – saw income growth and escaped poverty, but the incomes of the extreme poor lagged. Consequently, the remaining poor in 2007 can be characterized by a greater degree of severe poverty. To what extent do these poverty trends reflect the results of income growth rather than redistribution between richer and poorer groups? As noted earlier, on average rural incomes grew substantially between 2002 and 2007. Did this rising tide raise the boats of the poor? Two methods commonly used to differentiate between the impact of growth as opposed to redistribution are those of Datt and Ravallion (1992) and Shorrocks (1999). We have used both methods, which yield similar results, so here we report only the results of the Shorrocks approach. The level of poverty P is determined by the poverty line z, the mean income μ, and the cumulative distribution of income as measured by the Lorenz curve L(p), which gives the share of income going to the bottom p percent of the population. Let the subscript t denote time. Then, the change in the level of poverty from time 0 to time t can be expressed as P = P (μt , L t (p ), z) − P (μ0 , L 0 (p ), z).

(5)

According to Shorrocks (1999), the change in the level of poverty can be decomposed into the growth effect (G) and the redistribution effect (R) as follows: G = 0.5 ∗ {[P (μt , L 0 (p ), z) − P (μ0 , L 0 (p ), z)] + [P (μt , L t (p ), z) − P (μ0 , L t (p ), z)]}

(6a)

R = 0.5 ∗ {[P (μt , L t (p ), z) − P (μt , L 0 (p ), z)] + [P (μ0 , L t (p ), z) − P (μ0 , L 0 (p ), z)]}.

(6b)

The growth effect G (Equation 6a) is calculated as the change in poverty that results from the observed change in mean income, holding the distribution and poverty line constant. The redistribution effect R (Equation 6b)

Inequality and Poverty in Rural China

211

Table 5.8. Decomposition of changes in poverty, 2002–2007 PPP$1.25 per day Poverty headcount

Poverty gap

−13.60

−3.72

Of which: (percentage points) Growth −14.10 −4.66 Redistribution 0.50 0.94

Change in poverty (%)

Official poverty line

Squared poverty gap

Poverty headcount

Poverty gap

Squared poverty gap

1.32

−5.64

−0.72

5.82

−1.55 2.87

−6.61 0.97

−1.80 1.07

0.61 5.21

Note: Calculated using the Shorrocks (1999) method, with weights, and using the NBS income definition (excluding imputed rental income from owner-occupied housing). The calculation uses constant prices.

is calculated as the change in poverty that results from the observed change in the distribution of income, holding the mean income and the poverty line constant. In both cases, the effects are calculated as the average of the values obtained from holding the other variables constant at their 2002 and 2007 values. Table 5.8 reports the results of the decomposition calculated using the two absolute poverty lines. In all but one case, income growth reduced poverty. The largest effect of growth was on the poverty headcount. Indeed, the measured reduction in China’s rural poverty headcount was due entirely to income growth. In contrast, in all cases redistribution increased poverty, although for the poverty headcount and the poverty gap the effect was relatively small. For the squared poverty gap, FGT(2), the redistribution effect increased poverty and was the primary reason for the increases in this measure of poverty. These findings reveal the importance of across-the-board income growth for reductions in the number of rural poor and the poverty gap. Growth alone, however, has not been sufficient to reduce the severity of poverty as measured by the squared poverty gap. The fact that redistribution in all cases has been poverty increasing indicates that recent government transfer programs meant to benefit lower income areas and households have not, on balance, been sufficient to generate a poverty-reducing redistribution of income between higher- and lower-income groups. The structure of income differs between the poor and the nonpoor. Tables 5.9 and 5.10 show the composition of income for these two groups in 2002 and 2007, calculated using the PPP$1.25-per-day poverty line. In both years agriculture remained the most important source of income for

212

Luo Chuliang and Terry Sicular

Table 5.9. Per capita income and its composition for nonpoor and poor households 2002 Nonpoor Mean Wage earnings from migrant employment Wage earnings from local employment Net income from agriculture Net income from nonagricultural businesses Net transfer income Asset income TOTAL

%

2007 Poor

Mean

Nonpoor %

Mean

%

Poor Mean

%

382

12.0

138

13.6

918

19.4

186

16.6

875

27.4

157

15.6

1,048

22.2

191

17.0

1,281 481

40.2 15.1

620 51

61.4 5.1

1,861 541

39.4 11.5

605 33

53.8 2.9

146 4.6 25 0.8 3,189 100.0

42 4.1 1 0.1 1,009 100.0

216 4.6 136 2.9 4,720 100.0

79 7.0 29 2.6 1,124 100.0

Note: Calculated using the PPP$1.25-per-day poverty line, in current prices, weighted, and with the NBS income definition.

the poor. The poor received a large but declining share of their income from agriculture – 61 percent in 2002 and 54 percent in 2007. In comparison, the nonpoor received about 40 percent of their income from agriculture in both years. For the nonpoor, wage earnings were as important as agricultural income and in both years contributed roughly 40 percent of income. Furthermore, for the nonpoor wages from local employment were more important than wages from migrant work, although the gap between these two types of wage income shrank in 2007. For the poor, wages were a less important, although still significant, source of income, contributing 29 percent of income in 2002 and 34 percent in 2007. Nearly half of the wage income of the poor was from migrant employment, which suggests either that the poor tend to live in areas with fewer local job opportunities than the areas where the nonpoor live or that they do not fare as well in local job markets. Nonagricultural businesses were a significant source of income for the nonpoor, but contributed a small and declining share of income for the poor. Net transfer income was relatively small for both groups, although for the poor it increased from 4 percent of income in 2002 to 7 percent in 2007. This may reflect the impact of the dibao program (see Section VIII). Income from assets increased for both the nonpoor and the poor, but remained a relatively small share of income. Because our poverty calculations are done using the NBS income definition, the breakdown of income shown in Table 5.9 does not include imputed rents on owner-occupied housing.

Inequality and Poverty in Rural China

213

Table 5.10. Composition of the income difference between nonpoor and poor households 2002

Wage earnings from migrant employment Wage earnings from local employment Net income from agriculture Net income from nonagricultural businesses Net transfer income Asset income TOTAL

2007

Yuan

%

Yuan

%

244 718 661 429 104 24 2,180

11.2 32.9 30.3 19.7 4.8 1.1 100.0

731 857 1,256 508 137 107 3,597

20.3 23.8 34.9 14.1 3.8 3.0 100.0

Note: Calculated as the absolute gap between the mean incomes of the nonpoor and the poor, as shown in Table 5.9.

Table 5.10 provides additional information about the difference in income between poor and nonpoor households. In both 2002 and 2007 wage earnings, including those from local and migrant employment, accounted for more than 40 percent of the difference in income between these two groups. The importance of migrant wages increased, whereas that of wages from local jobs declined. Agricultural income contributed more than 30 percent of the income difference. Income from transfers and assets accounted for relatively small portions of the income gap.

VI. Migration and Rural Incomes China’s economic reforms have led to an ongoing and substantial flow of rural workers seeking migrant work in the cities. Although migration was already substantial before the change in leadership in the early 2000s, policies adopted since 2000 have more actively supported rural migration. Central government policies include programs to improve employment and living conditions for migrants, as well as some loosening of the household registration (hukou) regulations (Cai et al. 2009). With these policy measures has come growth in the number of migrants. As depicted in Figure 5.2, by 2006 the number of migrants reached about 130 million, equivalent to 26 percent of the rural labor force and up from about 50 million (less than 15 percent of the rural labor force) in 1999 (Sheng 2008). There are different ways to explore the effects of migration on rural incomes, inequality, and poverty, and there are different criteria for identifying migrants, including, for example, by workplace, time outside the

214

Luo Chuliang and Terry Sicular 30

140 120

total rural migrants, millions

25

share of migrants to rural labor, %

20

80

15 60

percent

millions

100

10 40 5

20

0 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006

0

Figure 5.2. Growth in Migrant Employment of Rural Labor. Notes: Taken from Sheng (2008). This source estimates the level of migration using data from the NBS rural household survey. Migrants are defined as members of rural households who receive migrant wage employment. The labor force is defined as the number of members of rural households of working age.

household, and so forth. Here our focus is on the level and distribution of rural household per capita incomes, and we are concerned with that portion of rural household income that is derived from migrant work by members of rural households. We use data from the CHIP surveys on household labor earnings from migrant employment to identify households that engage in migration. Households that report labor earnings from migrant employment are identified as migrant households; households with zero labor earnings from migrant labor are identified as nonmigrant households. This approach differs somewhat from that used in other studies, many of which examine individuals rather than households. Income in the CHIP data includes several types of income derived from migration: wage earnings from migrant employment by current household members, remittances from family and relatives who are not members of the household, and income from household nonagricultural businesses that operate in a location different from the place of residence. Unfortunately, we cannot identify the latter two types of income because they are not reported separately in the CHIP data. Remittances are included in transfer income, and business income earned in a location away from the place of residence is included in nonagricultural business income. The CHIP data

Inequality and Poverty in Rural China

215

55% 50%

45% 40% 35% 2002

30%

2007

25% bottom

2

3

4

5

6

7

8

9

top

Figure 5.3. Percentage of Households Reporting Wage Earnings from Migrant Employment, by Decile.

do provide information on wage earnings from migrant employment of current household members. By 2007 wage earnings from migrant jobs held by current household members exceeded the sum of total transfers and nonagricultural business income (see Table 5.2). Thus, even though we do not know exactly the amount of the remittances and business income earned in other locations, we do know that by 2007 they were less important for rural households than were wages from migrant jobs. As discussed previously, the CHIP data clearly show the growing importance of income from migrant employment between 2002 and 2007, especially for nonpoor households. Moreover, this source of income remained equalizing in both years. Figures 5.3 and 5.4 provide additional information about the distribution of migrant wages and employment. Figure 5.3 shows the percentage of households that reported wage earnings from migration, by decile of the distribution of income. These percentages can be interpreted as household participation rates in migrant employment. In 2002, 33 percent of rural households participated in migrant employment. By 2007, participation had risen to 41 percent. In 2002 participation in migrant employment was distributed fairly evenly across most income deciles, but by 2007 migrant participation was disproportionately concentrated in middle-income groups. The share of wages from migration in total household income (Figure 5.4) shows a similar pattern. Thus, in 2007 migrant employment and earnings were especially important to middle-income rural households. Participation in migrant employment differed markedly across provinces (Table 5.11). In 2007 provincial participation rates ranged from a low of

22%

20% 18% 16% 14%

12% 10%

2002

2007

8% bottom

2

3

4

5

6

7

8

9

top

Figure 5.4. Wage Earnings from Migration as a Percentage of Household Per Capita Income, by Decile.

Table 5.11. Percentage of households in each province of the CHIP rural survey reporting wage earnings from migrant employment Province

2002

2007

Beijing Hebei Shanxi Liaoning Jilin Jiangsu Zhejiang Anhui Fujian Jiangxi Shandong Henan Hubei Hunan Guangdong Guangxi Chongqing Sichuan Guizhou Yunnan Shaanxi Gansu Xinjiang Total

24.38 13.78 6.50 27.78 11.46 36.82 29.04 60.00 n.a. 57.44 18.57 34.34 30.19 43.11 45.66 49.25 38.50 44.60 44.75 21.54 36.22 31.56 13.00 32.95

24.00 32.40 15.00 23.00 n.a. 41.90 12.60 56.22 29.00 n.a. n.a. 48.20 62.60 56.13 50.50 n.a. 63.20 63.09 n.a. 16.14 n.a. 48.86 n.a. 41.39

Note: The provincial percentages are not weighted; the totals are weighted using household-level weights. n.a. indicates “not applicable.”

Inequality and Poverty in Rural China

217

Table 5.12. The relationship between migration and poverty Share of rural population (%)

Poverty headcount (%)

Share of poor rural population (%)

2002 No migrant workers With migrant workers

63.7 36.3

28.3 26.1

65.6 34.4

2007 No migrant workers With migrant workers

51.6 48.4

16.6 11.0

61.6 38.4

Type of household

Note: Migration is identified by whether the household reports wage earnings from migrant employment. Poverty is calculated using the PPP$1.25-per-day poverty line. Weighted; poverty calculations use the NBS income definition.

13 percent in Zhejiang to a high of 63 percent in Hubei, Chongqing, and Sichuan. Changes over time also differed among the provinces. Participation in migration rose sharply in Hebei, Henan, Hubei, Hunan, Chongqing, Sichuan, and Gansu, but declined in Liaoning, Zhejiang, Anhui, and Yunnan. Lagging participation by the poorer deciles, as shown in the preceding figures, raises questions about whether migration contributed to a reduction in poverty. Analyzing the contribution of migration to poverty reduction is difficult, as migration has multiple direct and indirect effects on income (Poverty Reduction and Economic Management Division 2009). Also, poor households may be less able to migrate due to a lack of resources and networks, thereby rendering the relationship between migration and poverty bidirectional (Poverty Reduction and Economic Management Division 2009). Nevertheless, some simple statistics in Table 5.12 provide an indication of the relationship between migration and poverty.3 In 2002 the poverty rates for individuals in migrant and nonmigrant households were similar – about 26 to 28 percent. In other words, individuals living in households without migrant earnings were no more likely to be poor than were those living in households with migrant earnings. Moreover, the share of poor living in households without migrant earnings was similar to the share of the total rural population in such households. By 2007, poverty rates had declined for households both with and without migrant earnings, but more so for households with migrant earnings. 3

We adapt this table from Poverty Reduction and Economic Management Division (2009: 102, Table 5.51), which provides the same statistics for 2003.

218

Luo Chuliang and Terry Sicular

Consequently, in 2007 the poverty rate for migrant households was lower than that for nonmigrant households; also, a larger share of the poor – nearly two-thirds – was living in households without migrant earnings. These statistics are consistent with a scenario in which migration contributed to poverty reduction and those who remained below the poverty line in 2007 were disproportionately in households that did not have migrant income. Thus, the relationship between migration and poverty appears to be evolving over time.

VII. The Elimination of Taxes and Fees In 2005 the Chinese government announced the abolition of agricultural taxes, effective January 1, 2006 (Xinhua News Agency 2005). This announcement was the final step in the “rural tax and fee reforms” that were initiated in the 1990s. As discussed in Sato, Li, and Yue (2008), since 2000 the Chinese government has carried out a comprehensive reform of agricultural taxes and fees. During the first phase of this reform (2000–2003), informal local levies were replaced by formal taxation (feigaishui). During the second phase (2004–2006), as part of its goal of eliminating agricultural taxes, the government implemented a program of gradual tax reductions and experimented with the full abolition of agricultural taxes in some regions (Sato et al. 2008; Xinhua News Agency 2005). As of January 1, 2006, the abolition of agricultural taxes was to be completed nationwide. Using earlier rounds of the CHIP rural data, Sato et al. (2008) analyze the distributional effects of the tax and fee reforms through 2002. Here we examine the changes between 2002 and 2007. In 2002 the tax and fee reforms were ongoing, with implementation varying regionally. In 2007 agricultural taxes and fees had been eliminated nationwide, at least in principle. The 2007 CHIP data allow us to verify whether, from the perspective of rural households, this goal was achieved. As discussed in Sato et al. (2008), rural households in China have paid a variety of taxes and fees. The CHIP rural data for 2007 contain a single “total” value of taxes and fees paid by the household, including both formal taxes paid to the state as well as levies and fees collected by the village and township. We do not have information on the composition of this total. Also, the reported taxes and fees do not include contributions of unpaid labor. Historically, an important component of rural taxation was in-kind taxation in the form of contributions of unpaid labor. This form of taxation was also eliminated as part of the rural tax reforms. We cannot examine it here due to lack of data for 2007, but the 2002 CHIP data indicate that

Inequality and Poverty in Rural China

219

Table 5.13. Taxes and fees paid by rural households (per capita), by deciles

Taxes and fees (yuan) Income decile Bottom 2nd 3rd 4th 5th 6th 7th 8th 9th Top Average

Before-tax income per capita (yuan)

Tax rate (%)

2002

2007

2002

2007

2002

2007

56.95 63.46 71.51 79.10 81.68 81.88 87.95 86.11 91.30 121.32 82.12

4.71 8.32 5.96 10.79 10.45 11.24 9.48 15.92 17.61 33.80 12.83

818.07 1,247.84 1,557.85 1,850.56 2,163.22 2,496.71 2,894.85 3,437.63 4,323.02 7,747.05 2,853.21

1,139.19 1,965.69 2,487.20 2,979.16 3,491.65 4,042.76 4,718.90 5,670.42 7,171.74 12,642.26 4,630.41

6.96 5.09 4.59 4.27 3.78 3.28 3.04 2.50 2.11 1.57 2.88

0.41 0.42 0.24 0.36 0.30 0.28 0.20 0.28 0.25 0.27 0.28

Note: The tax rate is equal to per capita taxes and fees divided by household per capita net before-tax income. In current prices, calculated with weights and using the CHIP income definition plus taxes, so that the tax rates are percentages of the before-tax income.

this form of taxation had already been substantially reduced by 2002, at which time only 28 percent of the rural households reported contributing unpaid labor, and the mean unpaid labor contribution was less than two days. Table 5.13 shows the level of taxes and fees reported by households in absolute terms and as a percentage of income. Rural taxes and fees declined markedly in both absolute terms and relative to income. Indeed, as of 2007 taxes and fees took a trivial fraction of rural household incomes. These data indicate that the government’s goal of abolishing taxes and fees was effectively accomplished. In 2002 taxes and fees were distributed regressively, as revealed in the higher tax rates for households in the lower deciles (Table 5.13). In 2007 the tax rate for the bottom two deciles was higher than that for the higher deciles, but for all deciles the tax rates were well below 1 percent. This pattern suggests that the abolition of agricultural taxes and fees was equalizing, although given the relatively low level of taxes in 2002, the net impact on income inequality may not have been very large. Indeed, in 2002, inequality of after-tax income was higher than that of before-tax income (0.354 vs. 0.338). In 2007 the Gini coefficients of before- and after-tax incomes were identical (0.358). (See Table 5.3.)

220

Luo Chuliang and Terry Sicular Table 5.14. Taxes and fees paid by poor and nonpoor households (per capita) 2002

2007

Nonpoor Yuan PPP$1.25 per day Official poverty line 0.5 × median income 0.6 × median income

89 85 85 87

Poor

Nonpoor

Poor

Tax rate Tax rate Tax rate Tax rate (%) Yuan (%) Yuan (%) Yuan (%) 2.56 2.73 2.71 2.63

62 60 61 62

5.36 7.00 6.63 5.88

14 13 14 15

0.27 0.27 0.27 0.27

5 6 5 5

0.40 0.63 0.38 0.32

Note: See the notes to Table 5.13. Households are grouped as poor or nonpoor using the NBS income definition (excluding imputed rents on owner-occupied housing). The tax rate is calculated as a percentage of the before-tax income, which is equal to CHIP income plus taxes. Note that the 2007 tax rates for the nonpoor in fact are slightly different, but all round to the same value.

Table 5.14, which shows taxes and fees paid by the poor versus those paid by the nonpoor, reveals the differential impact of taxes and fees for those in the lowest-income groups. In 2002 taxes and fees accounted for 5 to 7 percent of the before-tax income of the poor, more than double the tax rate for the nonpoor. The average amount of taxes and fees paid by the poor in 2002 was large enough to account for a significant share of the poverty gap. As shown in Table 5.15, in 2002 the average poverty gap, measured using the PPP$1.25 per day poverty line, was 442 yuan; on average, those who fell below this poverty Table 5.15. Taxes and fees paid by the poor relative to the poverty gap Average poverty gap per capita (yuan)

Average taxes and fees per capita (yuan)

Taxes and fees as a % of the poverty gap

2002 PPP$1.25 per day Official poverty line 0.5×median income 0.6×median income

441.74 255.64 287.58 363.79

62.88 60.29 60.93 62.09

14.23 23.58 21.19 17.07

2007 PPP$1.25 per day Official poverty line 0.5 × median income 0.6 × median income

565.70 452.55 572.93 676.69

5.41 5.77 5.28 5.20

0.96 1.27 0.92 0.77

Note: In current prices. Calculated with weights and using the NBS definition of income.

Inequality and Poverty in Rural China

221

line paid 63 yuan in taxes and fees; that is, taxes and fees were equivalent to 14 percent of the poverty gap. Using the other poverty lines, we find that taxes and fees were equivalent to larger percentages of the poverty gap. For example, on average, in 2002 taxes and fees paid by households below the official poverty line were equivalent to nearly one-quarter of the average poverty gap. By 2007 the average amount of taxes and fees paid by the poor was much lower, both in absolute terms and relative to the poverty gap. These statistics suggest that the abolition of rural taxes and fees was beneficial to the poor. However, some observers have noted that the abolition of rural taxes and fees may have had negative indirect effects on the poor, as it resulted in a loss of revenue for local governments and thereby negatively affected their ability to fund social welfare programs, such as the dibao program (Zhang and Sun 2009).

VIII. The Minimum Living Standard Guarantee A significant component of the government’s new rural policy program was the minimum living standard guarantee, or dibao, program. The government initiated the dibao program in urban areas in the early 1990s, and local experiments with rural dibao programs began not much later, largely in the more developed areas (Xu and Zhang 2010). By 2001 rural programs were quite widespread, but at that time they were locally funded and varied considerably in levels of support and criteria for eligibility, and many difficulties in implementation arose after the reform of rural taxes and fees, which reduced local revenue (Xu and Zhang 2010). After 2004, the rural dibao program was enlarged, especially during and after 2006. By the end of 2006, roughly 80 percent of the provinces and counties in China had adopted rural dibao programs (Xu and Zhang 2010). In early 2007 the central government announced that it would provide central subsidies for the program and that, by the end of that year, the program would be implemented nationwide in all counties (Xinhua News Agency 2007a, 2007b; Xu and Zhang 2010). According to official statistics, in 2007 35.7 million rural individuals (4.9 percent of the rural population) received relief under the dibao program, up from 4 million (0.5 percent) in 2002 (Department of Social, Science and Technology Statistics of the NBS 2008: 330; NBS 2009: 89, 939). The dibao program was expected to absorb or complement several previous programs that had provided subsidies for poor households, including the five-guarantee (wubao) program and subsidies for destitute households

222

Luo Chuliang and Terry Sicular

Table 5.16. Basic statistics on individuals in dibao versus non-dibao households, from the CHIP rural household survey, 2007

Percentage of individuals (%) Income per capita (yuan) Net transfer income per capita (yuan) Net transfer income per capita, as a share of the total income per capita (%) Estimated dibao subsidy per capita, as a share of the average household income per capita (%)

dibao

Nondibao

2.46 3,029 197 7.2

97.54 4,658 217 4.2

15.4

0

Note: Based on the reported national average expenditures of 38.8 yuan per person per month in 2007 the annual dibao subsidy per capita for dibao households is estimated to be 466 yuan (Ministry of Civil Affairs 2008). Non-dibao households are assumed to receive zero dibao subsidies. Weighted; CHIP income definition.

(tekun jiuzhu). The tekun program, which had provided targeted assistance to households that lacked labor because of age, illness, or death, was gradually to be absorbed, where and when local fiscal capacity and funding from higher levels made it possible to implement the more comprehensive dibao program (Xu and Zhang 2010). The five-guarantee program in principle has been separate from and complementary to the dibao program, although the distinction between the two programs is not always clear at the local level.4 By 2007 the dibao program was by far China’s broadest nationwide rural social relief program, accounting for three-quarters of the rural recipients of social relief, followed in a far second place by the five-guarantee program which covered 5 million recipients (Department of Social, Science and Technology Statistics of the NBS 2008: 330). In 2007 the average dibao threshold was 70 yuan per person per month (840 yuan per person per year), an amount slightly higher than the official poverty line that year (785 yuan). In that year, the average spending per recipient under the dibao program was 466 yuan (Ministry of Civil Affairs 2008; Poverty Reduction and Economic Management Division 2009; Xinhua News Agency 2007b; Zhang and Sun 2009), an amount close to the average poverty gap (Table 5.15). In principle, then, the dibao program had the potential to substantially alleviate poverty if it was well implemented and effectively targeted. Table 5.16 presents statistics on dibao households in the CHIP rural survey. In 2007, the prevalence (weighted) of rural individuals in dibao 4

Personal communication from the World Bank.

Inequality and Poverty in Rural China

223

households nationally was 2.5 percent.5 This percentage is lower than the percentage of the rural population receiving dibao subsidies as reported by the NBS (4.9 percent). The lower percentage of dibao households reported in the CHIP rural household survey may be due to an undersampling of poor households, a known feature of the NBS household survey samples from which the CHIP survey is drawn. It could also reflect misreporting. Participating households may have been unaware that they were receiving transfers under the dibao program, as opposed to some other programs such as the fiveguarantee household program. It is also possible that the official statistics are misreported. Local-level governments in China have been known to overstate their implementation of central government policies. Table 5.16 also shows the differences between dibao and non-dibao households. Income per capita is lower in dibao households than in non-dibao households, but at 3,029 yuan per year, it is still substantially higher than the national poverty line as well as the national average dibao threshold. The CHIP questionnaire did not ask about the amount of dibao subsidies received by the households, but in principle, dibao subsidies would be counted as transfer income. As shown in Table 5.16, net transfer income for dibao and non-dibao households in the CHIP survey is similar, although this may be due to the fact that non-dibao households received larger private transfers. If we assume that the average dibao subsidies were equal to the average monthly expenditure per capita on the dibao program in 2007, then the annual dibao subsidies would have been equivalent to 15 percent of the per capita income of dibao households. This amount is larger than their average reported net transfer income, which in 2007 was only 7.2 percent of per capita income. Such a discrepancy might arise if dibao expenditures reported by the Ministry of Civil Affairs overstate the subsidy amounts actually received by households, or if transfer income in the CHIP survey does not fully reflect the dibao transfers. In many areas village leaders are responsible for implementation of the dibao program and slippage is possible at the ground level. Dibao participation rates vary substantially among provinces, as shown in Figure 5.5 for the provinces covered in the 2007 CHIP rural survey. The dibao participation rate is by far the highest in Yunnan, where almost one out of ten 5

The CHIP rural household survey included a question asking households if they participated in the dibao program. The percentages reported here are calculated as the total number of individuals in dibao households divided by the total number of individuals in all households.

224

Luo Chuliang and Terry Sicular 10 9 8

percentage

7 6 5 4 3 2 1

ng J G ian u a gs ng u do ng H u Li bei ao C n ho ing ng qi ng An hu G i an s Sh u an x Fu i jia H n en Si a n ch ua Yu n nn an

an

ia

Zh

ej

ei

un H

eb H

Be

ijin

g

0

Figure 5.5. Percentage of Individuals in Rural Dibao Households, 2007, by Province. Note: Unweighted.

individuals resides in a dibao household. The lowest participation rate is in Beijing. This regional variation is not surprising given the differing poverty rates and the variations in implementation of the dibao program, which is largely dependent on local fiscal resources plus some central supplements in regions that face fiscal difficulties. It has been reported that income thresholds and subsidies vary among regions and generally are lower in poor localities (Poverty Reduction and Economic Management Division 2009; Xinhua News Agency 2007a). Does the dibao program effectively target the poor? The CHIP data suggest that the dibao glass is half full: individuals in poor households benefited more than those in nonpoor households, but there was leakage. As shown in Table 5.17, in 2007 between 15 and 45 percent of individuals in dibao households were poor, depending on the poverty line. The poverty rates for non-dibao households were substantially lower. Also, a much higher share of the poor than of the nonpoor lived in dibao households. The share of the poor receiving dibao benefits was well below 10 percent for all four poverty lines.6 In other words, the overwhelming majority of the 6

We note that our percentages of poor households participating in the dibao program are very different from those reported by official sources. Government announcements in 2007 reported that 70 percent or more of China’s rural poor benefited from the dibao program (Xinhua News Agency 2007a, 2007b). The reason for this large discrepancy is unclear.

Inequality and Poverty in Rural China

225

Table 5.17. The relationship between dibao participation and poverty, 2007 Poverty rate of individuals in non-dibao versus dibao households (%)

PPP$1.25 per day per person Official poverty line 0.5 × median income 0.6 × median income

% of nonpoor and poor individuals living in dibao households

Non-dibao

Dibao

Nonpoor

Poor

13.30 5.34 13.73 20.46

37.05 15.31 37.64 45.63

1.80 2.21 1.79 1.69

6.56 6.74 6.47 5.33

Note: Weighted. Poverty is calculated using the NBS definition of income.

poor – more than 90 percent – lived in households that did not receive dibao subsidies. Also, even for our highest poverty line, more than half of the dibao households were not poor. These statistics together with some reports about irregularities in implementation of the program at the local level (Deng and Wong 2008; Lin and Wong 2012) suggest that there is substantial room for improvement in implementation of the rural dibao program.

IX. Conclusion In this chapter we used the CHIP rural survey data to examine changes in rural household incomes and inequality between 2002 and 2007, a period of renewed emphasis on rural policy. Overall, between 2002 and 2007 conditions improved for rural households, reversing trends in the late 1990s through the 2000s. We find that rural incomes grew substantially and more rapidly than during the preceding period. The fact that the urban-rural income gap continued to widen thus was not due to stagnation in rural incomes, but rather to the more rapid growth in urban incomes. Income growth was the result of increases in income from multiple sources, including agriculture as well as off-farm employment and other sources. Growth was most rapid in asset income, although in rural areas this source of income remained small. Imputed rental income from owneroccupied housing also increased rapidly. By 2007 these two sources of income together constituted more than 10 percent of rural household income, reflecting the rising importance of property income in rural China. Income from migrant employment, narrowly defined as wages earned by rural household members from migrant jobs, also increased rapidly. Indeed,

226

Luo Chuliang and Terry Sicular

by 2007, such income accounted for nearly one-fifth of per capita income in the rural areas, approaching the amount of income from local wage employment. These trends suggest that the easing of restrictions on labor movement was beneficial for rural households. The importance of migrant income would have been even greater if our calculations had included remittances from migrant family members, not to mention income of former rural households that had relocated. Despite growth in nonagricultural forms of income, agriculture retained its place as the largest single source of income for rural households. Agricultural income grew at a fairly rapid pace, likely reflecting the recovery of farm prices and technological improvements, as well as new policies supporting agriculture. Rural income growth was fairly widely shared, so that inequality increased only slightly between 2002 and 2007. Stable inequality was partly due to the growth in migrant wage earnings as well as to the growth in agricultural income, both of which were relatively equally distributed. As measured using the absolute poverty lines, the poverty headcount rate and the poverty gap declined substantially. Yet, although income growth among the poor was sufficient to raise roughly one-half of the poor from absolute poverty, among those who remained poor the severity of their poverty increased. Also, relative poverty showed no improvement. Income growth in the low-income groups was thus insufficient to catch up with the median incomes. Using the CHIP data, we explored the impact of the elimination of rural taxes and fees. The data reveal the near elimination of tax and fee payments by rural households. As taxes and fees were regressive in 2002, their elimination reduced inequality, but because the level of taxes and fees was already low in 2002, the size of this impact was small. We also note that rural taxes and fees had been a source of local public revenues; thus, their near abolition may have had negative consequences on local public spending that, in turn, affected rural households. These indirect effects are not captured by our calculations. Our analysis of the dibao subsidies raises questions about the effectiveness of the minimum living standard guarantee program and its impact on poverty reduction, at least as of 2007. Although the program was more beneficial to the poor than to the nonpoor, we find that the overwhelming majority of the poor lived in households that did not report receiving dibao subsidies. Discrepancies between dibao numbers based on the CHIP rural survey data and those in official reports raise questions and suggest the need for additional research.

Inequality and Poverty in Rural China

227

References Adams, R.H., Jr. (1999), “Nonfarm Income, Inequality and Land in Rural Egypt,” World Bank Policy Research Working Paper No. 2178, The World Bank, Washington, DC. Brandt, L. and C.A. Holz (2006), “Spatial Price Differences in China: Estimates and Implications,” Economic Development and Cultural Change, 55(1), 43–86. Cai, F., Y. Du, and M. Wang (2009), “Migration and Labor Mobility in China,” Human Development Research Paper No. 2009/09, United Nations Development Programme, New York. Chen, S. and M. Ravallion (2008), “China Is Poorer than We Thought, but No Less Successful in the Fight against Poverty,” World Bank Policy Research Working Paper No. 4621, The World Bank, Washington, DC. Chen, X. (2009), “Review of China’s Agricultural and Rural Development: Policy Changes and Current Issues,” China Agricultural Economic Review, 1(2), 121–135. Chen, X. (2010), “Issues of China’s Rural Development and Policies,” China Agricultural Economic Review, 2(3), 233–239. Datt, G. and M. Ravallion (1992), “Growth and Redistribution Components of Changes in Poverty Measures: A Decomposition with Applications to Brazil and India in the 1980s,” Journal of Development Economics, 38(2), 275–295. Deng, D. and Z. Wang (2008), “Woguo nongcun dibao zhidu cunzaide wenti jiqi tantao: Yi xiancun nongcun dibao zhidu cunzaide wenti wei jiaodu” (Problems in Our Country’s Rural Minimum Living Standard Guarantee Program and Its Investigation: From the Perspective of Existing Problems in the Current Rural Dibao Program), Shandong jingji, no. 1, 61–64. Department of Rural Surveys, National Bureau of Statistics (NBS) (2008), China Yearbook of Rural Household Survey, Beijing: Zhongguo tongji chubanshe. Department of Rural Surveys, National Bureau of Statistics (NBS) (2010), China Yearbook of Rural Household Survey, Beijing: Zhongguo tongji chubanshe. Department of Social, Science and Technology Statistics of the National Bureau of Statistics (NBS) (2008), Zhongguo shehui tongji nianjian 2008 (China Social Statistical Yearbook 2008), Beijing: Zhongguo tongji chubanshe. Foster, J., J. Greer, and E. Thorbecke (1984), “A Class of Decomposable Poverty Measures,” Econometrica, 52(3), 761–766. Gale, F., B. Lohmar, and F. Tuan (2005), “China’s New Farm Subsidies,” USDA Outlook, WRS-05–01, United States Department of Agriculture, Washington, DC, http://www .ers.usda.gov/publications/WRS0501/WRS0501.pdf. Accessed February 22, 2011. Gustafsson, B., S. Li, and T. Sicular (2008), “Inequality and Public Policy in China: Issues and Trends,” in B. Gustafsson, S. Li, and T. Sicular, eds., Inequality and Public Policy in China, 1–34, New York: Cambridge University Press. Huang, J., Y. Jun, Z. Xu, S. Rozelle, and N. Li (2007), “Agricultural Trade Liberalization and Poverty in China,” China Economic Review, 18(3), 244–265. Khan, A. R., K. Griffin, C. Riskin, and R. Zhao (1992), “Household Income and Its Distribution in China,” China Quarterly, no. 132, 1029–1061. Khan, A. R. and C. Riskin (1998), “Income and Inequality in China: Composition, Distribution and Growth of Household Income, 1988 to 1995,” China Quarterly, no. 154, 221–253.

228

Luo Chuliang and Terry Sicular

Khan, A.R. and C. Riskin (2008), “Growth and Distribution of Household Income in China between 1995 and 2002,” in B. Gustafsson, S. Li, and T. Sicular, eds., Inequality and Public Policy in China, 61–87, New York: Cambridge University Press. Lin, W. and C. Wong (2012), “Are Beijing’s Equalization Policies Reaching the Poor? An Analysis of Direct Subsidies under the ‘Three Rurals’ (Sannong),” China Journal, no. 67, 23–45. Ministry of Civil Affairs (2008), 2007 nian minzheng shiye fazhan tongji baogao (Statistical Report on the Development of Civil Affairs Work 2007), Beijing, http://www.mca.gov. cn/article/zwgk/tjsj/. Accessed March 18, 2011. National Bureau of Statistics (NBS) (2008), Zhongguo tongji nianjian 2008 (China Statistical Yearbook 2008), Beijing: Zhongguo tongji chubanshe. National Bureau of Statistics (NBS) (2009), Zhongguo tongji nianjian 2009 (China Statistical Yearbook 2009), Beijing: Zhongguo tongji chubanshe. Osberg, Lars (2000), “Poverty in Canada and the United States: Measurement, Trends, and Implications,” Canadian Journal of Economics, 33(4), 847–877. Poverty Reduction and Economic Management Division, East Asia and Pacific Region (2009), “From Poor Areas to Poor People: China’s Evolving Poverty Reduction Agenda,” World Bank Report No. 47349-CN, The World Bank, Washington, DC. Ravallion, M. (1992), “Poverty Comparisons: A Guide to Concepts and Methods,” World Bank Living Standards Measurement Study Working Paper No. 88, The World Bank, Washington, DC. Ravallion, M. (1994), Poverty Comparisons, Philadelphia: Harwood Academic. Ravallion, M. and S. Chen (2007), “China’s (Uneven) Progress against Poverty,” Journal of Development Economics, 82(1), 1–42. Sato, H., S. Li, and X. Yue (2008), “The Redistributive Impact of Taxation in Rural China, 1995–2002: An Evaluation of Rural Taxation Reform at the Turn of the Century,” in B. Gustafsson, S. Li, and T. Sicular, eds., Inequality and Public Policy in China, 312–336, New York: Cambridge University Press. Sheng, L. (2008), Liudong haishi qianyi: Zhongguo nongcun laodongli liudong guocheng de jingjixue fenxi (Floating or Migration: Economic Analysis of Laborers’ Out-migration Behavior in Rural China), Shanghai: Yuandong chubanshe. Shorrocks, A.F. (1999), “Decomposition Procedures for Distributional Analysis: A Unified Framework Based on the Shapley Value,” Department of Economics, University of Essex. Sicular, T., X. Yue, B. Gustafsson, and S. Li (2007), “The Urban-Rural Income Gap and Inequality in China,” Review of Income and Wealth, 53(1), 93–126. Stark, O., J. E. Taylor, and S. Yitzhaki (1986), “Remittances and Inequality,” Economic Journal, 96(383), 722–740. Whyte, M.K., ed. (2010), One Country, Two Societies: Rural-Urban Inequality in Contemporary China, Cambridge, MA: Harvard University Press. Xinhua News Agency (2005), “China’s Legislature Abolishes 2,600-Year-Old Agricultural Tax,” People’s Daily Online, December 30, http://english.peopledaily.com.cn/200512/ 30/eng20051230 231752.html. Accessed March 9, 2011. Xinhua News Agency (2007a), “Chinese Government Decides to Subsidize All Rural Poor,” People’s Daily Online, May 24, http://english.peopledaily.com.cn/200705/24/ eng20070524 377380.html. Accessed March 9, 2011.

Inequality and Poverty in Rural China

229

Xinhua News Agency (2007b), “77% of Rural Poor Covered by Allowance System,” May 28, http://www.china.org.cn/english/government/212112.htm. Accessed March 9, 2011. Xinhua News Agency (2008), “No. 1 Central Document Focuses on Rural Issues,” China Daily, January 31, http://www.chinadaily.com.cn/china/2008-01/31/content 6432725.htm. Accessed February 22, 2011. Xinhua News Agency (2010), “Consecutive No. 1 Central Documents Target Rural Issues,” China Daily, February 1, http://www.chinadaily.com.cn/bizchina/2010-02/ 01/content 9409941.htm. Accessed February 22, 2011. Xu, Y. and X. Zhang (2010), “Rural Social Protection in China: Reform, Performance and Problems,” in J. Midgely and K.L. Tang, eds., Social Policy and Poverty in East Asia: The Role of Social Security, 116–127, New York: Routledge. Zhang, X. and L. Sun (n.d.), “Social Security System in Rural China: An Overview,” CATSEI Project Report, Chinese Agricultural Transition: Trade, Social and Environmental Impacts, Beijing, http://www.catsei.org/contents/73/1076.html. Accessed March 9, 2011.

SIX

The Evolution of the Migrant Labor Market in China, 2002–2007 John Knight, Deng Quheng, and Li Shi

I. Introduction Migration simply did not figure in the first of the China Household Income Project (CHIP) volumes, which was based on a 1988 national household survey (Griffin and Zhao 1993). This was partly because that survey relied entirely on samples drawn from the annual national household survey of the National Bureau of Statistics (NBS), which contained only rural households and urban hukou (household registration) households. That sampling procedure in turn reflected the underlying reality: rural-urban migration was restricted, limited, and unimportant. The same is true of the volume based on the 1995 CHIP survey (Riskin, Zhao, and Li 2001), although it contains an analysis of migrants based on the rural sample (Li 2001). The 2002 CHIP survey was the first to include a separate sample of rural migrants to the cities, and migrants were integrated into several of the chapters in the resultant volume (Gustafsson, Li, and Sicular 2008). A sample of rural-urban migrants was again included in the 2007 CHIP survey, on which the current volume is based. The greater emphasis given to migrants and migration in each succeeding CHIP survey reflects an important development in the Chinese economy. What has been referred to as the greatest migration in human history is now critical to an analysis of China’s economic growth, income distribution, poverty alleviation, and labor market. Indeed, it is the subject of a separate volume that is also based on the 2007 survey (Meng and Manning 2010), but that volume does not address the question posed in this chapter. This research was conducted while John Knight was visiting Beijing Normal University. We are grateful to Simon Cox, Fung Kwan, Gus Ranis, and Adrian Wood for helpful comments. We draw on our longer paper (Knight et al. 2011), which extends beyond an analysis of the CHIP national household surveys.

230

The Evolution of the Migrant Labor Market in China, 2002–2007

231

The famous Lewis model (Lewis 1954) provides a good framework for evaluating the success of a developing economy and for explaining the ways in which the fruits of economic development are spread. Within a competitive market economy, it is only when the economy emerges from the first, labor-surplus, classical stage of the development process and enters the second, labor-scarce, neoclassical stage that real incomes generally begin to rise. Up to that point, the benefits of economic growth can accrue in the form of the absorption of surplus labor, but not in the form of generally rising real incomes. Beyond that point, the scarcity of labor can be a powerful force for reducing inequality in labor income. When the economic reforms commenced, there is no doubt that China was an extreme example of a labor-surplus economy. There was surplus labor both in the rural areas (where it was disguised as underemployment in the communes) and in the urban areas (where it was disguised as underemployment in the stateowned enterprises [SOEs]). During the reform period, China achieved rapid economic growth, averaging more than 9 percent per annum during the three decades from 1978 to 2008. Nevertheless, during the same period the labor force grew by 380 million, or 90 percent, equivalent to 2.3 percent per annum. Has the surplus labor by now been absorbed productively into the economy? Reports or data on rising migrant wages, at least in various growth points of the Chinese economy, have led some researchers to argue that China has now reached the Lewis turning point (Cai, Du, and Zhao 2007; Park, Cai, and Du 2010; Wang 2008). However, others argue either that migrant wages have barely increased (Du and Pan 2009; Meng and Bai 2007) or that there is still evidence of widespread surplus labor in rural China (Kwan 2009; Minami and Ma 2009). The issue has become a lively and contentious topic in the Chinese media. For instance, State Council Councillor Ma Li is reported to have argued that China has a sufficient labor pool for the next forty years (Xin and Shan 2010). The inconclusive nature of the debate reflects both the use of different methodologies and the lack of the required data to test these alternative hypotheses. Nevertheless, it is possible that there is some truth in both arguments. Can the apparently contradictory pieces of evidence be reconciled? In this chapter we explore the light that the CHIP national household surveys of 2002 and 2007 throw on the debate. Section II briefly provides some background information on trends in the Chinese labor market. Section III describes relevant aspects of the surveys on which we draw. Section IV analyzes wage functions for the rural-urban migrant subsamples of the surveys in order to examine and explain migrant

232

John Knight, Deng Quheng, and Li Shi

wage behavior in urban China. An attempt is made in Section V to measure the remaining pool of potential migrant labor in rural China by means of the rural subsamples and probit analyses of migration functions. Section VI summarizes, reflects, and presents our conclusions.

II. Trends in the Chinese Labor Market China reached the limits of its land availability decades ago. The total land area sown in 1995 was no more than 6 percent more than that in 1952. During the same period, the rural labor force increased by 150 percent, reaching its peak in 1995. Surplus labor was present in the communes but it was camouflaged by the work-point system. Numerous attempts to measure the extent of surplus labor in rural China produced a range of estimates, with most economists suggesting that surplus labor represented one-third of the rural labor force in the 1980s (Knight and Song 1999: chap. 2; Taylor 1988). Reflecting the pro-population policies of the Maoist period, the rural labor force grew rapidly during the next generation, that is, the 1980s. It was only in the late 1990s that the effects of population control measures, such as the “late-sparse-few” and “one-child” family policies, introduced in the 1970s, began to have an effect on the labor market. Table 6.1 shows various measures of the labor force and employment over the 1995–2007 period. The rural labor force began to decline gently in the mid-1990s. As rural nonfarm employment grew (by 1.6 percent per annum), farm employment fell markedly (by 1.4 percent per annum). Urban employment increased rapidly (by 3.7 percent per annum). Formal sector employment, including SOEs and urban collective enterprises (UCEs), actually declined (by 2.2 percent per annum), whereas the most dynamic sector was urban informal employment (rising by 13.1 percent per annum). The natural increase in the urban-born labor force was far too slow to meet the growing demand for labor by urban employers, thus the increasing shortfall was met by rural-urban migration. According to Sheng (2008), using data taken from the NBS website, the number of rural-urban migrants rose from 30 million in 1995 to 132 million in 2006. Migrants accounted for 7 percent of the rural labor force in 1995 but they constituted no less than 26 percent in 2006. It is difficult to measure the number of migrants accurately on an annual basis, but such orders of magnitude are not in dispute: migrant labor was the most dynamic component of labor force activity during the decade, growing by perhaps 13.1 percent per annum. The table also shows that average urban real wages rose by 11.2 percent per annum over the 1995–2007 period. This rate of growth was far

The Evolution of the Migrant Labor Market in China, 2002–2007

233

Table 6.1. Labor force and employment in China, 1995–2007 Million 1995 2007

%

% per annum

1995–2007 1995–2007 1995–2007

Rural areas Labor force Employment TVEs, PEs, and self-employed Household farming Employment in primary industry

490 490 165 325 355

476 476 200 276 314

−14 −14 35 −49 −41

−2.9 −2.9 21.2 −15.1 −11.5

−0.03 −0.03 1.62 −1.36 −0.01

Urban areas Labor force Employment

196 190

325 294

131 104

66.8 54.7

4.43 3.70

Formal sector Informal sector Unemployment

149 41 6

114 180 31

−35 139 25

−23.5 339.0 416.7

−2.21 13.12 15.55

30

132

102

340.0

13.14

Yuan per annum, average (1995 prices) Urban real wage 5,348 19,904 Rural real income per capita 1,578 3,289

14,556 1,711

272.2 108.4

11.16 6.31

Rural-urban migrants

Note: TVE denotes township and village enterprises, and PE denotes rural private enterprises. Sources: NBS (2008: Tables 4–2, 4–3, 4–5, 4–8, 10–2) (and earlier versions of the same tables where necessary). For rural-urban migrants, see Sheng (2008).

higher than that of rural real income per capita (6.3 percent per annum). However, official sources report only the wages of urban residents and not those of rural-urban migrants. The pay of the former has been subject to institutional and politically motivated determination and, in recent years, informal profit-sharing associated with a form of efficiency wage theory, whereas the pay of the latter has often been determined separately (Knight and Li 2005; Knight and Song 2005: chap. 7). Thus, without information on migrant wages, it cannot be inferred from this officially reported wage increase that there has been a shortage of migrant labor. The 2007 CHIP national household survey shows the ratio of the average monthly wage of urban residents to that of rural-urban migrants to be 1.49. Although migrants are subject to market forces more than urban residents, the migrant wage is greater than the opportunity cost. The 2007 survey also asked rural-urban migrants about their income had they remained in the village. The ratio of the average migrant wage to the average counterfactual village income per month was 2.43. According to probabilistic migration

234

John Knight, Deng Quheng, and Li Shi

models, this urban-rural income differential should have induced an influx of labor and generated substantial urban unemployment among migrants. However, the restrictions on migrant employment and settlement in the cities imposed by the central and local governments held down migrant unemployment (Knight and Song 2005: chaps. 5 and 8). According to the 2002 CHIP survey, the unemployment rate of workers in migrant urban households was only 2.8 percent (Li and Deng 2004).

III. The Data The CHIP surveys for 2002 and 2007 cover three types of households: urban local households, rural households, and rural-urban migrant households. Each type of household was surveyed separately. The sample of urban local households and rural households is a part of the large NBS sample. The 2002 survey for rural households covers twenty-two provinces, with the condition that they are representative of various regions in rural China. The number of sampled households was distributed among the twentytwo provinces roughly in proportion to their populations. The provincial statistical bureaus were given autonomy to choose the number of sampled counties, but there had to be at least fifty households in each selected county, and villages within them had to be stratified by income level. In all, 9,200 households and 37,969 individuals were surveyed in 120 counties. The 2002 survey of registered urban households was conducted in twelve of the preceding twenty-two provinces. In all, 6,835 households and 20,632 individuals were surveyed in seventy cities. Income questions were posed with the objective of measuring household disposable income. Households were required to answer questions about wage income and other income of each working member and about income from family businesses. Rural households were asked questions on working time inside and outside the township. The 2002 rural-urban migrant survey sampled a total of 2,000 households: 200 households in each of the eastern and central provinces and 150 households in each of the western provinces. A person is defined as a migrant if he or she holds a rural hukou and has been living in an urban area for more than six months. Within each province, 100 households were sampled in the capital city and fifty households in each of the selected middle-sized cities. Within each city, rural-urban migrant households were sampled from residential communities, thus excluding migrant workers living at construction sites, in factories, or outside the city. The sample therefore excluded

The Evolution of the Migrant Labor Market in China, 2002–2007

235

short-term in-migrants and those not living in households. The questionnaires included questions regarding wage, business income, consumption, and job characteristics of individual members and households. Each of the 2007 CHIP surveys of rural, urban, and rural-urban migrant households was conducted in the same nine provinces. For the surveys of urban local households and rural-urban migrant households, fifteen cities were selected. For the rural household survey, 80 counties and 800 villages were selected. The samples contained 8,000 rural households, 5,000 urban local households, and 5,000 rural-urban migrant households. As in the 2002 surveys, the 2007 surveys of rural households and urban local households took subsamples from the national household survey of the NBS, whereas the rural-urban migrant survey was conducted separately. To ensure comparability between the 2002 and 2007 surveys, our analysis is confined to the nine common provinces: Hebei, Jiangsu, Zhejiang, Anhui, Henan, Hubei, Guangdong, Chongqing, and Sichuan. The questionnaires for the 2007 surveys included as many of the questions contained in the 2002 surveys as possible. In addition, some new questions on migration status and behavior were added to analyze the migration. The two rural-urban migrant surveys employed different sampling methods. In 2007 a migrant household was selected when one of its working members was drawn from his or her workplace, whereas in 2002 migrant households were drawn from residential communities. As a result, the 2002 survey has a higher proportion of self-employed migrants. As migrants living in communities tend to have higher incomes than do those living elsewhere, this might also produce some upward bias in the 2002 migrant wages. The best way to correct for this bias was to standardize based on housing: we selected only those 2007 migrants whose living conditions corresponded to those of the 2002 migrants. In both years, we included only migrants owning or renting their housing and not those living in dormitories or temporary shelters. The 2007 sample is effectively confined to migrants who have been in the city for at least six months and who live in households. Thus, we again exclude short-term migrants who are likely to regard their households as being in the village. We therefore cover only the fairly settled migrants in each year but as far as possible we are comparing like with like.1 1

More discussion on the rural-urban migrant samples in 2002 and 2007 is provided in Chapter 1 and in Appendix I and Appendix II. Note that our approach to delineating the migrant sample differs from that mentioned in Chapter 1 and the Appendices, which is based on the concept of “long-term, stable migrants” and is designed to address potential

236

John Knight, Deng Quheng, and Li Shi

IV. Migrant Wage Behavior The CHIP surveys are a potentially valuable source of information on migrant wages. Our analysis takes two forms. First, we explore the determinants of migrant wages in the 2007 survey. This analysis helps us examine the role that market forces play in migrant wage determination. Second, we combine the two surveys to examine the behavior of migrant real wages over the five years from 2002 to 2007. The purpose is to understand not only whether real wages rose but also, if that was the case, why it was so. Both the migrant and permanent urban resident questionnaires in the 2007 survey contained questions on monthly wage income and on net income from self-employment. We achieve income comparability across cities by means of the purchasing power parity (PPP)–adjusted deflator, as calculated at the province level by Brandt and Holz (2006). It is possible to show the influence of each city’s hukou worker income on migrant income. We do so by predicting the income that each migrant – with his or her particular characteristics – would have received if he or she had been rewarded according to the relevant city income function. This variable can be interpreted as a proxy for that city’s labor demand. With a perfectly elastic supply curve of migrant labor to any particular city and a segmented labor market within the city, the wages paid to permanent residents of the city have no effect on the market wages of migrants. However, if migrant wages are responsive to city wages, this might reflect competition for jobs between migrants and city residents (i.e., incomplete segmentation) or institutional wage determination that extends also to at least some of the migrants. There is information on the unskilled day wages in the migrants’ villages and the income that the migrants reported that they would have received had they remained in their villages. These variables serve as proxies for the migrants’ supply price. The proxies for migrant labor supply and demand can be helpful in interpreting migrant wage behavior. Consider a simple supply and demand model, bearing in mind that migrants and urban workers are imperfect substitutes (Knight and Yueh 2009). A rightward shift of the demand curve elicits a small supply response in the short run, owing to informational lags, inertia, and transaction costs. We expect the migrant wage will rise and marginal employees will enjoy a wage rent. In the long run, supply responds, the marginal rent is eliminated, and the equilibrium wage is determined by double counting of migrants when the rural and migrant CHIP samples are combined. A different approach is warranted in this chapter, as our analyses do not combine the rural and migrant samples.

The Evolution of the Migrant Labor Market in China, 2002–2007

237

the elasticities of the supply and demand curves; if the migrant supply curve is perfectly elastic, the wage in equilibrium returns to its initial level. If the labor supply curve is not perfectly elastic, we expect the proxy for city labor demand to exhibit a positive coefficient, not only in the short run but also in the long run. If instead the market shock is due to an upward (or leftward) shift of the supply curve, the wage rises only a little in the short term if the supply response is lagged, and indeed there may be a negative marginal rent. With time, the equilibrium wage rises further, and by the full amount of the supply shock if the supply curve is perfectly elastic. In that case, our proxy for labor demand does not influence the equilibrium wage. The relative importance of the proxies for supply and demand thus provides a pointer to the market forces influencing migrant wages. If our proxy for migrant labor demand has a relatively high coefficient, it suggests that demand is important in the determination of the wage level and of wage increases. If our proxy for the migrant supply price has a relatively high coefficient, it is likely that supply conditions are more influential in governing migrant wage behavior. However, caution is required because our cross-sectional data cannot deal with lags or distinguish equilibrium and disequilibrium situations. Table 6.2 presents estimates of the functions for migrant wage income and for migrant self-employment income in 2007, both variables in log form. The variables representing the migrant supply price have significantly positive coefficients: 0.161 for the reported opportunity cost and 0.046 for the village unskilled wage rate. Owing to possible colinearity between these variables, we also estimated the coefficient on opportunity cost when the unskilled wage is excluded from the specification (the final row in the table): the effect was a small rise in the coefficient, to 0.165. When the function was estimated with income expressed in levels and not logs (estimates not reported), this coefficient implied that an increase of 100 yuan in opportunity cost would alter migrant behavior in such a way as to raise the migrant wage by a significant 33 yuan. Precisely equivalent exercises for self-employment income showed the rural supply price to have larger effects (0.197 for the opportunity cost and 0.173 for the village unskilled wage, both significant). When the latter variable was excluded from the equation, the coefficient on opportunity cost implied that migrants with a rural supply price that was higher by 100 yuan would earn self-employment income in the city that was higher by 73 yuan. The evidence suggests that migrants with higher village opportunity costs will be found only in city jobs that pay more. The implication is that a rise in the rural supply price will indeed result in higher migrant wages.

238

John Knight, Deng Quheng, and Li Shi

Table 6.2. The determinants of migrant log wage income and log self-employment income, 2007 Mean value Wage ln-income if stayed in 6.277 village ln-village unskilled 6.958 wage ln-predicted city wage 7.107 Education (years) 9.522 Average performance 0.656 in school Poor performance in 0.077 school Possession of training 0.267 City experience 6.366 (years) City experience 73.218 squared Male 0.554 Manufacturing sector 0.263 Construction sector 0.072 Constant term Adjusted R-squared Observations Mean of dependent 7.007 variable Income if stayed in 0.165*** village (when village unskilled wage is omitted)

Self-empl. income

Coefficient Wage

Self-empl. income

6.233

0.161***

0.197***

6.977

0.046**

0.173***

7.333 8.431 0.710

0.086*** 0.020*** −0.021

0.074

−0.038

−0.006 0.004 0.066 0.070

0.148 10.024

0.037* 0.024***

0.096* 0.022***

141.523

−0.001***

−0.001***

0.646 0.038 0.022

0.102*** 0.063*** 0.165*** 4.714*** 0.212 2,026

0.173*** 0.158 0.237* 4.677*** 0.098 980

7.362 0.215***

Notes: The sample is confined to migrants who rented a house or owned a house in the city. The omitted categories in the dummy-variable analyses are good performance in school, no training, female, and “other” sectors. Certain explanatory variables relating to the employer, including firm size, contract type, and ownership type, were eliminated because their coefficients were found to be generally small and insignificant. The “predicted city wage” is the wage predicted for each migrant based on his or her individual characteristics and the city wage (or self-employment income) function estimated for the sample of urban-hukou residents. Nominal wages and incomes are corrected for provincial variations in the cost of living, by means of the PPP-adjusted price indices calculated by Brandt and Holz (2006). Statistical significance at the 1 percent, 5 percent, and 10 percent levels are denoted by ***, **, and * respectively. Source: 2007 CHIP national household survey, rural-urban migrant subsample.

The Evolution of the Migrant Labor Market in China, 2002–2007

239

The predicted migrant city wages were introduced as a potential proxy for pressures of demand for labor in the city. The coefficient for wage earners is positive (0.086) and significant but lower than the coefficient on the proxy for the migrant opportunity cost (0.165). This might reflect influences other than urban demand. The effect of variations among the cities in the cost of living in principle should be eliminated by our use of the PPP-adjusted deflator, but the provincial-level deflator has limitations, as acknowledged by its compilers (Brandt and Holz 2006: 83), and inaccuracy for a particular city within a province cannot be ruled out. Wages might be affected by institutional factors – because migrants are concentrated at the lower end of the city wage distribution – in particular by implementation of city minimum wage regulations. Therefore, it is relevant that the coefficient is not positive or significant for the self-employed (–0.006). We conducted robustness tests on our proxy for the urban demand for migrants. We tried replacing the predicted city wage/income variable with two alternative proxies: the average wage/income of urban residents in the city with no more than a junior middle-school education, and the urban wage/income of urban residents in the city, weighted by the occupational composition of migrants employed in the city. Whichever proxy was used, the coefficient on the predicted city income of self-employed migrants was small and not significantly different from zero. However, in the case of the predicted migrant wage, the occupation-based proxy had a coefficient of 0.148 and the education-based proxy had a coefficient of 0.300, both significant at the 1 percent level. Our evidence is therefore mixed: according to the proxy chosen, the demand side of the city labor market for migrants (coefficient varying from 0.086 to 0.300) might be more or less important than the supply side (coefficient varying from 0.046 to 0.165). Several control variables – interesting in themselves – are also included in the migrant income functions: we briefly discuss those that have both significant and substantive coefficients. The return to a year of education is positive and significant but low (2.0 percent per annum) in wage employment, and the wage is insensitive to the reported performance in school. These results might reflect the low quality of jobs that migrants generally take. The education variables are not significant at all in the self-employment equation. The possession of training, however, is rewarded both in wage employment and in self-employment. Similarly, city employment experience (years since migrating) has the usual inverted-U-shaped relationship in both forms of migrant employment. The fact that men and construction workers receive more wage income and self-employment income than women or workers in

240

John Knight, Deng Quheng, and Li Shi

the residual sectors (mainly sales and other services) is consistent with the arduous or unpleasant nature of some of the work performed by migrants and, in the case of self-employment, with the possibility of skill or capital barriers to entering certain activities. Table 6.3 combines the 2007 migrant survey with the 2002 migrant survey in order to examine the change in the logarithm of the wage over time. Sampling procedures were different in the two surveys: the 2002 sample was drawn from residential areas and thus contains only migrants living in households, whereas the 2007 sample was obtained by tracking all ruralurban migrants working in randomly selected areas. Because some of the latter were living in dormitories or workplaces provided by the employer, the coverage is broader. For comparability, we included 2007 migrants in the analysis only if they were living in their own houses or houses that they had rented. The Brandt-Holz (2006) PPP-adjusted deflator is used to correct both for differences in city price levels and for their rates of change. The specifications differ from those in Table 6.2. The key variable is the year dummy, with 2007 taking a value equal to 1 and 2002 taking a value equal to 0. Columns 1 and 5, both including only this dummy and an intercept term, show the raw increase in migrant real income: implying growth of 10.4 percent and 12.7 percent per annum for wage- and self-employment income respectively. Columns 2 and 6 add to this specification by introducing the set of individual variables available in both years. It is noteworthy that the proportionate increases in wage- and self-employment income fall only a little, to 9.7 percent and 12.1 percent per annum respectively, when personal characteristics are held constant. This represents the income change for migrants whose characteristics make them likely to be among the least skilled. We also standardize the urban predicted wage in columns 3 and 7: the increases come down further, to 8.9 percent and 11.7 percent respectively. Our best indicator of the rural supply price is the income that the migrant would have earned in the village: its addition, in columns 4 and 8, reduces the increases to 6.1 percent and 8.5 percent respectively. Nevertheless, there remains a substantial rise in wages and self-employment incomes that cannot be accounted for by the explanatory variables at our disposal. It is possible that changes in the supply of and demand for different worker characteristics altered the migrant wage structure. In particular, if there was a growing scarcity of young and educated migrants, this might have provided them with larger wage increases. We explore this possibility by distinguishing “young” (up to thirty-four years of age) and “old” (thirtyfive years of age and older) workers, and workers who were “more educated”

Table 6.3. The determinants of the proportionate change in the migrant wage and self-employment income, 2002–2007 Migrant Wage

241

Year 2007 Education (years) City experience (years) City experience squared Possession of training Male Manufacturing sector Construction sector Urban predicted wage Wage if stayed in village Constant Observations Adjusted R-squared

Self-employment income

1

2

3

4

5

6

7

8

0.643***

0.589*** 0.042*** 0.025*** −0.001*** 0.075*** 0.212*** 0.120*** 0.086***

0.342*** 0.021*** 0.023*** −0.001*** 0.050*** 0.148*** 0.096*** 0.099*** 0.085***

0.819***

0.771*** 0.032*** 0.040*** −0.002*** 0.066* 0.168*** 0.363*** 0.208***

0.737*** 0.032*** 0.038*** −0.002*** 0.078** 0.159*** 0.325*** 0.215*** 0.036***

6.362*** 3254 0.302

5.733*** 3254 0.409

0.531*** 0.030*** 0.023*** −0.001*** 0.064*** 0.170*** 0.118*** 0.098*** 0.098*** 0.158*** 5.254*** 3254 0.418

4.648*** 3254 0.459

6.539*** 2478 0.290

6.093*** 2478 0.343

5.855*** 2478 0.344

0.506*** 0.019*** 0.038*** −0.002*** 0.081 0.125*** 0.275*** 0.199*** 0.041*** 0.186*** 5.026*** 2478 0.385

Notes: Columns 1 and 5 contain only the dummy variable year 2007 (with year 2002 the omitted category). Columns 2, 3, and 4 add progressively to column 1, as do columns 6, 7, and 8 to column 5. The same explanatory variables as those in Table 6.2 are included, except for performance in school and the unskilled wage in the village, which were not available for 2002. The omitted categories in the dummy variable analysis are female, no training, and “other” services. Significance at the 1 percent, 5 percent, and 10 percent levels is denoted by ***, **, and * respectively. Source: 2002 and 2007 CHIP national household surveys, rural-urban migrant subsamples.

242

John Knight, Deng Quheng, and Li Shi

(having completed junior middle school) and “less educated” (not having completed junior middle school). Accordingly, we reestimate the wage functions corresponding to columns 2 through 4 of Table 6.3, now excluding the years of education but including a young worker dummy variable, plus a “young worker × 2007” interaction term and a more educated worker dummy variable, plus a “more educated worker × 2007” interaction term. The hypothesis is that the coefficients on the interaction terms are positive. The estimates (not reported) show the coefficients on the interaction term for young workers to be significantly positive in each specification (ranging from 0.08 to 0.11). By contrast, the coefficients on the interaction term for more educated workers are not positive and indeed are significantly negative in two of the three specifications. Whereas the wage premium on migrant education fell, young workers gained relative to old workers over the five years. However, this does not necessarily indicate a growing scarcity of young migrants. A minimum wage was introduced in some cities in the mid-1990s, and in subsequent years its coverage was broadened to more cities and its level was raised (Du and Pan 2009). In principle, it applies to all wage employees including migrants. It is plausible that young, poorly educated migrants in particular, as the lowest-paid workers in the cities, benefited the most from this development. Over time, the average migrant worker could be expected to become more educated and to have been working in the city for a longer period: both education and work experience are productive characteristics that are rewarded by the market. A more direct way of measuring the contribution of a change in characteristics to migrant wage growth is by means of decomposition analysis – permitting changes in the coefficients as well as in the characteristics. A standard decomposition of the change in the average migrant wages between 2002 and 2007, summarized in Table 6.4, shows that of the gross mean log wage increase (0.649), a minority (less than 30 percent) is due to differences in the coefficients of the two wage functions and a majority can be explained by changes in the mean characteristics. However, less than 5 percent is due to an improvement in the educational composition of the migrants, and there is no contribution due to a change in the length of their city experience. The main contributions come from the increase in the city demand price (32 percent or 42 percent, according to the weights being used) and the rural supply price (32 percent or 35 percent), both adjusted for price changes and for differences in provincial price levels. Labor-market forces were indeed largely responsible for the wage increase. The pattern is very similar for self-employment income, also shown in the table.

The Evolution of the Migrant Labor Market in China, 2002–2007

243

Table 6.4. Decomposition of the increase in the average real migrant wage, 2002–2007: Selective summary. Contribution of change in the mean characteristics to the gross mean wage increase: Percentage Wage

Self-employment income

2002 weights 2007 weights 2002 weights 2007 weights Education Length of city experience Predicted log city wage Log income if stayed in village Other Total

3.3 −0.4 31.6 35.4 0.4 70.3

4.1 −0.4 42.0 32.2 4.8 82.7

1.3 −0.5 8.0 36.2 −0.3 44.7

1.9 −0.9 30.3 26.4 2.9 60.6

Notes: The estimates are based on a standard Oaxaca-Blinder decomposition, using the coefficients for 2002 and 2007 as weights. The contribution of education as a whole is based on the change in composition among four levels: primary, junior middle school, senior middle school, and college education. The contribution of length of city experience is based on the change in composition among five experience groups: 0–5, 6–10, 11–15, 16–20, and 21+ years. The omitted categories in the dummy-variable analyses are the same as those in Tables 6.2 and 6.3, plus primary education and 0–5 years of city experience. The income if stayed in the village and the predicted city wage are as used in Tables 6.2 and 6.3. Source: 2002 and 2007 CHIP surveys.

To summarize what can be learned from these wage regressions: in Table 6.2 our proxies for rural supply (the rural opportunity cost) and urban demand (the employers’ valuation of the migrants) were indeed associated with a higher migrant wage. The rural proxy had a similar effect in the case of self-employment income, whereas the urban proxy did not. There was only a slight reward for education, probably reflecting the fact that most migrants perform menial jobs. We saw in Table 6.3 that the proportionate increase in the migrant real wage/income during the period from 2002 to 2007, both actual and standardizing for personal characteristics, was rapid. Table 6.4 indicates that the two variables most likely to reflect the contribution made by market forces to migrant wage behavior over time – the proxies for rural supply price and urban demand price – could together account for about two-thirds of the actual increase in migrant wages. The CHIP surveys provide some evidence suggesting that the market for migrants is becoming more integrated spatially. Table 6.5 reports the dispersion of the average city migrant wage for the seven cities that are common to the two surveys, the twenty-three cities in the nine common provinces, and for all cities in each survey. In the first of these cases, the Gini coefficient of the average city wage fell from 0.167 to 0.067, and the standard deviation of the log wage fell from 0.323 to 0.129. A similar dramatic

244

John Knight, Deng Quheng, and Li Shi

Table 6.5. Dispersion of migrant average city wage across cities, 2002 and 2007

Gini coefficient Standard deviation of log wage Standard deviation of wage

Common cities

Cities in common provinces

2002

2007

2002

2007

2002

2007

0.167

0.067

0.203

0.103

0.260

0.261

0.323

0.129

0.441

0.194

0.508

0.194

75

134

85

165

132

170

All cities

Source: 2002 and 2007 CHIP migrant samples.

reduction can be found for all the cities in the two surveys and for all twenty-three cities in the seven common provinces, as well as for the migrant self-employment income. However, both of these measures of dispersion are mean-dependent – falling as the mean increases, other things being equal – and the mean wage rose over the period. The standard deviation of the average real wage rose in each case. It is not clear which is the more appropriate measure of wage dispersion, but we assume that the sources of the wage differences, and their costs, are likely to rise along with incomes. On that basis, these results suggest that either minimum wages became more standardized across cities and more effective or, more likely, market forces were responding to the growing spatial mobility of migrants. Finally, using the CHIP urban and migrant surveys, we note that the average rural hukou wage in urban China was 70 percent of the average urban hukou wage in 2002, but it fell to 63 percent in 2007. Thus, migrant wages rose less rapidly than the wages of urban workers, although part of this was due to the changing returns to education – rising in the case of urban workers and falling in the case of migrants.

V. The Pool of Potential Migrants Our main concern in this section is to gauge the size of the pool of rural labor available to migrate to urban employment. Our method is to estimate the migration functions using the CHIP rural subsamples for 2002 and 2007, and then to assess how many nonmigrants have high probabilities of migration. Our cutoff probability in the probits is chosen to ensure that

The Evolution of the Migrant Labor Market in China, 2002–2007

245

Table 6.6. Probit equations predicting the probability of migrant status, 2002 and 2007 2002

Male Married without children With children ages 0−6 With children ages 7−12 With children ages 13+ A parent older than 70 Age group 21–25 26–30 31−35 36−40 41−45 46−50 51+ Schooling: Junior middle school Senior middle school College Health: Good Poor Arable land per household member Propn migrants in village Pseudo-R-squared Number of observations

2007

Coefficient

Marginal

Coefficient

Marginal

0.552*** −0.457*** −0.513*** −0.540 −0.526*** 0.049 0.172*** 0.041 −0.116 −0.301*** −0.530*** −0.719*** −1.022*** 0.217*** 0.168*** 0.041 0.181*** −0.089 −0.043** 2.021*** 0.195 9,321

0.145 −0.101 −0.113 −0.122 −0.136 0.013 0.049 0.011 −0.030 −0.073 −0.116 −0.150 −0.196 0.058 0.047 0.011 0.046 −0.023 −0.012 0.541

0.456*** −0.337*** −0.401*** −0.365*** −0.413*** −0.130*** 0.111** −0.021 −0.437*** −0.737*** −1.051*** −1.443*** −1.853*** 0.081** 0.014 −0.097 0.072* −0.271** −0.046*** 1.493*** 0.289 16,094

0.119 −0.079 −0.094 −0.086 −0.108 −0.034 0.031 −0.006 −0.099 −0.152 −0.198 −0.214 −0.298 0.022 0.004 −0.025 0.019 −0.064 0.012 0.401

Notes: The omitted categories in the dummy-variable analysis are female, not married, no parent older than 70, 16–20 age group, primary schooling or none, normal health. The symbols ***, **, and * denote statistical significance at the 1 percent, 5 percent, and 10 percent levels, respectively. Province dummy variables are included in the specifications but are not reported. Source: 2002 and 2007 CHIP rural samples.

the number of rural workers who are predicted to migrate is set equal to the number of workers who do migrate. We use the nine provinces that are common to both surveys. In 2002 the proportion of workers who actually migrated was 23.4 percent, and in 2007, it was 27.3 percent. In 2002 14 percent of the nonmigrants were predicted to migrate, and 46 percent of the migrants were predicted not to migrate; the corresponding figures in 2007 were 13 percent and 36 percent. Table 6.6 reports the probit equations, the dependent variable being migrant status and the omitted category being nonmigrant status. Several of the coefficients are not only statistically significant but also economically

246

John Knight, Deng Quheng, and Li Shi

substantial. The marginals show the effect of a unit change in a variable on the probability of migration. They imply that being male increased that probability by 15 percentage points in 2002 and by 12 percentage points in 2007. Marriage reduces the probability of migration, especially if there are children. The probability peaked for the 21–25 age group in both years. It fell sharply after age twenty-five in 2002 and after age thirty-one in 2007, and thereafter it declined more sharply in 2007. This is surprising: we would expect the probability of older workers to rise as migrant labor becomes scarcer. Age has a greater effect on the probability of migration than any other personal characteristic. With primary education or below as the omitted category, the probability of migration after junior middle school is 6 and 2 percentage points higher in 2002 and 2007, respectively. Senior middle-school enrollment is not significant in 2007. Although it is significant in 2002, its marginal effect on the probability of migrating (5 percent) is smaller than that of junior middleschool enrollment. Consistent with the low returns to education reported in Table 6.3, education is not an important determinant of migration in 2002 and becomes even less important over the next five years. Good health increases migration in both years and poor health decreases migration in 2007. The greater the area of arable land per member possessed by the household, the less chance there is of members migrating. Province dummy variables are included but not reported: the province of rural residence is a notable determinant of migration. Of great importance is the proportion of migrants among workers in the village. The mean proportion is 0.13 in 2002 and 0.22 in 2007; the standard deviations are 0.10 and 0.14. A one-standard-deviation increase in this proportion raises the migration propensity by 5.2 and 5.5 percentage points respectively. This result has several possible interpretations. One is that migration from the village sets in train a process of cumulative causation as information and support networks increase and the monetary and psychological costs of migration and job searches fall. In that case, the many villages still with low proportions of migrants might be ripe to become future migration villages. What keeps the nonmigrants from migrating? The 2007 survey contains a specific question asking the reason. The distribution of the replies is shown in Table 6.7. Three reasons were stressed: being too old, being unable to find an outside job, and needing to care for the elderly or for children. Each of these might prove to be flexible in the face of a rising demand for migrant labor. Older workers and caregivers might well be willing to move if policy is revised to meet the changing circumstances, such that .

The Evolution of the Migrant Labor Market in China, 2002–2007

247

Table 6.7. Reasons given by nonmigrant workers for not migrating: Distribution of the replies and the relationship of the replies to the probability of migrant status Regression explaining the probability of migrating

Too old, under 40 Too old, 40 or over Sick or disabled Cannot find a job outside Care of the elderly or children Has a local business Other

Reason given (%)

Regression coefficient

17.3 7.3 3.2 22.6 26.0 10.4 13.3

−0.118*** 0.195*** 0.000 0.021* 0.021* 0.006 −0.006

Partial correlation coefficient −0.107*** 0.161*** 0.019* 0.019* 0.004 −0.020

Source: 2007 CHIP rural sample.

family migration and urban settlement are made easier. Workers will find it easier to obtain outside jobs if the demand for migrants grows, especially if migrant networks are strengthened in the process. The table also shows the results of an ordinary least squares (OLS) regression equation for nonmigrants in which the dependent variable is the estimated probability of migrating, derived from Table 6.6, and the reported coefficients are those for the dummy variables representing the different reasons for not migrating. The coefficients cannot be interpreted as denoting a causal effect: they are merely associations that indicate which subjective reasons for not migrating are associated with a high probability of migrating, as predicted by the objective variables reported in Table 6.6. The higher the positive value of a regression or partial correlation coefficient, the more closely the reason is associated with a high probability of migration. This suggests that such a reason is important in explaining why rural workers with a high potential to migrate fail to do so. We see that the highest regression and partial correlation coefficient is the one for workers aged over forty who reported that they are too old. Over and above the effect of actual age (which is already incorporated into the estimated migration probability), the perception of being too old appears to be important in deterring migration. It is an important issue whether such a perception will be adjusted in response to improving migration opportunities and migration policies. It was possible to use the probit estimates of Table 6.6 to predict the probability of migrating for each worker – whether in fact a migrant or a

248

John Knight, Deng Quheng, and Li Shi

Table 6.8. Frequency distribution of the number of migrants and nonmigrants by predicted probability of migrating, and “expected value” of migration by nonmigrants, 2002 and 2007 (million) “Expected value” of migration by Predicted Migrants Nonmigrants Migrants Nonmigrants nonmigrants probability range 2002 2002 2007 2007 2002 2007 0–0.1 0.1–0.2 0.2–0.3 0.3–0.4 0.4–0.5 0.5–0.6 0.6–0.7 0.7–0.8 0.8–1.0 Total Total with p > 0.3

7.8 14.6 19.4 20.4 18.1 15.8 12.1 7.5 1.3 117.0

153.3 104.6 57.8 30.9 19.5 14.2 8.6 3.6 0.6 393.1

8.7 11.3 13.8 14.0 17.4 19.6 23.2 21.4 11.3 140.7

77.4

185.9 72.0 41.5 26.4 19.1 14.2 11.0 7.8 2.2 380.1

7.7 15.7 14.5 10.8 8.8 7.8 5.6 2.7 0.5 74.1

9.3 10.8 10.4 9.2 8.6 7.8 7.2 5.9 2.0 71.2

80.7

Note: The methods of estimation are explained in the text. Source: 2002, 2007 CHIP rural samples.

nonmigrant – in both 2002 and 2007, and from that to calculate the frequency distributions of workers by predicted probability. These can be expressed in millions of workers by using estimates of the number of migrants and nonmigrants in the two years. Calculated on this basis, Table 6.8 and Figure 6.1 show that in both years there were more migrants than nonmigrants among those rural workers with a predicted probability of migrant status exceeding 0.5. The disparity was small in 2002 but it increased in 2007. There were many migrants (31 million in 2007) with a probability of between 0.3 and 0.5, indicating that migration was quite possible in that range of probabilities; there were even more nonmigrants (45 million). Indeed, there were over 80 million nonmigrants with a migration probability of 0.3 or higher. This figure is actually slightly higher than the 77 million in the same category in 2002. Another method of assessing the potential pool of migrants is to find the “expected value” of migration by nonmigrants, that is, to multiply the number of nonmigrants in each migration probability range by that probability (taken to be the mid-point of the range). These estimates are also shown in Table 6.8. The total expected value of migration is 74 million in 2002 and 71 million in 2007.

The Evolution of the Migrant Labor Market in China, 2002–2007

249

190

180 170 160 150 140

Number of persons, million

130 120 110 100 90

80 70 60 50 40

30 20 10 0 [0,0.1)

[0.1,0.2)

[0.2,03)

[0.3,0.4) [0.4,0.5) [0.5,0.6) [0.6,0.7) [0.7,0.8)

[0.8,1]

Predicted probability of migration 2002, migrants

2002, nonmigrants

2007, migrants

2007, nonmigrants

Figure 6.1. The Distribution of the Number of Migrants and Nonmigrants by the Probability of Migrating (Million).

Because age is such an important determinant of migration, it is interesting to distinguish “young” and “old” nonmigrants (the dividing line again being set at age thirty-five). In both years, 67 million young nonmigrants had a probability of migrating higher than 0.3 (most old nonmigrants had

250

John Knight, Deng Quheng, and Li Shi

probabilities lower than 0.3), and the expected value of migration by young nonmigrants fell over time from 44 million to 41 million. Our results are based on binary probit equations distinguishing migrants and nonmigrants. As a robustness test, we also estimated multinomial logit equations for the two years. The base category was farming and the alternatives were local nonfarming and migration. The determinants of local nonfarm employment and migration employment are similar, but education is more important and age is less important for local nonfarm activities. Local nonfarm employment is better rewarded than farming (Knight and Song 2005: chap. 8), and it might be more attractive than migration for those with access to full-time local employment. The question that we seek to answer concerns the choice between migrating or not migrating rather than between migrating or farming, but the number of rural workers available to migrate in the future is likely to depend inversely on how rapidly rural nonfarm employment grows. A different approach to examine the extent of the rural labor surplus is to measure the number of days that are actually worked in relation to the number of days available for work. Although the 2007 CHIP rural survey does not contain this information, its 2008 continuation panel does record the number of days worked. Rural workers were asked to state their main economic activity. For those who said they were farmers, the average number of days worked was 183 (of which only 25 days were not in farming), with 49 percent of the farmers working fewer than 200 days. The corresponding figures for all rural workers (including those who classified themselves as local nonfarm workers and migrant workers) were 226 days and 32 percent, respectively. Clearly, rural people who obtain nonfarm jobs are more fully employed than are farmers. Assume that 300 days in the year are available for work. On that basis, the amount of surplus labor is 39 percent in the case of farmers – the group from which most potential migrants are likely to be drawn – and 25 percent in the case of rural workers as a whole. Our various measures illuminate different aspects of the potential to migrate. However, whichever measure is considered, it appears that a substantial supply of migrants is still available in rural China. Moreover, the potential pool of migrants barely declined over the five years. In any case, there are two reasons why the probabilities of migration are likely to rise as the urban economy grows. Rural workers will have better opportunities to migrate for employment, and older workers in particular will have a stronger incentive to move with their families as central and local governments respond to the economic need for a more settled urban labor force.

The Evolution of the Migrant Labor Market in China, 2002–2007

251

VI. Conclusion We have produced evidence of simultaneous surplus labor in rural areas and rising rural migrant wages in urban areas. The two phenomena appear to be inconsistent with the hypothesis of the Lewis model, and yet they are both observed in China. Our interpretation of the puzzle is that there is segmentation in the labor market – the result of constraints on rural-urban labor migration (Knight and Song 1999: chaps. 8–9; 2005: chaps. 5–7). The institutional constraints create difficulties for migrants living in urban areas in terms of good and secure jobs, housing, and access to public services, and these difficulties deter or prevent migrant workers from bringing their families with them to the cities. This in turn makes many rural workers reluctant to leave their villages, at least for long periods. Although there is evidence that the Chinese market for migrant labor is becoming more integrated, it is possible that the two phenomena will continue to coexist for several years: there will not necessarily be a neat Lewis turning point in a country as large and as regulated as China. In their revision of the Lewis model, Ranis and Fei (1961) formally incorporate a turning stage that reflects a gradually rising marginal product of rural labor. We envisage an even longer turning stage – the result not only of rural-sector heterogeneity but also of China’s labor-market institutions. Nevertheless, with evidence of sharp increases in migrant real wages since 2007 and projections of continuing rapid growth of urban employment over the next decade, on the one hand, and stagnation and then decline in the labor force, on the other, the turning stage cannot be far off and might even have already begun (Knight, Deng, and Li 2011). We adduced evidence that migrant wages indeed rose in real terms over the 2002–2007 period and that migrant wages are sensitive to urban labor-market conditions and to rural supply prices. Much of the increase can be explained by rising rural household incomes, although it is not possible to distinguish the increases that were exogenous (such as the abolition of the agricultural taxes and the fees for basic education) and the increases that were endogenous to the migration process. We had expected that the increased migrant wage was partly due to the improving human capital of migrant workers – both their educational attainment and their urban work experience – but this effect turned out to be surprisingly small over the five years. Our analysis of the 2002 and 2007 CHIP rural surveys shows that there is a large pool of nonmigrants with fairly high probabilities of migrating. Much depends on how far the three main perceived reasons for not

252

John Knight, Deng Quheng, and Li Shi

migrating – being too old, needing to care for dependents, and failing to find migrant work – will fade as work opportunities for migrants improve and labor-market policies adjust endogenously. Future trends in the labor market are likely to encourage both the urban settlement of migrants and the weakening of the hukou system. As more of the skilled jobs become vacant and migrants accordingly move up the job ladder, there will be an economic imperative for their permanent settlement. Skills and the associated training costs necessitate long-term employment. The Chinese system of “floating” temporary migration increasingly will become economically inefficient. The solution to this problem adopted by employers in many countries has been to try to stabilize the labor force by improving the rewards for staying. If long service becomes economically more efficient, governments have an incentive to permit and encourage staying, employers have an incentive to reward staying, and migrants have an incentive to stay. Long service, in turn, encourages migrants to settle with their families. Long-term residence in the city leads to the adoption of urban attitudes and to the transfer of the migrants’ social reference groups from the village to the city (Knight and Gunatilaka 2010). This process may well give rise to feelings of relative deprivation in relation to residents with urban hukou. As more former peasants make the transition from migrant to proletarian, the pressures on Chinese central and local governments to treat them on a par with urban-born residents is likely to grow, and hukou privileges will likely erode. The general scarcity of unskilled labor is probably the most powerful market force to reduce Chinese income inequality – inequality that has increased inexorably during the period of economic reform. It is likely to be the main market mechanism for narrowing the still-widening income divide between rural and urban China. Rapidly rising returns to unskilled labor will also require a change in development strategy toward more skillintensive and technology-intensive economic activities, and this will require long-term planning and investment in human capital. There is little evidence that these changes are yet taking place, other than the remarkable expansion of higher education enrollments that has occurred since 1998. However, given the continued rapid growth of urban employment and the rapid demographic transition that has been predicted, it is likely that these changes will occur increasingly over the coming decade. References Brandt, L. and C.A. Holz (2006), “Spatial Price Differences in China: Estimates and Implications,” Economic Development and Cultural Change, 55(1), 43–86.

The Evolution of the Migrant Labor Market in China, 2002–2007

253

Cai, F., Y. Du, and C. Zhao (2007), “Regional Labour Market Integration since China’s WTO Entry: Evidence from Household-Level Data,” in R. Garnaut and L. Song, eds., China: Linking Markets for Growth, 133–150, Canberra: Asia Pacific Press. Du, Y. and W. Pan (2009), “Minimum Wage Regulation in China and Its Applications to Migrant Workers in the Urban Labor Market,” China and World Economy, 17(2), 79–93. Griffin, K. and R. Zhao, eds. (1993), The Distribution of Income in China, Basingstoke: Macmillan. Gustafsson, B.A., S. Li, and T. Sicular, eds. (2008), Inequality and Public Policy in China, New York: Cambridge University Press. Knight, J., Q. Deng, and S. Li (2011), “The Puzzle of Migrant Labour Shortage and Rural Labour Surplus in China,” China Economic Review, 22(4), 585–600. Knight, J. and R. Gunatilaka (2010), “Great Expectations? The Subjective Well-being of Rural-Urban Migrants in China,” World Development, 38(1), 113–124. Knight, J. and S. Li (2005), “Wages, Firm Profitability and Labor Market Segmentation in Urban China,” China Economic Review, 16(3), 205–228. Knight, J. and L. Song (1999), The Rural-Urban Divide: Economic Disparities and Interactions in China, New York: Oxford University Press. Knight, J. and L. Song (2005), Towards a Labour Market in China, New York: Oxford University Press. Knight, J. and L. Yueh (2009), “Segmentation or Competition in China’s Urban Labour Market?” Cambridge Journal of Economics, 33(1), 79–94. Kwan, F. (2009), “Agricultural Labour and the Incidence of Surplus Labour: Experience from China during Reform,” Journal of Chinese Economic and Business Studies, 7(3), 341–361. Lewis, W.A. (1954), “Economic Development with Unlimited Supplies of Labour,” The Manchester School, 22(2), 139–191. Li, S. (2001), “Labor Migration and Income Distribution in Rural China,” in C. Riskin, R. Zhao, and S. Li, eds., China’s Retreat from Equality: Income Distribution and Economic Transition, 303–328, Armonk, NY: M. E. Sharpe. Li, S. and Q. Deng (2004), “Zhongguo chengzhen shiyel¨ude chongxin guji” (Reestimating the Unemployment Rate in Urban China), Jingjixue dongtai, no. 4, 44–47. Meng, X. and N. Bai (2007), “How Much Have the Wages of Unskilled Workers in China Increased? Data from Seven Factories in Guangdong,” in R. Garnaut and L. Song, eds., China: Linking Markets for Growth, 151–175, Canberra: Asia Pacific Press. Meng, X. and C. Manning, with S. Li and T.N. Effendi, eds. (2010), The Great Migration: Rural-Urban Migration in China and Indonesia, Cheltenham: Edward Elgar. Minami, R. and X. Ma (2009), “The Turning Point of Chinese Economy: Compared With the Japanese Experience,” Conference on Labor Market in the PRC and Its Adjustment to Global Financial Crisis, ADBI, Tokyo, June. National Bureau of Statistics (NBS) (2008), Zhongguo tongji nianjian 2008 (China Statistical Yearbook 2008), Beijing: Zhongguo tongji chubanshe. Park, A., F. Cai, and Y. Du (2010), “Can China Meet Its Employment Challenges?” in J.C. Oi, S. Rozelle, and X. Zhou, eds., Growing Pains: Tensions and Opportunity in China’s Transformation, 27–55, Stanford, CA: Walter H. Shorenstein Asia-Pacific Research Center and Baltimore: Brookings Institution. Ranis, G. and J. Fei (1961), “A Theory of Economic Development,” American Economic Review, 51(4), 533–565.

254

John Knight, Deng Quheng, and Li Shi

Riskin, C., R. Zhao, and S. Li, eds. (2001), China’s Retreat from Equality: Income Distribution and Economic Transition, Armonk, NY: M.E. Sharpe. Sheng, L. (2008), Liudong haishi qianyi: Zhongguo nongcun laodongli liudong guocheng de jingjixue fenxi (Floating or Migration? Economic Analysis of Floating Labor from Rural China), Shanghai: Shanghai yuandong chubanshe. Taylor, J.R. (1988), “Rural Employment Trends and the Legacy of Surplus Labour, 1978– 86,” China Quarterly, no.116, 736–766. Wang, D. (2008), “Liuyisi zhuanzhe dian yu Zhongguo jingyan” (Lewisian Turning Point: Chinese Experience), in F. Cai, ed., Zhongguo renkou yu laodong wenti de baogao 9 (Report on Chinese Population and Growth No. 9), 88–103, Beijing: Shehui kexue wenxian chubanshe. Xin, D. and J. Shan (2010), “China Has ‘Sufficient Labor Pool for Next 40 Years,’” China Daily, March 27, 2010, http://www.chinadaily.com.cn/cndy/2010-03/ 27/content 9651006.htm. Accessed August 24, 2011.

SEVEN

A New Episode of Increased Urban Income Inequality in China Deng Quheng and Bj¨orn Gustafsson

I. Introduction The development of income inequality in urban China is a hot topic. There is agreement that income inequality has tended to increase over the years, but evidence indicates that the development has not been smooth. For example, previous studies based on the China Household Income Project (CHIP) have found that earnings inequality at the individual level as well as income inequality at the household level in urban China increased profoundly from 1988 to 1995. However, although from 1995 to 2002 earnings inequality continued to increase, income inequality at the household level decreased modestly (Gustafsson, Li, and Sicular 2008). Rapid growth in incomes caused urban poverty, assessed by a poverty line representing constant purchasing power (“absolute poverty”), to diminish rather substantially (Appleton, Song, and Xia 2010). What has happened more recently, during the initial phase of the Hu Jintao–Wen Jiabao leadership (2002–2007)? In this chapter we aim to shed new light on developments during this period using data from the CHIP urban household survey. Our first research question is, How did income, income inequality, and poverty develop? To answer this question, we show income growth curves and report estimates of income inequality. Furthermore, we show cumulative density functions and report summary measures on absolute and relative poverty for 1988, 1995, 2002, and 2007. The second research question is, What were the forces for change during the period from 2002 to 2007? To understand this, we decompose the Gini coefficient of disposable household per capita income by income components for 2002 and 2007. The third research question is, How have various categories of the population fared during the period from 2002 to 2007? To answer this question

255

256

Deng Quheng and Bj¨orn Gustafsson

we look at differences among groups based on ownership, sector, age, and education. One major finding is that the period between 2002 and 2007 was characterized by a new episode of increased income inequality in urban China. However, if measured by summary indices such as the Gini coefficient, the increase was not as rapid as the increase between 1988 and 1995. Poverty among urban residents assessed by various poverty lines expressed in constant purchasing power decreased. It is also true, however, that a slightly larger proportion of urban residents were relatively poor; that is, they had per capita incomes falling below a relative poverty line defined as a fixed percentage of the median income. We find two sources to be the most important contributors to the increase in inequality between 2002 and 2007 – the rather rapid growth of business income (income from self-employment and from owning a private business) and the rapid growth of imputed rent from owner-occupied housing. These sources originated from policy changes introduced during the pre–Hu-Wen leadership period. We find substantial differences in a household’s economic situation across cities. China’s urban poverty problem is disproportionally concentrated in low-income cities and affluent households are more prevalent in high-income cities. China’s children grow up in households with rather different economic situations. There is also a wide variation in economic well-being among the elderly in urban China. There are many aspects of urban inequality in China, and we do not study all of them in this chapter. Following many other studies, our analysis concentrates on formal urban residents. In other words, we leave aside the important issue of how rural migrants are faring and how their increased number has contributed to the development of inequality among all persons and households living in urban China. We also note that our focus is on how individuals living in households and sharing income with other household members are faring, whereas other studies in this volume analyze inequality in workers’ earnings and wages (see, e.g., Chapter 9 in this volume). Although these two aspects of urban inequality are strongly related, they are not the same. This becomes apparent in Chapter 8 in this volume, which shows that redistribution within Chinese urban households to a large extent has counteracted impulses toward increased inequality due to increased unemployment and other forms of nonwork. Furthermore, we implicitly assume that resources within households are equally shared, an assumption that might not be correct in all cases. Yet it is rather difficult to replace this with another assumption due to the lack of information on intra-household allocation in the CHIP survey data. Finally, although our

A New Episode of Increased Urban Income Inequality in China

257

focus is on the distribution of income, income obviously is not the only indicator of well-being; our analysis is complementary to parallel studies that focus on other welfare indicators, such as education and health (see Chapter 4 on educational inequality). In the next section, we provide some background information on how changes in urban China during the 2002–2007 period are relevant to our research questions. Section III presents the data and definitions of some of the key variables. Section IV examines overall developments, and Section V analyzes the decomposition of the Gini coefficient by income components. Section VI describes how various categories of persons have fared, and the chapter concludes with a summary of our findings.

II. Background During the period from 2002 to 2007, the Chinese economy continued to grow at an astonishing rate – gross domestic product (GDP) rose by 82 percent. Many processes contributed to this development, affecting changes in the composition of the affected groups in the population. For example, the proportion of young children decreased, whereas the proportion of elderly increased. We discuss those changes considered to have had the most effect on the development of income inequality. Change in the types of work units in which Chinese households earn their incomes has been considerable. In the past, almost all economic activities in urban China took place in state-owned units (including state-owned enterprises [SOEs]) or collective units. During the second half of the 1990s, central policies promoted diversified ownership, allowed ineffective work units to go bankrupt, and abolished permanent job tenure. This led to many job losses as the aggregate number of those employed in state-owned and collective units declined from 140 million in 1995 to 80 million in 2002, an enormous loss of 60 million jobs, or 8.6 million jobs per year (National Bureau of Statistics [NBS] various years). As a consequence, an employment problem of unprecedented magnitude became a strong stimulus for the increased income inequality (see also Cai, Chen, and Zhou 2010). Although jobs in state-owned and collective units continued to decrease from 2002 to 2007, the reduction slowed to 1.8 million per year; in 2007, 64 million workers were employed in SOEs and 7 million in collective units. The downsizing and restructuring of the state and collective sector was counteracted by the growth of the private sector (see, e.g., Chen, Li, and Matlay 2006; Dickson 2008; Haggard and Huang 2008; and H. Li et al. 2008). From the second half of the 1950s and until 1978, the social and

258

Deng Quheng and Bj¨orn Gustafsson

political environment allowed little room for the development of either private enterprises or self-employment. Private enterprises were not officially recognized until April 1988 when China issued provisional regulations on private enterprises. The regulations gave legal status to privately owned firms that employed eight or more workers (called siying qiye). However, adoption of the regulations did not immediately change the environment for private business. For example, private entrepreneurs faced, and still face, problems of accessing credit via formal channels. Furthermore, complex rules govern private enterprise activities and owners must spend considerable time and resources interacting with bureaucrats. Most observers agree, however, that opportunities for operating private enterprises have improved. An indication of their increased acceptance is that at the 2002 Sixteenth National Congress of the Communist Party of China, the constitution was amended to allow private owners to become members of the Communist Party. In order to legally run a business as a private owner, one must register with the State Administration for Industry and Commerce at different levels. Official statistics show a growing number of registered private enterprises after the 1988 change in legal status. The number of private businesses was 139,000 in 1991, over 2 million in 2002, and as many as 5.5 million in 2007 (Zhongguo siying jingji nianjian 2009). Measured by the scale of their operations, private enterprises are rather heterogeneous. There are many small firms (e.g., in the retail and service sectors) and a few large units in, for example, manufacturing and mining. Thus, one would expect the earnings of private owners to be rather unequally distributed. Among private firms, in 2002 there were 20 million employees and 4.2 million employers; by 2007 the numbers had grown to 46 million employees and 9.8 million employers. Another part of the private sector is made up of the self-employed (see, e.g., Yueh 2009). During the period of the planned economy, SOEs provided stable employment, heavily subsidized housing and health care, as well as old-age security. Self-employment was illegal and politically dangerous. However, as the urban reforms proceeded and jobs disappeared and the various benefits and subsidies were phased out, the incentives to become self-employed increased. Particularly during the early stages of the reform process, switching to self-employment was an attractive alternative for low-skilled workers who risked being laid off. More recently, a substantial number of skilled workers and professionals have also moved into self-employment. The number of self-employed increased from 23 million in 2002 to 33 million in 2007 (NBS various years). This means that in 2007

A New Episode of Increased Urban Income Inequality in China

259

the number of persons engaged in the private sector (employees and owners as well as the self-employed) reached 79 million, a number higher than the 71 million who were employed in state or collective enterprises. Still, the latter number is larger than the 46 million employed in private firms. The expansion of the private sector means that business income, defined here as income from self-employment or from being an owner of a private business, expanded rapidly from its low base during the first phase of the Hu-Wen leadership. In Section V we report that during these years, business income increased more rapidly than total income. We also report that during the period under study property income increased more rapidly than total income. However, property income still constitutes a rather small proportion of the total income of Chinese households. Although enterprise and property income increased rapidly during the initial years of the Hu-Wen leadership, wages from working in an SOE or in a privately owned unit were still the primary sources of income. But wage earnings increased less rapidly than many other sources of income. We report that the share of wage earnings in total income actually fell. How much a specific household earned in wages depends on various household circumstances. These include changes in the household’s labor supply, with a long-run trend of fewer adult persons earning income from work, changes in wage rates due to changed methods for setting wages, changing demand, and changing supply. Regarding the latter, the increased number of ruralurban migrants, who most often are low skilled, presumably negatively affected the wages of low-skilled workers. Moreover, the expansion of higher education presumably exerted downward pressure on the wages of highly skilled workers. Chapter 9 in this book examines in more detail changes in wage inequality in urban China in the 2000s. In prereform China, an overwhelming majority of households were allocated low-rent housing, that is, they received large housing subsidies. Due to the various types of housing reform that proceeded at different speeds in different locations, by 2002 most housing in urban China had been privatized (see Chapter 3). The privatization followed a pattern by which the tenants were given an opportunity to buy the apartment where they were living at a price lower than the market price. The resulting wealth transfers were typically larger for better-off workers because these workers generally had been allocated larger apartments in better locations (Logan, Fang, and Zhang 2010). For this reason, and due to the transactions on the emerging housing market, one can assume that imputed rents from owner-occupied housing are positively related to household income.

260

Deng Quheng and Bj¨orn Gustafsson

No recent visitor to urban China can fail to note the intense construction activity taking place. During the first phase of the Hu-Wen leadership, the housing stock increased rapidly. Furthermore, housing demand increased rapidly as well. Many people had accumulated savings enabling them to afford housing and, at the same time, access to loans increased. One essential part of the picture is that urban residents typically expect future income increases. Furthermore, the rapidly increasing housing prices led to expectations of further price increases, making urban residents more inclined to invest in the housing market, thus feeding price increases even at the risk of creating price bubbles. We observe that housing prices in urban China increased rather rapidly during the initial phase of the Hu-Wen leadership (see Chapter 3). We report that the rental value of owner-occupied housing, on average, increased almost twice as rapidly as total household income. In urban China, a very large proportion of women over the age of fifty-five and men over the age of sixty receive pensions as former SOE, government, or collective employees. Few of the elderly work for wages; however, many live with their grown and economically active child and his or her spouse, and others live alone with their spouse and receive pensions as their dominant source of income (for the situation of the elderly during the Mao period see Davis-Friedmann [1991]; and for an analysis of income among the aged using CHIP data from 1988, 1995, and 2002, see Palmer and Deng [2008]). Pension payments are linked to work histories; from the perspective of Western observers, income replacement rates are considered to be high. An overwhelming proportion of all retirees has long work histories and thus they have substantial pension incomes. Many retirees with limited means have enjoyed increased real income as the minimum enterpriseemployee pension increased from 714 yuan per month in 2005 to 963 yuan per month in 2007 (908 yuan per month in 2005 yuan). With their long work careers leading to relatively large apartments, many of the elderly enjoy imputed rents from owner-occupied housing. On the whole, China’s older urban population has a living standard not significantly different from that enjoyed by the working population. Many of the situations that were described earlier have increased income inequality at the household level. However, most likely other forces are also at work. For example, rapidly increased incomes have moved income earners into higher tax brackets. Although tax schedules have been reformed, the progressive tax system presumably counteracted those forces leading to higher income inequality. For an analysis of the distributional impact of personal income taxes in urban China, see Chapter 10.

A New Episode of Increased Urban Income Inequality in China

261

III. Data and Definitions For our analysis we use data from the 2002 and 2007 CHIP urban surveys. The 2002 urban data cover twelve provinces: Beijing, Shanxi, Liaoning, Jiangsu, Anhui, Henan, Hubei, Guangdong, Chongqing (in 1988 still a part of Sichuan), Sichuan, Yunnan, and Gansu. The 2007 data include these provinces as well as Shanghai, Zhejiang, Fujian, and Hunan. For comparisons with earlier periods, if possible, we use data for the same provinces from the 1988 and 1995 urban surveys (Sichuan was not surveyed in 1988). The 1988 survey is described by Eichen and Zhang (1993), and information on the 1995 and 2002 surveys is found in S. Li et al. (2008). Chapters 1 and 2 in this volume as well as Appendices 1 and 2 in this volume provide details on the 2007 survey. We define household income per capita to include earnings, pensions, business income, housing subsidies, imputed rents from owner-occupied housing, and income in-kind. Business income includes self-employed income as well as income accruing to private entrepreneurs. Our definition of household income also includes imputed rents from owner-occupied housing. Following the approach in several other chapters in this volume, we have used the market rent approach to estimate the imputed rental income from owner-occupied housing for 2002 and 2007 (see Chapter 3). As this alternative is not available for 1988 and 1995, for those years we follow the approach of Khan et al. (1993) and define imputed rent of owner-occupied housing as 8 percent of the net worth of owner-occupied housing (current replacement value minus the outstanding debt). Taxes and fees are treated as negative income. We introduce province weights based on the published NBS population data as discussed in Appendix II. The total household income is divided by the number of household members and is then ascribed to each household member, making individuals the unit of analysis. Income is measured in 2002 constant prices using the NBS urban consumer price index. This study differs from Chapter 8 in this volume in that our population includes children and the elderly. Following Brandt and Holz (2006), we also take into account spatial price differences.

IV. Overall Developments In this section we study the overall trends in household income and poverty from 1988 to 2007. Although developments up to 2002 have been reported in earlier writings, information on the 2002–2007 period is new. We start by comparing the income growth curves (Ravallion and Chen 2003) computed

Deng Quheng and Bj¨orn Gustafsson

0

Income Growth .05 .1

.15

262

0

20

40

60

80

100

Percentile 1988-1995 2002-2007

1995-2002

Figure 7.1. Income Growth Curves for the 1988–1995, 1995–2002, and 2002–2007 periods (annual income growth at various percentiles). Source: Authors’ computations from the CHIP.

for percentiles, as shown in Figure 7.1 for the three periods 1988–1995, 1995–2002, and 2002–2007. Several interesting observations are revealed. Positive growth is reported for almost all percentiles and for all three periods. The exception is the lowest nine percentiles for the 1988–1995 period. Income growth was generally fastest during the 2002–2007 period: the growth curve for this period is located entirely above the other two. Thus income growth of Chinese households accelerated during this first phase of the Hu-Wen leadership. For example, income growth at the median was 2.7 percent per annum during the first period, 4.8 percent during the second period, and an impressive 10.6 percent during the third period. Figure 7.1 also shows that during the most recent period income growth generally was fastest at the top of the income distribution and lowest at the bottom; the upward slope means that income inequality increased. However, the growth curve for the 2002–2007 period is less steep than the slope for the 1988–1995 period. In contrast, the growth curve for 1995– 2002 is relatively flat: sloping upward at the lower percentiles and sloping slightly downward at the higher percentiles. From an examination of the slope of the three curves, we can conclude that income inequality developed

A New Episode of Increased Urban Income Inequality in China

263

Table 7.1. Income inequality 1988, 1995, 2002, and 2007, according to various inequality indices

1988 1995 2002 2007

Mean income

Median income

Gini

MLD

Theil index

Proportion having income above 200 percent of median income Percentage

4,520 6,037 8,078 13,796

4,173 5,034 6,993 11,593

0.2104 0.3340 0.3039 0.3229

0.0726 0.1931 0.1554 0.1790

0.0768 0.2422 0.1551 0.1753

3.60 8.80 6.08 6.82

Source: Authors’ computations using the CHIP data, in 2002 prices with adjustments for regional differences in living costs. MLD refers to mean logarithmic deviation.

differently during the three periods. The period between 1988 and 1995 was characterized by rapidly increasing income inequality, that between 1995 and 2002 witnessed few changes, and that between 2002 and 2007 represented a new period of increased income inequality. Table 7.1 provides estimates of three often-used income inequality indices, computed for 1988, 1995, 2002, and 2007. The indices reveal the same directional change in inequality as the growth curves, although the magnitude differs across the three indices. A period of rapid increases was followed by a small reduction and then by a new episode of increased income inequality. According to our estimates, in 2007 the Gini coefficient was 0.323, which by the standards of rich countries is not very high, but nor is it extremely low. Looking at the top of the distribution, we see that the proportion of individuals having a per capita income of at least 200 percent of the contemporary median (i.e., affluent persons) increased rapidly from 4 percent in 1988 to 9 percent in 1995, fell to 6 percent in 2002, and increased marginally to 7 percent in 2007. The rather rapid income growth at the lower part of the income distribution between 2002 and 2007 means that poverty, assessed by an absolute poverty line representing fixed purchasing power, decreased rapidly during the period. This is shown in Figure 7.2 where we report the cumulative density functions for 1988, 1995, 2002, and 2007. These curves show the cumulative proportion of individuals at each level of income. There is one curve for each year studied. In the figure we have drawn three alternative poverty lines, all expressed in constant purchasing power by using the consumer price index (CPI). Although this approach is used in several studies of changes in urban poverty in China (for a survey, see Riskin and Gao 2010), some analyses prefer a different approach (see Meng, Gregory,

Deng Quheng and Bj¨orn Gustafsson

0

Cumulative Popuplation Share .2 .4 .6 .8

1

264

0

1,761

3,522 5,283 Household Income per Capita (Yuan) 1988 2002

10,000

1995 2007

Figure 7.2. Cumulative Distribution of Income, 1988, 1995, 2002, and 2007. Source: Authors’ computations from the CHIP. Note: Income is expressed in 2002 prices using the spatial price index of Brandt and Holz (2006). For better visualization we have restricted the curves to income lower than 10,000 yuan.

and Wang [2005], who reestimate the cost of a basic needs poverty line for each year during the 1986–2000 period). There is no official poverty line for urban China, so in our analysis of poverty we use poverty lines based on the World Bank’s $1.25 purchasing power parity (PPP) per-person per-day standard. The lower poverty line in Figure 7.2 corresponds to the US$1.25 PPP per day. In 2002 prices, this was 1,761 yuan (Chen and Ravallion 2010). The second and third poverty lines correspond to two and three times this amount, respectively. Within that portion of the graph to the left of the poverty lines, the cumulative density function for 2007 is below that of 2002. Thus, we can conclude that poverty as assessed by these poverty lines has continued to decrease. We also note that although the decrease at the highest poverty line is substantial, at the lowest poverty line the decrease is not easy to detect, because by 2002 a very small proportion of urban residents fell below this low poverty line. In Table 7.2 we report the numerical values for the Foster, Greer, and Thorbecke (FGT; 1984) family of poverty index, computed for two “absolute” poverty lines in urban China for 1988, 1995, 2002, and 2007. For each poor unit this family of indices uses its normalized poverty gap, [(z − y )/z],

A New Episode of Increased Urban Income Inequality in China

265

Table 7.2. Absolute poverty in urban China, 1988, 1995, 2002, and 2007 FGT indices: 1,761 yuan as the poverty line

1988 1995 2002 2007

FGT(0), Poverty rate

FGT(1)

FGT(2)

0.0135 0.0269 0.0106 0.0014

0.0031 0.0062 0.0022 0.0004

0.0089 0.0027 0.0008 0.0002

FGT indices: 3,522 yuan as the poverty line 1988 1995 2002 2007

FGT(0), Poverty rate 0.3287 0.2439 0.1114 0.0241

FGT(1) 0.0648 0.0591 0.0261 0.0049

FGT(2) 0.0223 0.0228 0.0096 0.0016

Source: Authors’ computations from the CHIP.

which is a number indicating how far below the poverty line [z] the income falls on a scale bounded by 0 (in the case of no negative income) and 1. Those gaps are raised by a positive parameter before the average is taken and then multiplied with the headcount ratio. Higher numbers of the parameter give increasing weight to large poverty gaps, and thus greater “poverty aversion.” When the parameter is set to 0, the index collapses to the headcount ratio. Starting with the lowest line, the US$1.25 world poverty line, we see that the proportion of urban residents considered to be poor actually went up from 1 percent in 1988 to 3 percent in 1995, but thereafter fell to 1 percent in 2002 and was only 0.1 percent in 2007. However, when doubling the poverty line, not less than one-third of the urban residents were considered poor in 1988. The proportion thereafter decreased particularly rapidly between 1995 and 2002, reaching only 2 percent in 2007. The other two indices tell much the same story about the development of urban poverty. In a rapidly growing economy, does it make sense to assess the extent of poverty solely or predominantly against an “absolute” standard? There has been much debate on this issue during periods of growth in rich countries. For example, when Eurostat reports how many persons and households in the European Union are at risk of becoming poor, the assessment is made against a relative poverty line that is defined as a fixed percentage of the median income for the country where the person and household resides. For some years, a poverty line set at 60 percent of the median

266

Deng Quheng and Bj¨orn Gustafsson Table 7.3. Relative poverty in urban China, computed using various relative poverty lines, 1988, 1995, 2002, and 2007

Percentages of persons under various percentages of the median income 40% 50% 60% 70%

1988 percentage

1995 percentage

2002 percentage

2007 percentage

0.98 3.23 8.00 15.67

4.13 8.49 15.21 24.45

5.15 10.90 18.06 25.67

5.92 11.81 18.87 26.40

Source: Authors’ computations from the CHIP.

poverty line was used. A recent study on inequality and poverty in thirty rich countries uses the same approach (Organisation for Economic Cooperation and Development [OECD] 2008). In academic work on urban China, the approach of setting the poverty line at 50 percent of an urban location has been used. An early example is Wong (1995, 1997), in which the poverty line is defined as 50 percent of the median of the city under investigation (Guangzhou and Shanghai). Another example is Wang (2008) who, in a study of the 1986–2000 period, puts the poverty line at 50 percent of the median for urban areas in those provinces under investigation (Liaoning, Sichuan, and Guangdong) or, alternatively, at 50 percent of the median in the city where the person resided (in one of the three provinces covered in the study). Saunders (2007), in a international comparison of poverty among older people in urban China, uses a poverty line set to 50 percent of the median income for urban China. We follow this approach, putting the poverty line at 40, 50, 60, and 70 percent of the contemporary median income in urban China. The results are reported in Table 7.3. Table 7.3 shows that, for all alternatives applied, relative poverty in urban China has increased in all years under study. Whereas 8 percent of urban residents fell under a poverty line put at 60 percent of the median income in 1988, the proportion increased to 15 percent in 1995, to 18 percent in 2002, and to 19 percent in 2007. The latter number is within the range or above the average of similarly defined poverty rates for thirty OECD countries in the mid-2000s (OECD 2008). Note that when we compute the poverty rates, resources received by the households within the means-tested minimum living standard guarantee (dibao) program are considered. We can conclude that the expansion of the dibao program for urban residents from the mid-1990s and into the new millennium did not fully counteract the underlying increase in relative poverty.

A New Episode of Increased Urban Income Inequality in China

267

From the preceding two exercises conceptualizing and measuring poverty, we can conclude that China’s urban poverty record differs dramatically depending on the lens by which it is viewed. From a third-world perspective, China is a success story – in 2007 almost no one fell under the US$1.25 poverty line. However, seen through the lens of rich countries, the situation appears to be worrisome. Relative poverty rates in China are not low and urban poverty by this measure is not trivial. A similar conclusion follows from application of the Subjective Poverty Line approach to defining a poverty line for urban China. Gustafsson, Li, and Sato (2004) report poverty rates of 6 to 7 percent for a sample of twelve cities in 1999. Another concern is the secular upward trend in relative poverty; relative poverty rates in urban China have been rising steadily for as long as two decades.

V. How Changed Income Sources Have Affected Income Inequality In this section, by decomposing the Gini coefficient for total household income, as defined in Section III, we shed light on how income inequality has changed. The Gini coefficient can be written as the weighted sum of the concentration coefficients of the various income sources. The weights are the shares of the income source in the total per capita income. Thus we have  μk Ck, (1) G = μ k where μk and μ are the means of income source k and the total per capita income, respectively, and C k is the concentration coefficient of income source k. The concentration coefficient measures the association between income source k and the total per capita income, with values ranging from −1 to +1. If the concentration coefficient is negative, it means that lowincome earners are receiving larger amounts (in an absolute sense) than are high-income earners. Not only is the sign of the concentration coefficient of interest; its magnitude in comparison to the Gini coefficient is an indicator of the distributional profile of the income source. If the income source has a concentration coefficient that is equal to the value of the Gini coefficient of the total per capita income, the distribution of the income source is as equal as total per capita income. However, if the concentration coefficient of an income source is greater (or smaller) than the Gini coefficient of total per capita income, this income source is considered to be dis-equalizing (equalizing).

268

Deng Quheng and Bj¨orn Gustafsson

Table 7.4. Components and growth of household income per capita, 2002 and 2007 Household income per capita (yuan)

Earnings Pensions Imputed rents from owner-occupied housing Business income Property income Income in-kind Housing subsidies Net transfer income Total per capita income

Growth

2002

2007

Amount (yuan)

Annualized growth rate (%)

5,573.92 1,399.50 483.55

9,071.66 2,642.54 1,458.79

3,497.74 1,243.04 975.24

10.23 13.56 24.71

266.37 91.63 81.87 231.22 −49.70 8,078.37

985.65 209.81 88.40 86.74 −747.44 13,796.14

719.28 118.18 6.53 −144.48 −697.74 5,717.77

29.91 18.02 1.55 −17.81 71.97 11.30

Source: Authors’ computations from the CHIP. Amounts are in 2002 prices.

We define eight components of income and decompose the Gini coefficient for 2002 and 2007. Table 7.4 lists the components and reports the mean values for the two years under study as well as the changes in both absolute and relative terms. The largest component in both years is earnings, followed by pensions. Third are imputed rents from owner-occupied housing, a very rapidly increasing component. The fourth-largest component is business income, which more than tripled between 2002 and 2007. Although property income increased rapidly in 2007, it is still a minor component of income. Evidence that the planned economy generally had disappeared from urban China by 2007 shows up in the rapidly decreasing housing subsidies and the small in-kind income. Net transfer income is made up of income taxes and social security contributions, income from social relief, fees for participating surveys, private transfers, and so forth. The negative signs of income taxes and social security contributions, two main components of transfer income in terms of their absolute values, lead to the negative signs of net transfer income. In Table 7.5, let us first inspect the numerical values of the concentration coefficients for the income sources with a relative share of larger than 1 percent in 2007. We find that the distributional profile of earnings, pensions, and imputed rents from owner-occupied housing are all relatively close to the Gini coefficient in both years. In contrast, business income moved from being rather equalizing to being marginally dis-equalizing. Property income has the highest concentration coefficient of all income sources in 2007, and higher than that in 2002. The sign of the concentration coefficient

Table 7.5. Household income per capita and its decomposition, 2002 and 2007 2002 Proportion

269

Earnings Pensions Imputed rents from owner-occupied housing Business income Property income Income in-kind Housing subsidies Net transfer income Total per capita income Source: Authors’ computations from the CHIP.

Concentration coefficient

2007 Contribution

Proportion

Concentration coefficient

Contribution

69.00 17.32 5.99

0.2930 0.3341 0.3353

66.52 19.04 6.60

65.76 19.15 10.57

0.3101 0.3116 0.3421

63.15 18.48 11.20

3.30 1.13 1.01 2.86 −0.62 100

0.0580 0.4768 0.4836 0.3485 −0.2612 0.3039

0.63 1.78 1.61 3.28 0.53 100

7.14 1.52 0.64 0.63 −5.42 100

0.3650 0.7335 0.4840 0.2255 0.3439 0.3229

8.08 3.45 0.96 0.44 −5.77 100

270

Deng Quheng and Bj¨orn Gustafsson

Table 7.6. Decomposing differences in the Gini coefficient for 2002 and 2007 by income sources

Income source u02 × C02 u07 × C07 Column number Earnings Pensions Imputed rents from owneroccupied housing Business income Property income Income in-kind Housing subsidies Net transfer income Total per capita income

1

2

Contribution to changed Gini u02 C07 u07 (column 2 − C02 column 1) (u07 −u02 ) (C07 −C02 ) (u07 −u02 ) (C07 −C02 ) 3

4

5

6

7

0.2022 0.0579 0.0201

0.2039 0.0597 0.0362

0.0018 0.0018 0.0161

−0.0095 0.0061 0.0154

0.0118 −0.0039 0.0004

−0.0100 0.0057 0.0157

0.0112 −0.0043 0.0007

0.0019

0.0261

0.0241

0.0022

0.0101

0.0140

0.0219

0.0054

0.0111

0.0058

0.0019

0.0029

0.0029

0.0039

0.0049 0.0100

0.0031 0.0014

−0.0018 −0.0085

−0.0018 −0.0078

0.0000 −0.0035

−0.0018 −0.0050

0.0000 −0.0008

0.0016

−0.0186

−0.0203

0.0125

−0.0038

−0.0165

−0.0328

0.3039

0.3229

0.0190

0.0190

0.0141

0.0049

−0.0001

Source: See Table 7.5. Values in column 3 are equal to the sum of the values in columns 5 and 6, as well as the sum of the values in columns 4 and 7 (ignoring rounding errors).

for net transfer income changed across the years to become proportional to disposable income in 2007. We now use the decomposition to throw light on which channels have led to an increase in income inequality as measured by the Gini coefficient. Let us analyze the results in the following way: the difference between the two Gini coefficients for the different years can be written as  (u1k C 1k − u0k C 0k ), (2) G1 − G0 = where uik is the share of income source k in the total per capita income in year i (2002 and 2007), C ik is the concentration coefficient of the income source k in year i, and G i is the Gini coefficient of per capita disposable income in year i (2002 and 2007). The contribution to the changed Gini coefficient from each income source, reported in Table 7.6,

A New Episode of Increased Urban Income Inequality in China

271

column 3, in turn can be decomposed into changed relative shares (keeping the concentration coefficient constant) and changed concentration coefficients (keeping the relative share constant). As the latter exercise can be performed using different reference years, we report both alternatives in Table 7.6. Thus the numbers in columns 4 and 7 show one alternative decomposition whereas the numbers in columns 5 and 6 report the other. Table 7.6, column 3, shows that the two largest contributors to the increase in the Gini coefficient are the two rapidly expanding income sources – business income and imputed rents from owner-occupied housing. Business income not only increased its relative share but also became more concentrated among persons in the upper part of the distribution (the relative importance of these changes differ in the alternative decompositions). The increased contribution from imputed rents from owner-occupied housing is mainly due to its increased relative share. Compared to the trend toward increased income inequality from enterprise income and imputed rents, the impact of property income was relatively small. Notably, changes in earnings as well as in pensions, the two largest income sources, play only a small role in the increase in income inequality. Table 7.6 also reveals that the forces working against increased income inequality came mainly from net transfer income (as this component has a negative sign in total income). Other sources of income contributing to reducing income inequality include housing subsidies and income in-kind.

VI. How Various Groups Have Fared How did various groups in urban China fare during the initial phase of the Hu-Wen leadership? We divide the urban population into groups based on three dimensions: ownership sector, age of the individual, and education of the household head. We then describe changes for each category and estimate multivariate models. For each categorization we show growth curves and report means, measures of income inequality, relative poverty, and proportions of affluence. With respect to ownership sector, we find it useful to define three categories: (a) persons living in a household primarily earning wages from employment in SOEs or government institutions (the state sector); (b) persons living in households primarily connected to the private sector, that is, workers in privately owned firms, owners of a private firm, or those earning income from self-employment (the private sector); and (c) persons living in households with no working adult, that is, mainly elderly

272

Deng Quheng and Bj¨orn Gustafsson

persons living on pensions (nonworkers). Our divisions are based on the presumption that the trend toward higher income in the 2002–2007 period is strongest at the top of the income distribution within the dynamic and rapidly expanding private sector and that this income growth came not only from higher wages among skilled workers in private firms, but also from higher incomes earned by private owners as well as from rapidly increasing imputed rents from owner-occupied housing. We also hypothesize that incomes at the top of the income distribution in the slowly shrinking state sector have increased, but not as rapidly as those in the private sector. In contrast, income increases at the lowest end of the distribution in the two sectors are believed to be due to decreased labor supply and comparatively slow earnings development, for example among less-skilled workers. Furthermore, we are interested in how spatial characteristics measured by the mean income in the city where the household resides affect the income level. In our reading of the literature, differences in the distribution of income in urban China across cities have not attracted much research interest. One exception is Wang (2008) who studied urban income inequality among employed individuals in the three provinces of Liaoning, Sichuan, and Guangdong from 1986 to 2000. Based on his results, during the period under study, city differences played a large and increasing role in urban income inequality. Applying our categories, we find that the proportion of people primarily connected to the private sector increased from 25 percent in 2002 to 35 percent in 2007; mirroring this, during the same period the proportion primarily connected to the state sector decreased from 64 percent to 54 percent. In both years, 11 percent of people in urban China lived in households with no adult worker (see Table 7.7). Figure 7.3 shows that, as expected, income growth was fastest at the top of the private sector, but also among nonworkers in the lower part of the distribution. There is a pattern of people in the state sector experiencing slower income growth than people in the private sector. At the median, income growth was fastest in the private sector, followed by nonworkers, and finally in the state sector. The upward-sloping growth curves for the private and public sectors indicate that income inequality within those sectors increased, as is also shown by the Gini coefficients reported in Table 7.7. In contrast, the growth curve for nonworkers is sloping downward rather than upward for most of the distribution, and the Gini coefficient for this category did not change between the two years. Similarly, although the relative poverty rates for people in the private and state sectors increased from 2002 to 2007, among nonworker households the development was the opposite.

A New Episode of Increased Urban Income Inequality in China

273

Table 7.7. Population shares, mean income, income inequality, relative poverty, and proportion of affluence among individuals living in households primarily connected to the state sector, the private sector, and those with no workers, 2002 and 2007 2002

Proportion of all individuals (%) Average income (yuan) Gini coefficient Percentage of persons under 40% of the median income Percentage of persons under 50% of the median income Percentage of persons under 60% of the median income Percentage of persons under 70% of the median income Percentage of persons above 200% of the median income

Primarily in the state sector

Primarily in the private sector

63.99

25.13

2007 Primarily in the private sector

No workers

53.99

34.78

11.23

Primarily No in the state workers sector 10.89

8,537

6,718

8,519

14,646

12,112

14,924

0.287 4.02

0.317 5.18

0.330 8.68

0.306 4.51

0.337 5.95

0.329 6.73

8.94

9.42

15.51

10.38

12.79

12.95

15.05

16.92

22.78

17.65

19.91

19.41

23.12

26.25

29.82

25.44

27.47

25.70

8.59

11.2

12.72

9.95

11.88

11.13

Note: A household is classified as primarily linked to the state sector (private sector) if most workers are employed in the state sector (private sector). If the number of workers in the state sector is equal to the number of workers in the private sector, the household is classified as primarily linked to the state sector. As a consequence, we report a larger proportion of households primarily linked to the state sector than to the proportion of state-employed individuals, as according to the Statistical Yearbook of China. Source: Authors’ computations from the CHIP. Amounts are in 2002 prices.

Developments in the three sectors to a certain extent mirror those in two other alternative disaggregations of the population. In Figure 7.4 and Table 7.8, we divide the population into children (a category with a decreasing share of the population), and adults and elderly (a category with an increasing share of the population). In contrast to the case in many rich countries, the mean income of the elderly is higher than that of adults. Although the overall impression from Figure 7.4 is that income growth has not been

Deng Quheng and Bj¨orn Gustafsson

.8 .6

.7

Income Growth

.9

1

274

0

10

20

30

40

50

60

70

80

90

100

Percentile Primarily in state sector No worker in the household

Primarily in nonstate sector

Figure 7.3. Growth Curves for Individuals Living in Households Primarily Connected to the State Sector, the Private Sector, and Those with No Workers, 2002 and 2007. Source: Authors’ computations from the CHIP.

different for the three age groups, there are certain noteworthy differences. The elderly stand out in terms of a rapid increase at both tails of the distribution, but not in the middle. Income inequality measured by the Gini coefficient within this category increased whereas relative poverty decreased slightly. Income inequality also increased among children and adults. Relative poverty rates increased somewhat for both children and adults. It should be noted that the highest growth rates are observed at the top of the distributions for children and the elderly, but not for the adults. As opposed to rural China, few persons in urban China live in households headed by a person with only a primary education (see Table 7.9). In Figure 7.5, showing growth curves for persons living in households with the head having different levels of education, we find a difference between the less educated, many of whom are elderly, and all others. Incomes grew fastest among the less educated at the lowest part of the distribution. For those with education at the senior middle-school level and higher, the growth curve indicates increased income inequality and increased rather than decreased relative poverty rates.

275

.8 .7 .6

Income Growth

.9

1

A New Episode of Increased Urban Income Inequality in China

0

10

20

30

40

50

60

70

80

90

100

Percentile Children Elderly

Adults

.9 .8 .7 .5

.6

Income Growth

Figure 7.4. Growth Curves for Children, Adults, and the Elderly, 2002 and 2007. Source: Authors’ computations from the CHIP.

0

10

20

30

40

50

60

70

80

90

100

Percentile Primary school and below High school

Middle school University and above

Figure 7.5. Growth Curves for Individuals Where the Heads of the Household Have Various Levels of Education, 2002 to 2007. Source: Authors’ computations from the CHIP.

276

Deng Quheng and Bj¨orn Gustafsson

Table 7.8. Population shares, mean income, income inequality, relative poverty, and proportion of affluence among children, adults, and the elderly, 2002 and 2007 2002

Population shares Average income (yuan) Gini coefficient Percentage of persons under 40% of the median income Percentage of persons under 50% of the median income Percentage of persons under 60% of the median income Percentage of persons under 70% of the median income Percentage of persons above 200% of the median income

2007

Children

Adults

Elderly

Children

Adults

Elderly

14.31 7,084

74.95 8,155

10.74 8,899

12.64 12,498

74.74 13,751

12.62 15,365

0.291 4.97

0.304 5.02

0.301 6.94

0.330 6.17

0.320 6.00

0.328 5.18

10.26

10.67

12.36

12.32

12.00

10.15

17.02

17.69

19.56

18.92

18.82

17.74

24.60

25.48

26.77

26.38

26.52

24.37

8.59

9.99

9.95

11.60

10.80

12.42

Note: A person is regarded as a child if he or she is under the age of sixteen and as elderly if he or she is age sixty-one or older. Source: Authors’ computations from the CHIP. Amounts are in 2002 prices.

The overall impression from the bivariate analysis is that at the middle of the income distribution, the changes were similar for the various subgroups. This is confirmed when we run regression models for 2002 and 2007 and compare the coefficients across years. The explanatory variables measure the schooling of the household head, the age of the household head, and the age of the household head squared. Continuous variables measure the number of children in the household, the number of adults working in the state sector, the number of adults working in the private sector, the number of nonworking adults, the number of elderly with pensions, and the number of elderly without pensions. A dummy variable for Han ethnicity as a control variable is included in the specification, as is the log of city per capita income and dummies for the province. Descriptive statistics for the explanatory variables are presented in Table 7A1 in the Appendix to this chapter.

A New Episode of Increased Urban Income Inequality in China

277

Table 7.9. Population shares, mean income, income inequality, relative poverty, and proportion of affluence among individuals living in households with the heads of households having different levels of education, 2002 and 2007 2002

2007

Primary Junior Senior Primary Junior Senior and middle middle University and middle middle University below school school and above below school school and above Population shares Average income (yuan) Gini coefficient Percentage of persons under 40% of the median income Percentage of persons under 50% of the median income Percentage of persons under 60% of the median income Percentage of persons under 70% of the median income Percentage of persons above 200% of the median income

7.45 29.51 36.98 26.07 5,949 6,815 7,988 10,243

5.75 25.58 35.27 10,121 11,063 13,041

0.296 4.65

0.303 6.05

0.308 6.45

33.39 17,319

0.297 5.05

0.284 4.75

0.287 3.36

0.288 4.08

0.311 3.95

10.47

9.29

9.77

8.22

8.07

10.59

11.67

9.23

17.96

16.89

16.60

16.11

14.51

19.45

19.08

16.22

26.62

25.77

24.34

24.32

24.29

26.18

26.77

24.75

7.60

9.43

8.82

9.52

9.31

9.36

9.17

11.16

Source: Authors’ computations from the CHIP. Amounts are in 2002 prices.

The regression estimates are reported in Table 7.10. They show that household per capita income is closely and positively linked to the mean income of the city where the household resides. The estimates for the coefficients for the years of schooling of the household head are 0.047 in 2002 and 0.050 in 2007; that is, they are quite similar. Household per capita income decreases with the number of adult household members, and most rapidly if the household member is not employed. Although the number of elderly with pensions positively affects per capita income, the

Table 7.10. Income function: Dependent variable, log of household per capita income

Schooling of household head Age of household head Age of household head squared No. of children in the household No. of adults working in the state sector No. of adults working in the nonstate sector No. of nonworking adults No. of elderly with pensions No. of elderly without pensions Han ethnicity Log of city per capita income Beijing (omitted category) Shanxi Liaoning Jiangsu Anhui Henan Hubei Guangdong Chongqing Sichuan Yunnan Gansu Constant Adj. R2 No. of observations

2002

2007

0.047*** [0.001] −0.006** [0.002] 0.0001*** [0.00002] −0.066*** [0.007] −0.147*** [0.004] −0.196*** [0.004] −0.232*** [0.006] 0.101*** [0.007] −0.126*** [0.011] −0.068*** [0.016] 0.861*** [0.016]

0.050*** [0.001] −0.003* [0.002] 0.0001*** [0.00002] −0.065*** [0.007] −0.152*** [0.004] −0.185*** [0.004] −0.247*** [0.007] 0.087*** [0.007] −0.070*** [0.012] 0.016 [0.017] 0.843*** [0.012]

−0.070*** [0.020] −0.019 [0.016] −0.032** [0.015] −0.049** [0.019] −0.027 [0.017] −0.029 [0.018] 0.047*** [0.015] −0.056*** [0.020] −0.049*** [0.019] −0.025 [0.018] −0.054*** [0.020] 1.264*** [0.158] 0.447 20,624

0.010 [0.017] −0.003 [0.015] −0.014 [0.014] −0.007 [0.017] −0.012 [0.016] −0.017 [0.018] 0.130*** [0.014] 0.050*** [0.019] −0.031* [0.018] 0.043** [0.018] −0.035* [0.018] 1.238*** [0.136] 0.479 21,545

Note: Authors’ estimates from the CHIP. ** indicates statistical significance at the 5 percent level, and *** indicates statistical significance at the 1 percent level. Standard errors provided in brackets.

A New Episode of Increased Urban Income Inequality in China

279

opposite is the case for elderly without pensions. Among the coefficients for the province dummies, the positive coefficient for Guangdong stands out as having a high t-value in both years. In a second step, we focus on individuals at the two tails of the income distribution. We specify one probit model in which the dependent variable is relative poverty, defined as household per capita income below 70 percent of the median per capita income. In another model, we investigate the determinants of affluence, defined as living in a household with a per capita income of at least 200 percent of the median per capita income. The explanatory variables are the same for both models and for the linear regression model. The estimates are documented in Tables 7.A2 and 7.A3 in the Appendix to this chapter. In Table 7.11, we present the main results as predicted probabilities for some typical individuals. The overall impression from Table 7.11 is that differences in the mean city income can make a rather large difference in terms of the probability of being relatively poor or being well-to-do. Consider the typical individual A who lives in a household consisting of two employed adults and a child and where the household head has nine years of education. The probability of being poor in 2002 ranges from less than 1 percent if the household resides in a high-income city and up to 5 percent if the household resides in a low-income city. In 2007, the corresponding variation increases from 7 percent to as much as 55 percent. This example illustrates that although the relative poverty rate in the 2007 sample is only slightly higher than the relative poverty rate in the 2002 sample, there may be hidden substantial increased poverty risks for households with certain characteristics. The predictions in Table 7.11 also show that children and the elderly fare rather differently depending on their household. Among the elderly, there is substantial variation based on city income, the type of household, and whether or not the elderly receive a pension. It is striking that an elderly person without a pension living in a multigenerational household (individual B) in a low-income city in 2007 is predicted to have a 67 percent probability of being poor and a less than 1 percent probability of being rich. In contrast, a person living with one’s spouse (individual G) in a high-income city has less than a 1 percent probability of being poor and a 93 percent probability of being affluent. The simulations also illustrate how the probabilities are affected if one adult loses his or her job (compare individual A and individual C), the importance of the level of education of the household head (compare individual C and individual D), whether or

280

Deng Quheng and Bj¨orn Gustafsson

Table 7.11. Predicted probabilities of relative poverty and affluence, 2002 and 2007 (percentages)

Individual A

B

City income

Description of the individual

Relative poverty (percentage)

Affluence (percentage)

2002 4.67 2.13 0.31

2007 55.36 25.60 6.56

2002 1.09 3.11 29.99

2007 0.35 1.55 9.24

Low Middle High

Household head aged 47.9 years, 9 years of education, 2 adults employed in the state sector, 1 nonworking adult, 1 child, Han

Low

The same as A, but the household

9.06

66.81

0.65

0.35

increases by one elderly person

4.24

35.83

1.86

1.55

Middle

0.62

10.23

20.18

9.23

C

High Low Middle High

The same as A, but one worker becomes a nonworker

without a pension

6.84 3.16 0.46

66.09 35.10 9.94

0.85 2.44 25.01

0.24 1.09 6.62

D

Low Middle High

The same as C, but the household head has 16 years of education

1.91 0.86 0.12

32.33 11.70 2.63

3.67 9.99 59.66

1.70 7.23 33.43

E

Low Middle High

The same as D, but there is no child in the household

1.39 0.62 0.09

28.57 9.99 2.21

5.84 15.32 70.68

2.52 10.42 42.84

F

Low Middle High

The same as E, but the household increases by one elderly person with a pension

0.81 0.36 0.05

19.31 6.23 1.34

7.41 18.91 75.66

3.83 15.17 53.56

G

Low Middle High

An elderly couple living alone. The household head is 65 years old, has nine years of education, there is no child in the household, and one elderly person has a pension

0.24 0.11 0.02

3.54 1.01 0.21

51.98 75.93 97.68

31.67 67.58 93.07

Note: Low/median/high city income is defined as the mean for the first decile/the median/ the tenth decile for the year under study. Poverty is defined as living in a household with average disposable income that is less than 70 percent of the median income in urban China during the same year. Source: Estimates presented in the Appendix to this chapter.

not there is a child (compare individual D and individual E), and whether or not there is an elderly person receiving a pension (compare individual F and individual E). The findings in this section reveal differences in how various categories of Chinese urbanites fared between 2002 and 2007. For example, households

A New Episode of Increased Urban Income Inequality in China

281

closely connected to the expanding private sector and at the top of the income distribution experienced more rapid income increases than most other households. Furthermore, although relative poverty increased from 2002 to 2007 for children as well as for adults, this was not the case for the elderly. Overall, however, the data do not indicate any dramatic change in income determination from 2002 to 2007. In contrast, we find substantial differences in the economic situation of households across cities. China’s urban poverty problem is disproportionally concentrated in low-income cities and affluent households are disproportionately concentrated in high-income cities. We have reported a wide variation in household income among urban households with children or with elderly. Elderly couples living alone, particularly if they live in high-income cities, fare much better than elderly living in multigenerational households, particularly in households in low-income cities.

VII. Conclusion In this chapter we study income changes among Chinese formal urban residents between 2002 and 2007, with comparisons to earlier periods. Using the CHIP urban household survey data, we investigate trends in real income, income inequality, and poverty. The reasons for the changes in income inequality are investigated by decomposing the Gini coefficient for household per capita income by income components. Furthermore, we describe how various categories of people have fared by breaking down the population along three dimensions: ownership of the workplace (or, alternatively, not working), age of the individual, and education of the household head. We show the bivariate analyses and estimate income functions for these different population groups. We report that overall income increased more rapidly in urban China between 2002 and 2007 than it did during the two preceding periods of 1988– 1995 and 1995–2002. For example, although median per capita income grew by 2.7 percent per annum from 1988 to 1995, it grew by 4.8 percent from 1995 to 2002, and it grew by as much as 10.6 percent from 2002 to 2007. In contrast to the 1995–2002 period, income inequality increased between 2002 and 2007, although the increase was not as rapid as that between 1988 and 1995. The increases in real income at the bottom of the income distribution from 2002 to 2007 mean that, assessed against absolute poverty lines representing constant purchasing power, the proportion of people considered

282

Deng Quheng and Bj¨orn Gustafsson

to be poor decreased. However, as such income gains were slower than were those at the median, the trend of increased relative poverty in urban China continued. Therefore, views about China’s poverty problem very much depend on the perspective by which it is viewed. If households are observed through a lens that is used to view low-income countries, poverty is not a problem in urban China today. However, if viewed through a lens used to view high-income countries, the poverty problem among Chinese urban residents is similar in magnitude to that in many rich countries. Income inequality among urban residents increased through two major channels. The most important channel was the rapid increase in income from private businesses and self-employment at the top of the income distribution. In 2007 China had more private entrepreneurs and persons who were self-employed than it had in 2002, and their incomes were increasingly concentrated in the higher segments of the income distribution. The secondmost important factor contributing to increased urban income inequality was the rather rapid increase of imputed rents from owner-occupied housing. This may be due to increases in the stock of owner-occupied housing as well as to the rapid increases in housing prices. Interestingly, neither wage earnings outside of the private sector nor pensions were a major factor contributing to the increase in inequality. Between 2002 and 2007 Chinese urbanites did not enjoy a uniform rate of income growth. For example, households closely connected to the expanding private sector and at the higher end of their income distribution experienced more rapid income increases than most other households. However, the overall impression has been that no dramatic changes in income determination occurred between 2002 and 2007. In contrast, we have reported substantial differences in the economic situation of households across cities. China’s urban poverty problem is disproportionally concentrated in lowincome cities and affluent households are most prevalent in high-income cities. We have also illustrated that urban children and urban elderly reside in households with rather diverse economic circumstances. Elderly couples living alone, particularly if they live in high-income cities, fare much better than do those living in multigenerational households, particularly if they are living in low-income cities. Thus, in this chapter we have shown that China’s road toward increased income inequality did not come to a halt during the first phase of the Hu-Wen leadership. On the contrary, both income inequality and relative poverty increased. It should be stressed, however, that our analysis

A New Episode of Increased Urban Income Inequality in China

283

indicates that the major factors driving increased income inequality were the rapid increases in income from the private sector, particularly at the top of the distribution, as well as increased imputed rents from owner-occupied housing. Both of these factors can be attributed to policy changes initiated before the Hu-Wen leadership period.

APPENDIX Table 7A.1. Descriptive statistics

Schooling of the household head Age of the household head Age of the household head squared No. of children in the household No. of adults working in the state sector No. of adults working in the nonstate sector No. of nonworking adults No. of elderly with a pension No. of elderly without a pension Han ethnicity Log of city per capita income Beijing Shanxi Liaoning Jiangsu Anhui Henan Hubei Guangdong Chongqing Sichuan Yunnan Gansu Source: Authors’ computation from the CHIP.

2002

2007

10.67 47.67 2,394.96 0.49 2.10 0.83 0.30 0.27 0.07 0.96 8.94 0.07 0.09 0.10 0.10 0.07 0.10 0.10 0.09 0.04 0.08 0.09 0.06

11.99 48.99 2,535.29 0.44 1.78 1.15 0.27 0.32 0.06 0.97 9.46 0.11 0.08 0.10 0.08 0.07 0.09 0.05 0.11 0.06 0.08 0.08 0.08

284

Deng Quheng and Bj¨orn Gustafsson

Table 7A.2. Poverty function (poverty line set at 70 percent of the median income)

Schooling of the household head Age of the household head Age of the household head squared No. of children in the household No. of adults working in the state sector No. of adults working in the nonstate sector No. of nonworking adults No. of elderly with a pension No. of elderly without a pension Han ethnicity Log of city per capita income

2002

2007

−0.189*** [0.006] 0.047*** [0.015] −0.001*** [0.0002] 0.323*** [0.043] 0.480*** [0.028] 0.729*** [0.029] 0.884*** [0.039] −0.553*** [0.052] 0.709*** [0.068] 0.238** [0.103] −3.334*** [0.105]

−0.201*** [0.007] −0.005 [0.014] −0.0002* [0.0001] 0.178*** [0.041] 0.683*** [0.026] 0.837*** [0.026] 1.135*** [0.043] −0.514*** [0.047] 0.484*** [0.070] 0.013 [0.105] −3.403*** [0.097]

3.018*** [0.515] 2.762*** [0.512] 3.001*** [0.511] 3.117*** [0.513] 3.025*** [0.512] 2.753*** [0.512] 2.414*** [0.514] 3.089*** [0.516] 3.028*** [0.513]

−0.092 [0.122] 0.185 [0.115] 0.285** [0.122] −0.107 [0.120] −0.089 [0.117] 0.153 [0.123] −0.409*** [0.122] −0.299** [0.126] 0.416*** [0.117]

Beijing (omitted category) Shanxi Liaoning Jiangsu Anhui Henan Hubei Guangdong Chongqing Sichuan

A New Episode of Increased Urban Income Inequality in China

Yunnan Gansu Constant Pseudo R2 No. of observations

285

2002

2007

2.794*** [0.512] 2.930*** [0.514] 24.832*** [1.155] 0.257 20,626

−0.067 [0.120] 0.236* [0.121] 31.496*** [1.023] 0.279 21,545

Note: We set the poverty line at 70 percent of the median income because at a poverty line of 60 percent of the median income, no one in Beijing is poor, which makes it impossible to estimate the poverty function. ** indicates statistical significance at the 5 percent level; *** indicates statistical significance at the 1 percent level. Standard errors are provided in brackets. Source: Authors’ estimates from the CHIP.

Table 7A.3. Affluence function, with 200 percent of the median income as the threshold

Schooling of the household head Age of the household head Age of the household head squared No. of children in the household No. of adults working in the state sector No. of adults working in the nonstate sector No. of nonworking adults No. of elderly with a pension No. of elderly without a pension Han ethnicity Log of city per capita income

2002

2007

0.217*** [0.010] 0.019 [0.022] 0.00004 (0.0002) −0.466*** [0.076] −1.250*** [0.053] −1.460*** [0.058] −1.526*** [0.081] 0.233*** [0.073] −0.591*** [0.176] −0.344** [0.143] 4.369*** [0.209]

0.277*** [0.011] 0.012 [0.017] 0.00001 (0.0002) −0.364*** [0.069] −0.915*** [0.045] −0.963*** [0.046] −1.297*** [0.082] 0.432*** [0.062] 0.112 [0.147] 0.223 [0.161] 3.890*** [0.143] (continued)

286

Deng Quheng and Bj¨orn Gustafsson 7A.3 (continued) 2002

2007

0.555*** [0.188] 0.134 [0.144] 0.316** [0.126] 0.065 [0.195] 0.354** [0.154] −0.207 [0.205] 0.683*** [0.103] 0.592*** [0.162] 0.467*** [0.172] 0.044 [0.167] 0.032 [0.210] −41.185*** [2.062] 0.298 20,626

0.237 [0.176] −0.049 [0.119] 0.116 [0.088] −0.066 [0.146] −0.161 [0.121] 0.032 [0.155] 0.636*** [0.083] 0.275* [0.154] 0.541*** [0.128] 0.738*** [0.161] −0.336 [0.214] −41.205*** [1.501] 0.303 21,545

Beijing (omitted category) Shanxi Liaoning Jiangsu Anhui Henan Hubei Guangdong Chongqing Sichuan Yunnan Gansu Constant Pseudo R2 No. of observations

Note: A person living in a household with a disposable per capita income of at least 200 percent of the median income as observed during the year under study is classified as affluent. ** indicates statistical significance at the 5 percent level; *** indicates statistical significance at the 1 percent level. Standard errors are provided in brackets. Source: Authors’ estimates from the CHIP.

References Appleton, S., L. Song, and Q. Xia (2010), “Growing Out of Poverty: Trends and Patterns of Urban Poverty in China 1988–2002,” World Development, 38(5), 665–678. Brandt, L. and C.A. Holz (2006), “Spatial Price Differences in China: Estimates and Implications,” Economic Development and Cultural Change, 55(1), 43–86. Cai, H., Y. Chen, and L-A. Zhou (2010), “Income and Consumption Inequality in Urban China: 1992–2003,” Economic Development and Cultural Change, 58(3), 385–413.

A New Episode of Increased Urban Income Inequality in China

287

Chen, G., J. Li, and H. Matlay (2006), “Who Are the Chinese Private Entrepreneurs? A Study of Entrepreneurial Attributes and Business Governance,” Journal of Small Business and Enterprise Development, 13(2), 148–160. Chen, S. and M. Ravallion (2010), “China Is Poorer Than We Thought, but No Less Successful in the Fight against Poverty,” in S. Anand, P. Segal, and J. Stiglitz, eds., Debates on the Measurement of Global Poverty, 327–340, Oxford: Oxford University Press. Davis-Friedmann, D. (1991), Long Lives: Chinese Elderly and the Communist Revolution, expanded edition, Stanford, CA: Stanford University Press. Dickson, B. (2008), Wealth into Power: The Communist Party’s Embrace of China’s Private Sector, New York: Cambridge University Press Eichen, M. and M. Zhang (1993), “Annex: The 1988 Household Sample Survey: Data Description and Availability,” in K. Griffin and R. Zhao, eds., The Distribution of Income in China, 331–346, Basingstoke: Macmillan. Foster, J., J. Greer, and E. Thorbecke (1984), “A Class of Decomposable Poverty Indices,” Econometrica, 52(3), 761–765. Gustafsson, B., S. Li, and H. Sato (2004), “Can a Subjective Poverty Line Be Applied to China? Assessing Poverty Among Urban Residents in 1999,” Journal of International Development, 16(8), 1089–1107. Gustafsson, B.A., S. Li, and T. Sicular (2008), Inequality and Public Policy in Urban China, New York: Cambridge University Press. Haggard, S. and Y. Huang (2008), “The Political Economy of Private-Sector Development in China,” in L. Brandt, and T. G. Rawski, eds., China’s Great Economic Transformation, 337–374, New York: Cambridge University Press. Khan, A.R., K. Griffin, C. Riskin, and R. Zhao (1993), “Household Income and Its Distribution in China,” in K. Griffin and R. Zhao, eds., The Distribution of Income in China, 25–73, Basingstoke: Macmillan. Li, H., L. Meng, Q. Wang, and L-A. Zhou (2008), “Political Connections, Financing and Firm Performance: Evidence from Chinese Private Firms,” Journal of Development Economics, 87(2), 283–299. Li, S., C. Luo, Z. Wei, and X. Yue (2008), “Appendix: The 1995 and 2002 Household Surveys: Sampling Methods and Data Description,” in B.A. Gustafsson, S. Li, and T. Sicular, eds., Inequality and Public Policy in China, 337–354, New York: Cambridge University Press. Logan, J., Y. Fang, and Z. Zhang (2010), “The Winners in China’s Urban Housing Reform,” Housing Studies, 25(1), 101–117. Meng, X., R. Gregory, and Y. Wang (2005), “Poverty Inequality and Growth in Urban China 1986–2000,” Journal of Comparative Economics, 33(4), 710–729. National Bureau of Statistics (NBS) (various years), Zhongguo tongji nianjian (China Statistical Yearbook), Beijing: Zhongguo tongji chubanshe. Organisation for Economic Co-operation and Development (OECD) (2008), Growing Unequal? Income Distribution and Poverty in OECD Countries, Paris: OECD. Palmer, E. and Q. Deng (2008), “What Has Economic Transition Meant for the WellBeing of the Elderly in China,” in B.A. Gustafsson, S. Li, and T. Sicular, eds., Inequality and Public Policy in Urban China, 182–202, New York: Cambridge University Press. Ravallion, M. and S.H. Chen (2003), “Measuring Pro-Poor Growth,” Economic Letters, 78(1), 93–99.

288

Deng Quheng and Bj¨orn Gustafsson

Riskin, C. and Q. Gao (2010), “The Changing Nature of Urban Poverty in China,” in S. Anand, P. Segal, and J. Stiglitz, eds., Debates on the Measurement of Global Poverty, 300–326, Oxford: Oxford University Press. Saunders, P. (2007), “Comparing Poverty Among Older People in Urban China Internationally,” China Quarterly, no. 190, 451–465. Wang, F. (2008), Boundaries and Categories: Rising Inequality in Post-Socialist Urban China, Stanford, CA: Stanford University Press. Wong, C.K. (1995), “Measuring Third World Poverty by the International Poverty Line: The Case of Reform China,” Social Policy and Administration, 29(3), 189–203. Wong, C.K. (1997), “How Many Poor People in Shanghai Today? The Question of Poverty and Poverty Measure,” Issues and Studies, 33(2), 32–49. Yueh, L. (2009), “China’s Entrepreneurs,” World Development, 37(4), 778–786. Zhongguo siying jingji nianjian (2006.6–2008.6) (Economic Yearbook of the Private Economy [June 2006-June 2008]) (2009), Beijing: Zhonghua gongshang lianhe chubanshe.

EIGHT

Unemployment and the Rising Number of Nonworkers in Urban China Causes and Distributional Consequences Bj¨orn Gustafsson and Ding Sai I. Introduction Before the economic transition, almost all urban women between the ages of sixteen and fifty (for manual workers) or fifty-five (for professional workers) and urban men between the ages of sixteen and sixty worked for an income. This situation changed when China modernized and moved toward a market economy. In this chapter we show that whereas only 6 percent of those of working age who had an urban residence permit were nonworkers in 1988, the proportion increased to 15 percent in 1995, to 29 percent in 2002, and to as much as 36 percent in 2007. Such a rapid change in the expenditure burden is difficult to cope with in most economies. China, however, has experienced very rapid economic growth, as well as favorable demographic changes because many young persons were entering the labor force and few children were being born. In addition, an increasing proportion of paid work in urban China was performed by rural migrants, who generally work long hours and are paid less than urban residents. The rise of nonworkers in urban China is the result of various processes that to some degree have affected persons differently, both over time and in terms of age. One process that is shared with many rich countries is the rapid expansion of education since the late 1990s, leading to more young adults remaining students and not working for an income. Furthermore, during the planning era, the transition from being a student to one’s first Previous versions of this paper were presented at the Workshop on Income Inequality, Beijing Normal University, Beijing, May 2009 and May 2010; the Chinese Economist Society Annual Meeting, Xiamen, China, June 2010; the 31st General Conference of the International Association for Research on Income and Wealth, St. Gallen, Switzerland, August 2010; and the 2nd CIER/IZA Workshop on Research in Labor Economics, Bonn, October 2010. We are grateful to the participants on these occasions as well as to Terry Sicular and an anonymous referee for useful comments.

289

290

Bj¨orn Gustafsson and Ding Sai

job (which often became life-long) typically did not involve periods of enforced nonwork. In contrast, more recent changes mean that schoolleavers in contemporary China may experience periods of unemployment before gaining a foothold in working life, as is also the case in many rich countries. In this respect, therefore, the Hu Jintao–Wen Jiabao leadership period represents a continuation of circumstances that appeared at the end of the preceding period. Another kind of process began in the 1990s when state and collective enterprises were restructured to function more like capitalist enterprises. Two elements were involved: restructuring ownership and abolishing permanent job tenure. The latter was carried out under the slogan of “smashing the iron rice bowl” (za tiefanwan). In most enterprises, the goal of reducing the number of employees and increasing efficiency resulted in large-scale layoffs. In parallel, enterprises were no longer responsible for housing, pensions, health insurance, and social services. As consequence of these changes, many workers left the workforce before general retirement age. Such processes did not proceed at random. For example, unskilled workers have met increased competition from the recent large influx of rural-urban migrants, a situation that rarely affects skilled workers. Women and older workers are often considered less attractive potential employees and for them the option of nonwork is more socially acceptable than it is for men in their prime. Furthermore, China’s economic restructuring has been a spatially uneven process, affecting some locations and cities more than others. The structural changes in the economy during the Hu-Wen period to date differ in one important respect from the changes in the years immediately preceding their leadership (when the total number of jobs in urban China decreased drastically). The Hu-Wen period so far has been characterized by increasingly more urban jobs in China as a whole, leading to a decline in employment problems, although not to the level in the 1990s before the restructuring of ownership and the abolition of permanent job tenure. This chapter aims to increase our knowledge about nonwork in urban China in several directions. At the descriptive level, it provides unprecedented details on the changed gender/age profile of work and nonwork. We use large samples taken from many cities across China to study the situation in 1988, 1995, 2002, and 2007. We argue that it is meaningful to distinguish between five different categories of nonworkers: students, the unemployed, the early retired, homemakers, and the residual category. Of these, the unemployed together with the employed are regarded as a part of the labor force, and the key indicator, “the unemployment rate,” is defined as the proportion of unemployed in the labor force. In contrast, persons

Unemployment and the Rising Number of Nonworkers in Urban China

291

who are students, early retired, homemakers, or belong to a residual category of nonworkers are not regarded as part of the labor force. From this, it follows that the unemployment rate does not necessarily capture the entire employment problem. In the case of urban China we illustrate that the probability that a person is in one of the following states – students, early retired, homemaker, or the residual category – actually is affected by the employment situation in the city. However, at least to some extent, the determinants of being grouped in one of these five categories of not being employed can be assumed to differ. In one research question, we ask, “What characterizes people who belong to the various categories of nonworkers?” Starting from the assumption that the determinants differ throughout the life cycle, we investigate the role played by gender, education, household characteristics, as well as whether the employment situation in the city where the person resides affects the probability of being in a different state of nonwork, and if so, how? As a second research question, we ask, “What is the economic well-being of the various categories of nonworking people?” Not only the personal income of the nonworker plays a role, but also the income of the other household members as well as the expenditure burden of the household are important. We expect to find considerable variations across, as well as within, the different categories of nonworkers. E.g., a considerable number of students are the only children of well-off, or relatively well-off, parents, whereas other students are not as privileged. Some early-retired individuals receive relatively generous pensions and live with a partner who has above-average personal income, whereas other early-retired individuals are less fortunate in one or both of these respects. The economic situation of unemployed persons can be expected to be worse, because, if it is received at all, unemployment compensation is often meager. It is an open question whether or not the situation is similar for homemakers, which would be the case if the respondents to our survey communicated this alternative, instead of the alternative “unemployed.” Has urban China seen a reappearance of traditional housewives, that is, spouses of well-off workers? Our research aims to provide new information about family roles and gender differences in urban China during the two decades of rapid economic expansion and structural change. By now, a number of authors have addressed issues about the changed labor market in urban China since the introduction of the reforms. Several authors have studied issues related to laid-off workers and the unemployed. Other authors have estimated earnings or wage functions, sometimes linked to the study of the gender-wage gap. There are also writings about

292

Bj¨orn Gustafsson and Ding Sai

inequalities of remuneration and of household income. In contrast to the situation in the West, we have not found much research on the transition from school to work among young adults in urban China. The same is the case for the issue of transition into early retirement. Our study has the potential of contributing to knowledge on the economic well-being of persons who are nonworkers in urban China (a topic that, to the best of our knowledge, has attracted very little attention in the literature), and thereby to the literature on the changing income distribution in urban China. Turning to our results, we show that the age at which a majority of young men begin to work increased from age seventeen in 1988 to age twenty-four in 2007. The situation for young women is similar. Parallel to this, larger proportions of people in most age brackets did not work for pay. Furthermore, there was a clear change between 1988 and 2002 of a lowering of the ages when the person exited the labor force, but thereafter, the real retirement age bounced back to be about the same in 2007 as it was in 1988. A major result is that from 1988 to 2002 there was an outspoken education and gender bias in the process of increased nonwork in urban China. Among middle-aged and older workers, a low level of education and being female profoundly increased the probability of belonging to various states of nonwork in 2002. However, between 2002 and 2007, the first five years of the Hu-Wen era, more jobs were created, thereby reducing the influence of lower levels of education or of being female. In contrast, we also find that there is little indication of nonwork gender differences among young persons. Furthermore, local labor-market conditions affect the probability of belonging to some state of nonwork. Open unemployment is only one of the consequences of a low urban employment rate. When the employment rate is low, young adults are more likely to continue their studies, whereas middle-aged and older workers face an increased probability of retiring early, becoming a homemaker, or belonging to the residual category of nonworkers. Another major finding is that in several cases, income deficits due to nonwork are cushioned most importantly by income from working household members or by transfers. Several students have parents with high personal incomes. A substantial number of those who retire early make sizable contributions to the household income with their pensions. These are important reasons why we can report that students and the early retired are relatively evenly represented across the household income distribution. In contrast, the unemployed and homemakers are concentrated at the lower deciles of the distribution, particularly in 2002. The increased proportion of unemployed and homemakers thus directly contributed to making the distribution of income at the household level more unequal, whereas

Unemployment and the Rising Number of Nonworkers in Urban China

293

possible contributions from the increased proportion of students and the early retired are not equally straightforward. Although during the first five years of the Hu-Wen leadership the number of jobs increased in urban China, there was also continuity in widening inequality along some dimensions. One dimension was spatial as the dispersion in employment rates across cities continued to increase. Another dimension was the continued trend toward increased gender inequality as the wife’s share of the couple’s income continued to decrease, whereas the husband’s contribution continued to rise. The topic in this chapter is closely related to that in Chapter 7, which also analyzes changes in urban income inequality with an emphasis on the Hu-Wen period up to 2007 using the same data. Chapter 7 describes the development of poverty and analyzes how changes in various income components affected the development of income inequality – topics not addressed in this chapter. Furthermore, although Chapter 7 investigates, on the one hand, how the relation between education, household size, and some other characteristics has changed and, on the other, how income has changed, this chapter focuses on households with members who do not work. The remainder of this chapter proceeds as follows: In the next section, we describe the context, and in Section III we present our data and describe how nonwork varies across the life cycle for men and women during the three years under study. In the same section, we also introduce five categories of nonworkers. In Section IV we analyze the characteristics of the various categories. The question of the economic well-being of the various categories of nonworkers is addressed in Section V. The chapter ends with a summary and a discussion of our findings.

II. Context and Conjectures Newer birth cohorts in China have considerably more education than do earlier birth cohorts because of both supply and demand reasons. But the expansion has not always been smooth. For example, the data do not reveal an increasing trend in the transition rate from primary to middle school during the 1980s (Hannum et al. 2008: 231). In contrast, starting at a rate of slightly over 70 percent, the transition rate went up to over 90 percent in the 1990s. At the beginning of the reform period, there were relatively few students, relative to the large population, in universities and colleges in urban China. In 1980, as few as 5 percent of graduates from middle schools continued to the tertiary level. In 1984 the transition rate increased

294

Bj¨orn Gustafsson and Ding Sai

to 25 percent where it remained until the end of the decade, when it then began to increase, reaching over 50 percent at the end of the 1990s. The expansion of higher education in China has been rather impressive. The number of new students enrolled in regular institutions of higher education increased from 0.7 million in 1988, to 0.9 million in 1995, to 3.2 million in 2002, and to a stunning 6.6 million in 2007.1 The expansion of higher education has resulted in large costs for the public sector, but a part of the costs has been shifted to the students and their parents because universities and colleges now charge tuition fees. As a result of the expansion, China’s share of the world’s more highly educated labor force is growing, a process that unquestionably will have global consequences. Complementing the supply reasons, the expansion of education should also be understood from the demand side. As the Chinese economy has grown and become more complex, the job requirements faced by potential workers have become higher. This, together with the transition from a planned economy (where income disparities from education were rather small), as well as the increased competition less-qualified workers have faced from migrant workers, has meant that incentives for pursuing a longer education have become greater. Recent studies on changes in the rates of return to education in China confirm this (e.g., Knight and Song [2008]; Zhang et al. [2005]). Rapid economic growth has also made it possible for parents to finance longer periods of education for their children. Working in the same direction, China’s adoption of the one-child policy means that few children born in urban China since 1979 have siblings who compete for parental resources. The growth rates of spending on education typically have been higher than the growth rates of GDP. Hannum et al. (2008: 222) provide information for the 1994–2004 period. The expansion of education is a major reason why young adults are entering the labor force later in urban China. Students in China typically do not work for pay during the school terms or holidays. Another difference with the West, where the school-to-work transition has attracted considerable research interest (see, e.g., Ryan 2001), is that the nature of the process of leaving school has changed dramatically. Before 1984 the government assigned jobs to graduates. Thereafter, a mixed system prevailed whereby young job seekers could also find jobs on their own. Since 1993, however, graduates have been left entirely on their own to find employment (Zhao and Wen 2008). These changes have taken place in an environment where 1

National Bureau of Statistics, Zhongguo tongji zhaiyao 2011 (China Statistical Abstract 2011) (Beijing: Zhongguo tongji chubanshe, 2011), 166–167.

Unemployment and the Rising Number of Nonworkers in Urban China

295

supply is greater than demand and, like the situation in rich countries, many school-leavers in China now have difficulties finding a first job.2 On average, urban parents are much better off than are their rural counterparts and can afford schooling fees and other educational expenses as well as the forgone income when a studying child is not economically active. Similarly, school administrations in urban areas are much richer than are those in rural areas (particularly those in rural poor areas), and therefore can provide more varied opportunities for learning (Tsang and Ding 2005). As a consequence, school enrollments and education attainments vary profoundly by location (Connelly and Zheng 2003; Hannum and Wang 2006). In terms of spatial variations, to what extent does employment at the city level affect the probability of belonging to the different categories of nonworkers? To what extent is there evidence that student status is more likely when employment opportunities are limited? We are also interested in determining the extent to which intergenerational links in education activities affect the probability of nonwork. During the Cultural Revolution period (1966–1976), Chinese education policy focused on eliminating the educational disparities brought about by social background. Yet such policies are now history, and the restructuring of the educational system has been portrayed as “embracing neo-liberalism” (Mok, Wong, and Zhang 2009). Education is becoming more commercialized, public institutions are no longer entirely public, and there has been a rapid emergence of private schools. For example, in 2007 almost half of the funds for institutions of higher education came from tuition and miscellaneous sources. This was not the case in the past when fewer funds were allocated for education (National Bureau of Statistics [NBS] various years). Statistical data show that the average yearly tuition fee paid by households in 2007 was several thousand yuan, or a sum equivalent to the average household income for three to four months.3 Increasingly, Chinese parents are helping their children financially, but for many, the paying of tuition fees is difficult. Financing a child’s educational expenditures has become a top savings objective for Chinese residents.4 Thus, one should expect that, 2

3 4

Publication of the book Ant Tribe, built on interviews by Lian Si, in September 2009, led to a discussion of living conditions among a group of university graduates in large cities. They dream of a better life but struggle with low-paying jobs, spending too much time traveling to and from work, and residing in crowded housing. See http://en .wikipedia.org/wiki/Ant tribe. Accessed June 5, 2012. See “Lai kan gedi daxue xuefei biaozhun” (Standard University Fees in Each Region), April 4, 2007, http://bbs.edu5a.com/showtopic-422.html. Accessed July 23, 2011. See http://www.pbc.gov.cn/publish/diaochatongjisi/193/1685/16850/16850 .html. Accessed September 19, 2011.

296

Bj¨orn Gustafsson and Ding Sai

as is observed almost universally, the probability of being a student in urban China today is affected by the economic situation of the parents. Turning to gender, however, it is unclear whether gender differences affect the probability of being a student. The one-child policy in urban China means that parents do not have a choice between investing in the education of a son or a daughter. We now turn to the other major process leading to nonwork, that is, the economic restructuring that resulted in both the growth of unemployment and the decline in labor-force participation. Before the mid-1990s, there was virtually no open unemployment in urban China; thus, there was no need for a system of unemployment insurance. This changed, however, as central policies aimed to reduce the number of employees and to “smash the iron rice bowl.” As a consequence, the number of wage earners in stateowned enterprises (SOEs) and institutions and urban collectively owned units fell from 141 million in 1995 to 80 million in 2002. The huge drop of 61 million jobs in the combined state and collective sectors between 1995 and 2002 was only partly cushioned by 17 million new jobs created in other types of ownership, including 7 million jobs created from selfemployment. Because more jobs were lost than were created and the workactive population increased, the employment problem among registered urbanites was exacerbated. This situation changed between 2002 and 2007, the first years of the HuWen period. Although the number of wage earners in state and collectively owned units continued to decrease, the decline was down to fewer than 2 million persons per year, not the almost 9 million per year as was the case during the preceding period. This means that in 2007 the state and collectively owned units together employed 73 million workers, as compared with 49 million wage earners in other ownership sectors, 33 million selfemployed, and 10 million owners of private enterprises. The employment shocks that hit urban China in the mid-1990s and for some years thereafter were particularly serious in certain localities, for lessqualified workers, among older workers, and for females (Appleton et al. 2002; Cai, Park, and Zhao 2008; Giles, Park, and Cai 2006). Although in a growing economy many displaced workers are able to find new jobs, in China the demand for labor did not keep pace with the number of schoolleavers, displaced workers, and new rural-urban migrants. Furthermore, urban Chinese labor markets were not well developed, and due to residence restrictions, geographic mobility was difficult. Hence, since the mid-1990s open unemployment has become a reality for many urban residents and others who have left the labor force.

Unemployment and the Rising Number of Nonworkers in Urban China

297

It is difficult to provide a precise picture of the extent and evolution of open unemployment and the drop in labor-force participation in urban China. Currently, there are no official, reliable, and timely countrywide labor-force surveys comparable to those in the Organisation for Economic Co-operation and Development [OECD] countries from which unemployment rates and labor-force rates can be computed annually.5 However, a survey of five large cities reported the unemployment rate of urban residents, computed by international standards, to be 6.8 percent in 1996 and 11.1 percent in 2002 (Giles et al. 2006). These figures indicate an unemployment rate similar in magnitude to that of contemporary rich countries and considerably higher than the rate of registered unemployed, which stood at 2.9 percent and 4.0 percent, respectively, for 1996 and 2002 (NBS various years). Prior to the Hu-Wen period, several policy initiatives were aimed at reducing the consequences of employment problems in China. Thus, in 1999 the National Unemployment Insurance Rules were extended beyond the state enterprise sector and made mandatory for all urban employees (Duckett and Hussain 2008). Funded by contributions from employers and employees, this system provided benefits (xiagang butie) or retraining (zai jiuye peixun) for those laid-off workers who registered and paid contributions. Some work units provided laid-off workers with an early retirement (zaotui) for the years until retirement – age fifty or fifty-five for women and age sixty for men. Another measure to ease the consequences of job losses was the creation of the xiagang category, a situation where workers were laid off but still kept their ties with the work unit. If the work unit could afford it, the workers received a low wage and some welfare benefits (Wong and Ngok 2006). During the period of very rapid job losses, it became common for work units to buy out middle-aged and older employees with a lump sum related to their cumulative future earnings up to regular retirement (maiduan gongling). Together with those workers who had accumulated a work history of thirty years or more, they voluntarily chose to terminate their employment upon receiving a monthly stipend. A major policy shift prior to the Hu-Wen period was the expansion of the system of social assistance to urban 5

Solinger (2001) draws attention to a number of problems in defining and estimating unemployment in China. However, starting in 2009 the NBS has conducted labor-market surveys, but at the time of the editing of this text, March 2012, the results have not yet been published. There was a pilot survey in 2005 conducted by the NBS, the Ministry of Human Resources and Social Security, the Ministry of Agriculture, and the All-China Federation of Trade Unions. The results are published in State Statistical Bureau (SSB) (2006).

298

Bj¨orn Gustafsson and Ding Sai

residents, resulting in an increase in the number of recipients from fewer than 2 million in 1998 to 22 million in 2003 (Gao, Garfinkel, and Zhai 2009; Gustafsson and Deng 2011; Leung 2006; NBS various years).6 However, as labor-market conditions improved after the beginning of the Hu-Wen leadership, many of the policies affecting jobless workers changed. Work units no longer have xiagang workers, and it has been rare for work units to buy out middle-aged and older workers with a lump sum. The dibao (minimum living standard guarantee) program is no longer expanding in terms of the number of recipients, although the number on the welfare rolls has not decreased.7 We can conclude that the income consequences of restructuring have attracted much policy-making attention. Nevertheless, the measures have not been able to counteract the trend toward greater income inequality. Results from two studies on the pre–Hu-Wen period point in this direction. Based on an analysis of household income data for 1988, 1995, and 1999, Meng (2004) concludes that unemployment and other effects of the economic restructuring were the main contributors to the increase in the Gini coefficient for urban household income. Cai, Chen, and Zhou (2010) constructed panel data at the provincial level for the period from 1992 to 2003 and related measures of inequality to variables indicating SOE reform, urbanization, and globalization. They conclude that the SOE reform was the most significant factor contributing to the rise in urban inequality during the period. To what extent has the restructuring affected men and women differently? Two ideal types describing how public policies in rich countries structure gender roles are the “breadwinner model” and the “dual-earner model.” Urban China possesses several characteristics of the latter. During the planning era, the income of women was almost as high as that of men. Similarly, the earnings of employed women in urban China during the reform period on average were almost at the same level as those of men, and this did not 6

7

In our literature review we came across very few attempts to analyze the target efficiency of the various policy measures aimed at alleviating the consequences of job losses in urban China. Furthermore, there seem to have been no attempts to quantify how the various measures might have affected individuals and a household’s decision to stay out of the labor force because of the lessened negative income consequences. We cannot rule out the possibility that one part of the drop in employment in urban China reported here was due to the introduction of policy measures aimed at alleviating the consequences of job losses. In 2008, the year after the period under study here, the Law of the People’s Republic of China on Employment Contracts went into effect. It defines some categories of workers who are not to be terminated first if an enterprise lays off workers. One example is that of a sole breadwinner in a family or of a worker with at least fifteen years of seniority and who has a maximum of five years left before general retirement age.

Unemployment and the Rising Number of Nonworkers in Urban China

299

change much up to the first years of the new millennium (Chi and Li 2008; D´emurger, Fournier, and Chen 2007; Gustafsson and Li 2000). However, there have been signs of a rapid increase thereafter, as reported in Chapter 11. The Chinese social insurance systems provide benefits for being a worker, not for being a caregiver. When the income tax is assessed, it is based on how much the individual worker earns, without consideration of the household characteristics. In all respects, the situation is similar to the Nordic prototypes of the dual-earner model. In other respects, however, gender roles and the role of the family in urban China have elements of the breadwinner model. Parents in urban China cannot depend on heavily subsidized out-of-home child care or publicly funded home care for the elderly, services that tend to otherwise be women’s unpaid work. Similar to the situation in southern Europe, and different from that in the Nordic countries, adult children in China typically remain members of the parental household until they marry and become parents. This means that there are considerable interhousehold transfers as well as a sizable demand for caregiving within urban households. It also means there is a strong demand for financial redistribution within households, as housing, food, and education expenses are generally borne by the parents. In contrast, in the Nordic countries where adult children form their own households shortly after completing middle school, most living expenditures for college students are funded by state loans and stipends and there are no tuition fees, thereby limiting the need for parental funding. Women (particularly married women) in urban China, as elsewhere, can be assumed to structure their lives to meet expectations to perform caregiving and thus they do not seek full-time paid work during all phases of their lives (Maurer-Fazio et al. 2009; Zhang, Hannum, and Wang 2008). Employers and potential employers uphold this stereotype and assume that female workers are less productive; therefore, males are given preferential treatment in terms of hiring and in terms of retaining their jobs. It is thus not surprising that during periods of restructuring, spells of unemployment were longer for women than for men (Du and Dong 2009). The same authors, in an effort to weigh the relative importance of different factors leading to longer unemployment spells among women, conclude that structural and institutional factors play a more decisive role than gender preferences in contributing to gender disparities. At the household level, the rapid withdrawal of middle-aged women, as opposed to middleaged men, from paid employment in urban China has led to a decrease in the contribution of women to household income, thereby possibly weakening women’s bargaining power within the household (Li et al. [2006]

300

Bj¨orn Gustafsson and Ding Sai

City Employment Rates in Deciles (%)

110 100 90 1988 1995 2002 2007

80 70 60 50 40 1

2

3

4

5

6

7

8

9

10

Deciles

Figure 8.1. City Employment Rates By Deciles, 1988, 1995, 2002, and 2007.

for the 1988–1999 period; Li and Gustafsson [2008] for the 1995–2002 period). Finally, we turn to the issue of spatial differences. It is well known that the process of restructuring has hit some locations in urban China rather hard, for example, the northeastern part of the country, resulting in many job losses but the creation of relatively few new jobs. In other locations, such as the large metropolitan cities where local governments have more resources to combat job losses, employment problems, when viewed in terms of population size, have been less serious. Figure 8.1, derived from our data described in the next section, reports on the dispersion in employment rates at the city level. It shows that urban employment rates widened notably from 1995 to 2002. The data also show that this trend continued during the HuWen period here. The Gini coefficient computed for city employment rates stood at 0.031 in 1988 and 0.032 in 1995 but increased to 0.060 in 2002 and rose further to 0.093 in 2007.8

III. The Data and Descriptions of Nonworkers Our data come from the urban household surveys collected from the China Household Income Project (CHIP) for 1988, 1995, 2002, and 2007. Although there were six years between the first three pairs of years, there 8

Future research should look at whether the development of increased dispersion in employment rates across cities, as reported here, can be found in other data as well, as the sampling strategies for cities are not identical across surveys.

Unemployment and the Rising Number of Nonworkers in Urban China

301

were four years between the last two years. We have worked with data collected in the same provinces for all four years: Beijing, Shanxi, Liaoning, Yunnan, Gansu, Jiangsu, Anhui, Henan, Hubei, Guangdong, Chongqing, and Sichuan. These samples are subsamples taken from the larger samples that the National Bureau of Statistics (NBS) uses when collecting the official household statistics that are published in the annual Statistical Yearbook of China. The NBS also carried out the fieldwork. The target population for the samples was registered urban residents, and did not include rural-urban migrants who did not have urban registrations. Our samples were collected from a varying number of cities: 158 cities in 1988, 69 in 1995, 77 in 2002, and 219 in 2007.9 Further details on the sampling procedures are provided in Eichen and Zhang (1993) for the 1988 survey, Li et al. (2008) for the 1995 and 2002 surveys, and the Appendix to this volume for the 2007 survey. To a large extent, the same or similar questions were asked in all four surveys, but there were also certain differences. For example, in 2007 there were fewer questions about individual characteristics. Most of our analysis focuses on women between the ages of eighteen and fifty-five and men between the ages of eighteen and sixty. This provided us samples of 20,426 persons for 1988, 14,238 persons for 1995, 14,304 persons for 2002, and 13,808 persons for 2007. Using our definition of work-active ages, we report that in 1988, only 6 percent were nonworkers. However, this number increased to 15 percent in 1995, to 29 percent in 2002, and to 36 percent in 2007. Much of the reduction in paid work takes place at the beginning and end of one’s working life. Figure 8.2 illustrates the situation for young males during each of the four years. In 1988, 50 percent of an age cohort worked for pay at age seventeen, whereas in 1995, it was age twenty, and in 2002, it increased to age twenty-three. However, between 2002 and 2007 the increase in the age at which 50 percent of an age cohort was employed was not more than one year, indicating that the speed was slowing. This means that over the two decades under study, the average age for entering the workforce rose by as much as seven years. The graphs for women (Figure 8.3) are rather similar to those for men in all the years under study. Continuing to persons over the age of thirty in Figures 8.4 and 8.5, we find many examples of a sizable reduction in employment rates beginning in 1995. Initially, the reductions were larger among women than among men. The developments for older workers are particularly interesting. Although 9

Following the practice of most users of the CHIP data, we do not use sample weights. Although the sample sizes are more or less constant across provinces, the sizes of the populations are not.

302

Bj¨orn Gustafsson and Ding Sai

Percentage Males Working

120 100 80 60 40 20 0 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 Age 1998

1995

2002

2007

Figure 8.2. Percentage of Workers among Males between the Ages of Sixteen and Thirty, 1988, 1995, 2002, and 2007. Source: Authors’ calculations from the CHIP.

Percentage Females Working

there were changes, they did not (as among the young adults) represent a twenty-year trend. Although the age at which half of an age group left employment, which can be considered “the real retirement age,” decreased from 1995 to 2002 by two years among men and women alike; by 2007, the real retirement age had returned to age sixty-one for men and age fifty-two for women, almost identical to the real retirement rates in 1988. To sum up, Figures 8.2 to 8.5 show that the increase in nonwork in urban China from 1988 to 2007 was due to the considerably higher age at which young adults became economically active, in combination with a larger proportion of individuals of many age brackets (particularly women) not 120 100 80 60 40 20 0 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 Age 1988

1995

2002

2007

Figure 8.3. Percentage of Workers among Females between the Ages of Sixteen and Thirty, 1988, 1995, 2002, and 2007. Source: Authors’ calculations from the CHIP.

Unemployment and the Rising Number of Nonworkers in Urban China

303

120.00% 100.00% 80.00% 1988 1995 2002 2007

60.00% 40.00% 20.00% 0.00%

30

33

36

39

42

45

48

51

54

57

60

Age

Figure 8.4. Percentage of Workers among Males between the Ages of Thirty and SixtyTwo, 1988, 1995, 2002, and 2007. Source: Authors’ calculations from the CHIP.

working. However, as observed from 1988 to 2002, the trend toward a lower age when exiting the labor force, the real retirement age, was only temporary, not a long-run phenomenon. These changes are consistent with how the labor-market situation developed in urban China during those years. One can speculate that as the health of the Chinese population continues to improve, the age at which people exit the labor force in the future will increase rather than decrease. In what kinds of activities are nonworkers of work-active age involved? Focusing on their main activities as reported by the respondents to the questionnaires, we define the following: “early retired” means that the person 120.00% 100.00% 80.00%

1988 1995 2002 2007

60.00% 40.00% 20.00%

56

52 54

48

50

42

44 46

38 40

34

36

30 32

0.00% Age

Figure 8.5. Percentage of Workers among Females between the Ages of Thirty and FiftySeven, 1988, 1995, 2002, and 2007. Source: Authors’ calculations from the CHIP.

304

Bj¨orn Gustafsson and Ding Sai

receives some kind of early retirement benefit and is not working, studying, unemployed, or a homemaker. The fifth state is a residual category including, for example, people who cannot work for temporary or permanent health reasons. The operational definitions can be found in the notes to Table 8.1 in which we report the frequencies of the five categories for each of the four years under study. In Table 8.1, we see that in 1988 most of the nonworkers belonged to the residual category, whereas less than 1 percent of the population between the ages of eighteen and fifty-five or sixty belonged to each of the other four categories. The ongoing increases in the proportion of students were particularly rapid, reaching 11 percent by 2007. The proportion of unemployed increased rapidly between 1995 and 2002, after which it decreased, but to a level higher than that in 1995. The proportion of early retired rose to 4 percent in 2002,10 whereas the proportion of respondents labeled as homemakers increased continuously, reaching 7 percent in 2007. As a consequence, measured in this way signs of a reemergence of traditional gender roles in urban China are visible, though not massive, during the two decades under study here. In Table 8.2 we report the five rates of nonwork for persons between the ages of eighteen and twenty-nine, thirty and forty-five, and forty-six and older by gender for each year of the study. Unsurprisingly, students are concentrated in the youngest category, which in 1995, 2002, and 2007 also had a relatively large proportion of unemployed persons. Another unremarkable finding is that there are no early-retired persons in the youngest age category in any year. Shifting to persons between the ages of thirty and forty-five, we find that in 1995, 2002, and 2007 unemployment was the most frequent state of nonwork. In contrast, among those forty-six and older we find unemployment to be the third-most-frequent state of nonwork. Instead, the residual category, followed by early retirement, ranks highest among this group in all three years.11 From Table 8.2, we also learn predictably that being a homemaker is predominantly a female activity. By convention, the unemployment rate is defined as the percentage of not employed persons who are actively searching for a job in the labor force. Some unemployed register at employment offices, others do not. The 10

11

As questions on early retirement in the 2007 survey were not phrased identically to the corresponding questions in the previous surveys, it is difficult to judge whether the small reduction in the rate for 2007, as reported in the table, is real. Although this is true for all subgroups and years under investigation, our estimate of 10 percent of males aged thirty to forty-five in 2007 appears high and is possibly due to how we have constructed the classification (see note to Table 8.1).

Table 8.1. Categories of nonworkers 1988, 1995, 2002, and 2007, men ages eighteen to sixty and women ages eighteen to fifty-five 1988

1995

2002

2007

Persons in Persons in Persons in Persons in Proportion the sample Proportion the sample Proportion the sample Proportion the sample

305

Early retired Unemployed Students Homemakers Others Total nonworkers Unemployment rate defined as: unemployed/(worker + unemployed)

0.77 0.33 0.56 0.48 4.36 6.50 0.38

157 67 114 97 890 1,325 17,670

1.73 2.96 3.92 1.19 5.48 15.28 3.37

246 421 558 170 780 2,175 12,484

4.45 9.24 6.02 2.11 7.31 29.06 11.53

637 1,322 851 302 1,045 4,157 11,387

4.16 5.38 11.21 6.67 9.47 36.89 7.86

575 743 1,548 921 1,307 5,094 9,457

Notes: In the surveys for 1988, 1995, and 2002 there are ten alternative responses to the question on employment. We classify those who reported being employed as “employed,” students as “students,” and those who stated retirement and were between the ages of forty and fifty as “early retired.” Persons who indicated one of the three alternatives, laid off (xiagang), left post (ligang), or unemployed (shiye), are classified as “unemployed.” The residual category consists of all other nonworkers. In the 2007 survey there are fourteen alternative responses to the question on employment. We define “employed” as those who indicated one of six alternatives: working in a state-owned unit, working in a collective unit, self-employed or owner of a private business, employee in a private business, employed after having been retired, and other employed. Those who communicated that they were students were classified as “students,” together with persons with an undergraduate degree who were younger than age twenty-five, but indicated a status other than student. The motivation for this is that the Chinese term, zai xiao xuesheng, might be interpreted to refer to enrolled students. People who referred to themselves as homemakers were classified as “homemakers,” together with those who were younger than age forty-six, married, and indicated that they were retired. This is motivated by the fact that those employed in state-owned enterprises who started to work at age sixteen needed a thirty-year working career to be eligible for early retirement benefits and that people who were under age forty-six in 2007 were unlikely to have been bought out with a lump sum from their work unit. The early retired are persons who indicated this alternative, with the exception of those described previously. People who indicated one of three alternatives, waiting for employment, unemployed, or waiting for further studies, were classified as unemployed. The residual category consists of all other nonworkers older than the age of eighteen and younger than the age of forty who were not married and who referred to themselves as retired. Source: Authors’ calculations from the CHIP.

Table 8.2. Nonworkers by category, age, and gender, 1988, 1995, 2002, and 2007 (percentage of persons in different age categories) 1988

1995

2002

2007

306

Males

Females

Males

Females

Males

Females

Males

Females

18–29 Students Unemployed Unemployment rate Homemakers Other nonworkers Observations

2 1 1.27 0 1 2,749

2 1 1.05 0 1 2,954

20 10 11.98 0 2 1,584

18 9 10.87 0 3 1,667

31 17 33.77 0 3 1,345

29 14 22.19 0 6 1,430

36 6 11.16 3 6 2,383

31 8 15.29 6 12 2,267

30–45 Unemployed Unemployment rate Early retired Homemakers Other nonworkers Observations

1 0 0 0 0 4,464

1 1.21 0 1 0 4,891

10 10.22 0 0 0 3,441

9 20.16 1 1 1 3,770

6 6.12 0 0 1 3,027

12 15.17 4 1 2 3,393

4 4.54 0 10 1 2,425

7 8.89 0 15 1 2,608

46+ Unemployed Unemployment rate Early retired Homemakers Other nonworkers Observations

0 0 2 0 10 2,904

0 0.06 4 2 24 1,944

0 0.26 4 0 13 2,235

1 0.88 9 7 26 1,541

6 6.78 9 0 9 2,861

6 11.08 13 6 29 2,248

3 3.82 12 0 11 2,304

3 6.65 16 3 33 1,821

Note: For definitions of the various nonwork states, see Table 8.1. The unemployment rate is defined as: the number of unemployed divided by the sum of the number of unemployed and persons working, expressed in percent. Source: Authors’ calculations from the CHIP.

Unemployment and the Rising Number of Nonworkers in Urban China

307

labor force is defined as the sum of the employed and unemployed. This means that the number of people who are not working for reasons other than unemployment do not affect the numerical value of the unemployment rate. Applying this definition to our data, we find that in 1988, the Chinese urban unemployment rate was 0.4; in 1995, it increased to 3.3 percent; and in 2002, it jumped to 11.5 percent.12 Thereafter, the unemployment rate declined to 7.9 percent in 2007, still a level more than twice as high as that in 1995. This number is also twice as high as the registered unemployment rate of 4.0 percent (NBS China Statistical Yearbook various years) in the same year, indicating that a considerable proportion of workers searching for a job were not registered at employment offices. Our measurement of unemployment in urban China has not been harmonized with the definitions used for the OECD countries. However, we find it interesting to note that taken at face value an unemployment rate in urban China of 7.9 percent in 2007 is within the range, or slightly above, the mean of the unemployment rates observed in the same year for the OECD countries.13 Figure 8.6 shows the unemployment rates for men by age and Figure 8.7 shows the unemployment rates for women by age. We report that among middle-aged and older workers, the unemployment rates for women are higher than are those for men. On changes over time, we report that the unemployment rates increased for all age groups up to 2002. The highest unemployment rates were reported among young adults in 2002, as high as 22 percent for men and 34 percent for women. In contrast, the unemployment rates among people ages forty-six and older were not higher than 7 percent for men and 11 percent for women in 2002.

IV. Processes Leading to Nonwork In order to better understand the factors leading to various states of nonwork, we conducted statistical analyses focusing on 1995, 2002, and 2007, when the largest numbers of nonworkers are observed.14 We split each of 12 13 14

The number for 2002 is rather close to the 11.1 percent reported by Giles, Park, and Zhang (2005) from samples of five large cities. “OECD Unemployment Rate Stable at 8.8% in November 2009,” January 11, 2010, http://www.oecd.org/dataoecd/30/61/44367840.pdf. Accessed July 23, 2011. Using the CHIP data for urban China in 1988, 1995, and 2002 Liu (2012) estimates the probability models for belonging to the labor force (employed or unemployed). In addition, she estimates the probability models of being employed for persons belonging to the labor force. The analysis presented in this section differs by defining different categories of nonworkers and by considering, for example, the employment rate in the city as an explanatory variable. Furthermore, we also include 2007.

308

Bj¨orn Gustafsson and Ding Sai

20 18 16 14 12 Unemployment 10 Rate (%) 8 6 4 2 0

18-29

29-45

46-60

18-60

Age 1988

1995

2002

2007

Figure 8.6. Unemployment Rates among Men, by Age, 1988, 1995, 2002, and 2007 (percentage of the labor force). Source: Authors’ calculations from the CHIP.

20 18 16 14 12

Unemployment 10 Rate (%) 8 6 4 2 0

18–29

29–45

46–60

18–60

Age 1988

1995

2002

2007

Figure 8.7. Unemployment Rates among Females, by Age, 1988, 1995, 2002, and 2007 (percentage of the labor force). Source: Authors’ calculations from the CHIP.

Unemployment and the Rising Number of Nonworkers in Urban China

309

the samples into two: one sample with persons under the age of thirty and the other with persons aged thirty and older. This division is partly motivated by the fact that students were observed only among the young adults and there were no early retirees in this group. Furthermore, the processes leading to nonwork can be assumed to be different in certain respects between young adults and the other groups. What kinds of patterns do we expect to find? Starting with the young adults, we hypothesize the existence of intergenerational links in education, so that the probability of being a student is positively related to the education of the parents. Furthermore, we hypothesize that the activity of being a student is more probable in cases where the city employment rate is low. The probabilities of being a student or of being unemployed, respectively, are hypothesized to decrease by age. We do not expect to find clear gender differences in these probabilities. Among middle-aged and older workers, we expect to find a gender difference, meaning that being a female elevates the probability of being in all states of nonwork, and this probability was the greatest in 2002. The presence of children or of an elderly member in the household is expected to increase the probability of being a homemaker. Education is hypothesized to negatively affect the probability of belonging to various states of nonwork, particularly in 2002, whereas the city employment rate is hypothesized to negatively affect the probability of belonging to various categories of nonworkers. The statistical analysis consists of estimating multinomial logit models with employed persons as the omitted category. For young adults, we define three states of nonworkers: students, unemployed, and other; for the latter group, there are relatively few observations and a very small number of persons indicating they are homemakers. Among people aged thirty and older, we define the four states of nonworkers: unemployed, early retired, homemakers, and other. Some of the explanatory variables are the same for both samples: age, gender, and the employment rate in the city where the person resides (with the latter computed from the data). In our analyses of the young adults we also include the average years of education of the parents as explanatory variables. The specifications for persons aged thirty and older include a variable indicating the number of years of education of the person as well as a variable indicating the number of years of education of the spouse.15 Furthermore, there is one dummy variable indicating the 15

The small number of households where there is no spouse is omitted from the statistical analysis.

310

Bj¨orn Gustafsson and Ding Sai

presence of a child in the household and another dummy variable indicating the presence of a person aged sixty-five or older in the household. Table 8A.1 in the Appendix presents descriptive statistics for the variables in the analysis of young adults, and Table 8A.2 presents descriptive statistics for the variables in the analysis of middle-aged and older adults. Table 8.3 reports the estimates for young adults. We find that, as hypothesized, for all years in this study the probability of being a student, as well as the probability of being unemployed, decreases with age. Similar results are found in 2002 and 2007 with respect to the probability of belonging to the “other” category. There are no indications that being a female affected the probability in any year of belonging to any of the nonwork categories. In all the years of the study, we find, not surprisingly, that the employment rate in the city where the respondent resides negatively affects not only the probability of being unemployed but also, and more interestingly, the probability of being a student and the probability of belonging to the residual category. As hypothesized, we find the level of the parents’ education increases the probability that a young adult is a student. There are also indications that in 1995 and 2002 the parents’ level of education reduced the probability of being unemployed. Figure 8.8 illustrates some key findings about the probability that young adults with certain characteristics would belong to the various states of nonwork in each of the three years of the study. The first panel shows the probability of studying. In 1995 this probability is 41 percent for a twenty-year-old man having typical characteristics (person A), whereas it is 18 percent for a twenty-three-year-old male (person B). However, in 2007 these two predicted probabilities are as high as 75 percent and 44 percent, respectively. The figure also illustrates that the predicted probabilities vary according to the city employment rate for a typical twenty-year-old man (panels C and D) as well as according to the parents’ level of education (persons E and F). Measured in this way, the variation due to the parents’ education level has a greater influence on the probability of studying than on the variation in the city employment rate. The second panel in Figure 8.8 shows how age, the parents’ level of education, and the city employment rate affect the probability of being unemployed for a typical man. It is worth noting that comparisons between 2002 and 2007 in Figure 8.8 show a rapid decrease in the probability of unemployment, as opposed to Figures 8.6 and 8.7, which show an increase in the unemployment rate between the same two years. This difference can be reconciled by the fact that the rapid expansion of education during the period led not only to a decrease in the number of employed young adults

Table 8.3. Determinants of various states of nonwork among persons ages eighteen to twenty-nine, in 1995, 2002, and 2007 Student

Unemployed

Others

Coefficient z-value Coefficient z-value Coefficient z-value 1995 Age of individual Gender (male = 0; female = 1) Employment rate in the city Average years of education of parents Constant Pseudo R2 Number of observations 2002 Age of individual Gender (male = 0; female = 1) Employment rate in the city Average years of education of parents Constant Pseudo R2 Number of observations 2007 Age of individual Gender (male = 0; female = 1) Employment rate in the city Average years of education of parents Constant Pseudo R2 Number of observations

−0.912 −0.2148

−20.96 −1.60

−0.213 −0.132

−7.62 −0.92

−0.008 0.387

−0.16 1.30

−0.054 0.197

−3.48 8.41

−0.0605 −0.067

−3.61 −2.91

−0.088 −0.067

−2.59 −1.51

20.535 0.2499 538

12.84

8.674

5.52

4.586

1.42

243

48

−0.816 −0.165

−24.24 −1.29

−0.189 −0.061

−7.33 −0.46

−0.173 0.084

−4.04 0.38

−0.020 0.160

−2.01 6.08

−0.039 −0.108

−3.98 −4.49

−0.109 −0.035

−7.11 −0.84

17.408 0.2635 865

16.70

7.102

7.24

9.499

6.12

325

95

−0.854 0.048

−20.95 0.29

−0.181 0.187

−4.85 1.01

−0.417 0.360

−4.99 0.81

−0.038 0.232

−3.74 5.89

−0.053 −0.055

−5.10 −1.41

−0.067 −0.165

2.76 −1.76

19.104 0.3798 716

16.06

6.688

5.39

12.159

4.57

150

22

Notes: The omitted category for the dependent variable is “employed.” The test statistic z is the ratio of the coefficient to the standard error of the respective predictor. The z value follows a standard normal distribution which is used to test against a two-sided alternative hypothesis that the coefficient is not equal to zero. The note to Table 8.1 defines the various categories, with the exception that a rather small number of persons who indicated they were homemakers are included in the category “Others.” In the 1995 sample, 1,382 are employed, 538 are students, 243 are unemployed, and 48 belong to the residual category. The work sample of 2,211 is due to missing information for some of the variables, fewer than the 3,251 who were in the sample. In the 2002 sample, 948 are employed, 865 are students, 325 are unemployed, and 95 belong to the residual category. The work sample of 2,233 is due to missing information for some of the variables, fewer than the 2,551 who were in the sample. In the sample for 2007, 688 are employed, 716 are students, 150 are unemployed, and 22 belong to the other category. Source: Authors’ estimates from the CHIP.

312

Bj¨orn Gustafsson and Ding Sai (a) Probability of Studying 90.00% 80.00% A Probability B Probability C Probability D Probability E Probability F Probability

70.00% 60.00% 50.00% 40.00% 30.00% 20.00% 10.00% 0.00% 1995

2002

2007

(b) Probability of Being Unemployed 20.00% 18.00% 16.00% 14.00%

A Probability B Probability C Probability D Probability E Probability F Probability

12.00% 10.00% 8.00% 6.00% 4.00% 2.00% 0.00% 1995

2002

2007

Figure 8.8. Predicted Probabilities of Various Rates of Nonwork among Persons between the Ages of Eighteen and Twenty-Nine, 1995, 2002, and 2007. Notes: A is a male aged twenty, living in a city with an employment rate at the sample mean and whose parents’ level of education is at the sample mean. B. Differs from person A by being aged twenty-three. C. Differs from person A by living in a city with an employment rate equal to that observed in the top decile during the same year. D. Differs from person A by living in a city with an employment rate equal to that observed in the bottom decile during the same year. E. Equal to person C but the level of education of the parents is that observed in the top decile during the same year. F. Equal to person C but the level of education of the parents is that observed in the bottom decile during the same year. Source: Authors calculations from the CHIP.

Unemployment and the Rising Number of Nonworkers in Urban China

313

but also, and less rapidly, to a decrease in the number of unemployed young adults. The figure also illustrates substantial differences in the probability of being unemployed based on the city employment rate. It should be noted that the predictions indicate that in 1995 and 2002, unemployment problems among twenty-year-olds were concentrated among the offspring of parents with low levels of education. We now turn to the estimates for persons between the ages of thirty and fifty-five to sixty, as reported in Table 8.4. We find that some coefficients are estimated with high z-values for all states for almost all the years under study. This is the case for the positive coefficients for being female and for the negative coefficients for the city employment rate. This is also the case for the negative coefficients for one’s own education. The coefficients for the variables for being female and for one’s own education are typically lower when based on the 2007 data than they are when based on the earlier data. In contrast, evidence that the level of the spouse’s education will affect the state of nonwork is most clear in the 2007 data. Another clear result is that the presence of a child in the household increases the probability of being a homemaker, whereas this is not the case with the presence of an elderly person in the household. Furthermore, in all years, positive coefficients for the age variable are estimated with high z-values for both the early retired and the residual categories.16 Figure 8.9 illustrates some key results for the probabilities for a fifty-yearold man or woman with low or high levels of education, for 1995, 2002, and 2007. It also illustrates the variation due to the city employment rate. The first panel shows a more rapid drop in the probability of employment for women with a low level of education than for men with a high level of education between 1995 and 2002. Although between those years, there was a widening employment gap between men and women with low levels of education, between 2002 and 2007 the gap narrowed. In 2007 a woman with a low level of education is predicted to have a clearly higher probability of employment than she did in 2002, and in 2007 the probability of employment for a man with a high level of education is actually lower than in 2002. Finally, Figure 8.9 shows that the spatial gap in employment probabilities, consistent with the descriptive results reported in Figure 8.1, continuously widened. 16

The coefficients indicate that age positively affects the “homemaker” state in 1995 and 2002, but negatively affects it in 2007. One can speculate whether this may be a reflection that the label “homemaker” changed over time, making it more socially acceptable for middle-aged women to select this alternative in their responses to the questionnaire.

Table 8.4. Determinants of various states of nonwork among persons ages thirty to fifty-five or sixty, 1995, 2002, and 2007 Unemployed

314

Coefficient

z-value

1995 Age Female Individual years of education Dummy for children in the household Dummy for elderly in the household Education of spouse City employment rate Constant Pseudo R2 Number of observations

−0.039 0.775 −0.184 0.752 0.755 −0.013 −0.069 3.737 0.3464 98

−1.74 3.37 −4.56 2.45 2.48 0.32 3.04 1.68

2002 Age Female Years of education of the individual Dummy for children in the household Dummy for elderly in the household Education of spouse City employment rate Constant Pseudo R2 Number of observations

−0.011 0.872 −0.193 0.098 0.397 −0.038 −0.056 3.863 0.2775 764

−1.34 10.31 −12.77 0.92 2.96 −2.62 −10.35 6.95

Early retired Coefficient 0.146 1.241 −0.174 −0.958 −0.128 0.003 −0.086 −1.797

z-value 9.54 7.24 −7.03 −4.24 −0.47 0.11 −5.34 −1.13

225 0.173 1.318 −0.151 −0.387 0.120 −0.023 −0.081 −5.103 608

Homemaker Coefficient

z-value

Coefficient

z-value

0.238 4.190 −0.356 0.635 −0.699 −0.056 −0.067 −9.285

11.35 8.69 −10.65 2.76 −1.58 −1.83 −3.09 −4.21

0.496 2.877 −0.175 0.505 −0.252 −0.009 −0.052 −22.943

26.78 17.07 −8.81 3.01 −1.13 −0.47 3.64 −14.61

130 17.80 12.96 −8.75 −2.51 0.69 −1.39 −12.31 −7.02

Others

0.117 3.568 −0.388 1.194 −0.101 −0.014 −0.058 −7.398 251

495 8.83 12.75 −17.07 7.05 −0.35 −0.58 −6.24 −6.74

0.580 4.044 −0.195 0.171 0.261 −0.008 −0.060 −31.652 831

31.43 26.46 −10.83 1.07 1.38 −0.48 −8.50 −26.04

2007 Age Female Years of education of the individual Dummy for children in the household Dummy for elderly in the household Education of spouse City employment rate Constant Pseudo R2 Number of observations

0.0070 0.2896 −0.1914 0.0009 0.0946 −0.0525 −0.0618 2.3994 0.2919 85

0.37 1.10 −4.44 0.01 0.18 −1.19 −5.12 1.94

0.107 0.677 −0.080 −0.295 0.222 −0.138 −0.059 −1.859 250

8.02 4.11 −2.88 −1.40 0.70 −5.37 −7.45 −2.16

−0.1987 0.5758 −0.0878 −1.7062 0.4714 −0.1689 −0.0671 12.8326 252

−15.38 3.58 −2.92 −8.86 1.29 −5.93 −8.56 15.65

0.651 3.686 −0.167 0.189 0.456 −.044 −0.056 −32.248

17.55 14.99 −4.85 0.67 1.22 −1.38 −5.67 −15.27

281

315

Notes: The ages are from thirty to sixty for males and thirty to fifty-five for females. The note to Table 8.1 defines the various categories, with the exception that a rather small number of persons who indicated they were homemakers are included in the category “other.” In the 1995 sample, 8,945 are employed, 98 are unemployed, 225 are early retired, 130 are homemakers, and 495 belong to the residual category. The work sample consists of 9,893 due to missing information for some of the variables, fewer than the 10,987 who were in the sample. In the 2002 sample, 8,171 are employed, 764 are unemployed, 608 are early retired, 251 are homemakers and 831 belong to the residual category. The work sample consists of 10,625 due to missing information for some of the variables, fewer than the 11,753 who were in the sample. In the 2007 sample, 3,077 are employed, 85 are unemployed, 250 are early retired, 252 are homemakers, and 281 belong to the residual category. Source: Authors’ estimates from the CHIP.

316

Bj¨orn Gustafsson and Ding Sai (a) Probability of Being Employed 100.00% 90.00%

A Probability

80.00% 70.00%

B Probability

60.00%

C Probability

50.00%

D Probability

40.00% 30.00%

E Probability

20.00%

F Probability

10.00% 0.00% 1995

2002

2007

(b) Probability of Being Unemployed 12.00% 10.00%

A Probability B Probability

8.00%

C Probability

6.00% D Probability

4.00%

E Probability F Probability

2.00% 0.00% 1995

2002

2007

Figure 8.9. Predicted Probabilities of Employment and Various States of Nonwork among Persons Aged 50, 1995, 2002, and 2007.

The second panel in Figure 8.9 shows that the probability for being unemployed among persons aged fifty was very low in 1995 but thereafter increased substantially for both men and women with lower levels of education and continued to be low for those with higher levels of education. The figure also shows that the expansion of unemployment had a very clear spatial character that narrowed in 2007. Compared to differences due to education and space, gender differences in the probability of being unemployed are small. The main impression from the third panel in Figure 8.9 is that over time-spatial differences in the probability of retiring early were more important than were gender and educational differences.

Unemployment and the Rising Number of Nonworkers in Urban China

317

(c) Probability of Being Early-Retired 16.00% 14.00% A Probability

12.00% B Probability

10.00% C Probability

8.00% D Probability

6.00% E Probability

4.00%

F Probability

2.00% 0.00% 1995

2002

2007

Figure 8.9 (continued) Predicted Probabilities of Employment and Various States of Nonwork among Persons Aged 50, 1995, 2002, and 2007. Notes: Person A is a male aged fifty with a short education. There are no children or elderly persons in the household. Education of the spouse is at the sample mean, as is the city employment rate. Person B has all characteristics identical to person A with the exception that his education is long. Person C is identical to person A with the exception of being female. Person D is identical to person A but female and has a long education. Person E is female aged fifty with an average education. There are no children or elderly persons in the household. Education of the spouse is at the sample mean. She lives in a city with a low employment rate (the mean for the bottom of the city employment rates). Person F has characteristics identical to person E with the exception that the city employment rate is high (the mean of the top decile of the city employment rates). Source: Table 8.4.

What do the changed employment patterns imply for gender roles? We address this question by looking at the extent to which husbands and wives contributed to the couple’s earnings during the period under study. Over time, the contributions change not only due to changed employment patterns, but also due to changed remuneration and marriage patterns. The trend in Table 8.5 is consistent with the results of the studies we refer to in Section II, studies that capture much, but not all, of the period studied here. Whereas at the beginning of the period three-fifths of the couples’ earnings were earned equally by the husbands and wives, the proportion declined to two-fifths in 2007. This development was primarily driven by the fact that a large proportion of the wives earned less than their husbands, a development most conspicuous between 1995 and 2002. By comparison, the accelerated increase in the proportion of wives who did not have any earnings is of lesser importance, but it is still visible.

318

Bj¨orn Gustafsson and Ding Sai Table 8.5. Economic dependency of married women in urban China, 1988, 1995, 2002, and 2007 (ages eighteen to sixty for males and ages eighteen to fifty-five for females)

Mean dependency Woman has no earnings (%) Woman earns less than man (%) Equality in earnings (%) Woman earns more than man (%) Woman as sole earner (%)

1988

1995

2002

2007

0.03 0.79 25.92 63.09 10.99 0.09

0.118 1.03 30.43 60.93 8.64 0.14

0.158 2.11 41.57 46.47 11.96 0.62

0.216 5.64 44.50 42.60 9.06 0.71

Note: Equality is defined as each partner contributing between 40 and 60 percent of the combined earnings. Source: Authors’ calculations from the CHIP.

V. The Economic Well-Being of Nonworkers In this section we analyze the economic well-being of nonworkers. For this purpose, and following usual practice when analyzing the distribution of household income, we construct the variable disposable equivalent income for each household in our data set by adding the personal income of all the household members. We also add the income received at the household level, including income from owner-occupied housing and the minimum living standard guarantee (dibao) program.17 Taxes and transfers are entered by a negative sign. During a second step, we divided the household income by the number of persons in the household to arrive at the variable for the disposable household per capita income, which we assign to all household members. In this manner, we obtained a variable defined for individuals of all ages, a variable that considers the economic situation of the household. During the next step, we examined how nonworkers were represented in the various deciles of this variable in 1988, 1995, 2002, and 2007. Figure 8.10 shows that in addition to the levels of nonworkers increasing across the surveys, the main difference is that the profile for 2002 deviates from those for all other years by having a very clearly downward slope. In 2002 nonworkers were concentrated in the lower deciles. However, in 2007 the profile is again more horizontal. The increase in nonwork thus changed from a force toward greater inequality to a force toward greater equality. In order to better understand this finding, we disaggregated the nonworkers 17

For 2002 and 2007, we use the market rate approach for valuing the imputed rent of owner-occupied housing, as described in Chapter 3. For 1995, we use the market rate approach, based on our own calculations.

Percentage (%)

Unemployment and the Rising Number of Nonworkers in Urban China 50.00 45.00 40.00 35.00 30.00 25.00 20.00 15.00 10.00 5.00 0.00

319

1988 1995 2002 2007

1

2

3

4

5

6

7

8

9

10

Deciles of Disposable Household Per Capita Income

Figure 8.10. Percentages of Nonworkers by Decile of Disposable Household Per Capita Income, 1988, 1995, 2002, and 2007. Source: Authors’ calculations from the CHIP.

into our five categories and reported the results in separate figures for each year. The main finding from Figures 8.11, 8.12, 8.13, and 8.14 is that although some categories of nonworkers are rather evenly spread out over the distribution of household income, others are concentrated at the lower household incomes. Remarkably, the growing category of students is evenly represented in all deciles. This is also the case for the residual category as well as for the early retired. The other pattern is represented by the unemployed and

8.00%

Unemployed

7.00%

Student

6.00% Early Retired

5.00% Homemaker

4.00%

Residual Category

3.00% 2.00% 1.00% 0.00% 1

2

3

4

5

6

7

8

9

10

Deciles of Disposable Household Per Capita Income

Figure 8.11. Percentage of Various Categories of Nonworkers by Decile of Disposable Household Per Capita Income, 1988. Source: Authors’ calculations from the CHIP.

320

Bj¨orn Gustafsson and Ding Sai

8.00%

Unemployed

7.00% Student

6.00%

Early Retired

5.00% 4.00%

Homemaker

3.00% Residual Category

2.00% 1.00% 0.00% 1

2

3

4

5

6

7

8

9

10

Deciles of Disposable Household Per Capita Income

Figure 8.12. Percentage of Various Categories of Nonworkers by Decile of Disposable Household Per Capita Income, 1995. Source: Authors’ calculations from the CHIP.

the homemakers, two categories both disproportionally located in the lowest deciles. It thus appears that, to a large extent, the label “homemaker” functioned as an alternative to the category “unemployed.” However, there are also differences across years. In 2002 as many as 57 percent of the unemployed and 59 percent of the homemakers were located in the four lowest deciles, whereas only 2 percent were located in the top decile. During that year, the average household disposable household 25.00%

Unemployed Students

20.00%

Early Retired

15.00%

Homemaker Residual Category

10.00% 5.00%

0.00% 1

2

3

4

5

6

7

8

9

10

Deciles of Disposable Household Per Capita Income

Figure 8.13. Percentage of Various Categories of Nonworkers by Decile of Disposable Household Per Capita Income, 2002. Source: Authors’ calculations from the CHIP.

Unemployment and the Rising Number of Nonworkers in Urban China

321

25.00% Unemployed Students

20.00%

Early retired

15.00% Homemaker

10.00%

Residual Category

5.00% 0.00% 1

2

3

4

5

6

7

8

9

10

Deciles of Disposable Household Per Capita Income

Figure 8.14. Percentage of Various Categories of Nonworkers by Decile of Disposable Household Per Capita Income, 2007. Source: Authors’ calculations from the CHIP.

income for homemakers was only 66 percent of the average for the entire population, and among the unemployed, it was 68 percent (numbers calculated from Table 8.6). The concentration of the unemployed at the bottom of the income distribution is similar to the situation in 2007. The next step of the analysis focuses on personal income earned by the employed and the various categories of nonworkers in 1995, 2002, and 2007. For this purpose, we define personal income as that earned by men between the ages of eighteen to sixty and by women between the ages of eighteen to fifty-five. The means and measures of inequality for this variable, together with the variable for the household per capita disposable income, are reported in Table 8.6. The main finding is that on average the unemployed, homemakers, and students have low personal incomes, but this is not the case for the early retired and the residual category. The personal income among the unemployed, homemakers, and especially students is rather unequally distributed. The mean values for personal income among the unemployed and students are much lower than are the mean values for the disposable income for each of the categories. In contrast, employed persons on average have lower disposable income per capita than personal income. There is thus a considerable redistribution of income taking place within urban Chinese households and, as we will see, this redistribution increased over time. Note also that the negative difference between personal income and household per capita income among employed persons is, absolutely and relatively, smaller than the corresponding positive difference

Table 8.6. Personal income and household disposable per capita income among the employed and various categories of nonworkers, 1995, 2002, and 2007 (means and Gini coefficients)

322

Employed

Unemployed

Student

Early retired

Homemaker

Others

Total

1988 Personal income Household per capita income Gini index of personal income Gini index of household per capita income

1,459 1,882 0.246 0.224

736 1,238 0.275 0.228

799 1,340 0.262 0.201

1,171 1,902 0.271 0.257

533 1,264 0.553 0.249

612 2,145 0.675 0.226

1,251 1,885 0.277 0.227

1995 Personal income Household per capita income Gini index of personal income Gini index of household per capita income

6,661 5,841 0.309 0.324

2,933 4,102 0.530 0.342

750 5,653 0.799 0.343

4,884 5,336 0.324 0.297

1,268 4,555 0.577 0.463

5,258 6,243 0.308 0.345

6,453 5,780 0.320 0.330

2002 Personal income Household per capita income Gini index of personal income Gini index of household per capita income

12,100 8,636 0.347 0.319

3,771 5,503 0.565 0.314

1,647 8,288 0.714 0.325

8,384 7,836 0.311 0.282

1,952 5,219 0.684 0.310

8,821 8,975 0.301 0.303

11,011 8,243 0.375 0.324

2007 Personal income Household per capita income Gini index of personal income Gini index of household per capita income

22,283 16,703 0.379 0.327

3,074 14,866 0.757 0.355

10,895 16,561 0.773 0.326

16,364 17,788 0.325 0.323

13,878 14,349 0.480 0.386

14,199 17,515 0.397 0.309

18,400 16,478 0.463 0.331

Note: In current prices. Source: Authors’ calculations from the CHIP.

Unemployment and the Rising Number of Nonworkers in Urban China

323

among the unemployed and students. This is because the number of workers is higher than the number of nonworkers. The rapidly increasing redistribution within households also appears in the Gini coefficient for personal income among people of work-active ages and in the Gini coefficient for household disposable per capita income among the entire population. The former increased rapidly, from 0.27 in 1988, 0.32 in 1995, 0.37 in 2002, and finally 0.46 in 2007, whereas the latter more or less remained stable, 0.23 for 1988, 0.33 for 1995, 0.32 for 2002, and 0.33 for 2007. Since 1995, the increased income inequality due to the labor market in urban China to a large extent was counteracted within the household. An important part of this story is that an ever-larger proportion of young adults have no or only small personal incomes while they are studying or are unemployed and they live with their parents and are supported by their working parents. It is important to note that a proportion of the disposable income of parents co-residing with students is used to pay school fees and other educational expenses, and thereby is not available for other expenditures. In addition, this redistribution of income taking place within the household refers to an accounting period of one year. Nevertheless, one can expect that young adults with a higher level of education will have a relatively high personal income in the future. In this way, one can expect that the higher levels of education among urban young adults in China will contribute to increased income inequality in the future. To further illustrate the importance of the redistribution of income within urban Chinese households, Table 8.7 shows the relation between the personal income of adults and the disposable household per capita income by transition matrixes using deciles for 1988, 1995, 2002, and 2007. A rather low proportion of individuals in the first decile of personal income are also in the lowest decile of disposable household income. In 2007 the proportion was as low as 21 percent, much lower than that in 1995 or 2002, indicating the increased importance of redistribution within the households during this period. About half of the individuals in the top decile of the distribution of personal income are also in the same decile in the distribution of disposable household per capita income for both years of the study. A rather low proportion, a maximum of 1 percent, of the persons located in the top decile of personal income are located in the bottom decile of household disposable per capita income. This illustrates that the redistribution of income within the household is more powerful in improving the lot of those with no or low personal income than in lowering the relative position of those with high personal incomes.

Table 8.7. Adult persons by deciles of personal income and disposable household per capita income, 1988, 1995, 2002, and 2007 Decile of Characteristic of individual household per Decile of personal income capita income

Total

324

1988 1 2 3 4 5 6 7 8 9 10 Total Number of observations

1 22.1 19.7 15.5 10.8 10.0 7.3 6.3 4.2 3.2 0.9 100 2,043

2 15.3 14.4 15.2 13.1 10.5 10.8 7.9 6.2 4.7 2.0 100 2,038

3 11.4 11.1 14.1 13.5 12.9 9.7 10.2 7.5 6.6 3.0 100 2,038

4 9.5 10.1 12.7 13.3 12.6 12.4 11.3 9.5 5.9 2.8 100 2,038

5 9.6 8.6 9.2 11.9 12.6 11.9 12.1 10.3 8.7 5.2 100 2,039

6 8.0 6.6 7.7 10.8 12.3 12.8 13.4 12.4 9.3 6.7 100 2,041

7 7.8 8.9 8.3 7.7 9.6 10.7 13.0 13.6 12.5 7.8 100 2,038

8 6.8 9.5 7.6 7.7 8.1 9.2 9.8 13.0 15.4 12.9 100 2,041

9 6.3 7.0 6.7 6.9 6.3 7.5 9.8 12.6 18.6 18.4 100 2,038

10 3.2 4.1 3.2 4.4 5.1 7.7 6.1 10.9 15.1 40.3 100 2,038

1995 1 2 3 4 5 6 7 8 9 10 Total Number of observations

33.7 25.9 17.5 7.9 6.6 3.8 2.0 2.0 0.5 0.2 100 1,325

17.8 18.9 20.0 17.2 11.2 6.4 5.1 1.8 1.3 0.3 100 1,326

11.0 13.7 14.8 17.4 15.1 12.2 8.9 4.5 2.7 0.5 100 1,324

8.7 10.3 11.7 15.2 16.9 14.9 10.4 7.5 3.8 0.7 100 1,327

7.1 8.2 10.4 11.9 14.4 16.0 14.4 9.5 5.9 2.3 100 1,323

4.8 8.5 8.6 9.6 10.4 14.7 17.8 14.3 8.9 2.4 100 1,324

5.4 5.4 5.1 8.5 8.5 13.1 14.2 19.0 15.7 5.3 100 1,325

4.2 3.6 5.7 6.0 7.7 8.1 12.5 19.9 19.3 13.0 100 1,326

3.9 3.0 3.9 4.4 5.3 6.9 10.1 14.9 21.9 25.8 100 1,325

3.6 2.4 2.3 2.0 3.9 4.2 5.2 6.9 19.9 49.7 100 1,323

100 100 100 100 100 100 100 100 100 100 20,392 100 100 100 100 100 100 100 100 100 100 13,248

325

2002 1 2 3 4 5 6 7 8 9 10 Total Number of observations

1 35.3 26.9 14.5 8.2 6.6 3.3 2.3 1.3 0.8 0.7 100 1,132

2 18.5 22.3 20.7 15.5 10.0 5.8 3.8 2.0 1.0 0.5 100 1,181

3 14.0 16.5 14.4 17.4 14.3 9.9 6.1 5.0 1.1 1.2 100 1,228

4 9.8 12.1 13.9 15.1 15.5 13.4 8.4 6.4 4.3 1.2 100 1,267

5 7.9 8.0 12.4 14.0 13.0 16.5 10.8 9.9 5.8 1.7 100 1,261

6 7 8 5.4 5.5 3.1 6.5 3.1 2.2 8.5 7.3 5.2 10.6 7.1 6.7 12.2 11.1 8.8 16.2 13.6 10.8 16.0 15.4 14.8 12.6 16.8 17.5 8.9 14.1 21.3 3.2 6.00 9.7 100 100 100 1,317 1,290 1,313

2007 1 2 3 4 5 6 7 8 9 10 Total Number of observations

1 20.6 33.7 20.3 11.8 7.0 4.2 1.9 0.4 0.0 0.2 100 1,346

2 16.6 18.3 19.4 15.6 12.7 8.6 4.8 2.8 1.0 0.2 100 1,344

3 13.9 11.2 14.9 17.3 15.6 10.8 8.9 5.0 2.1 0.3 100 1,344

4 10.6 8.8 13.2 15.2 13.2 14.5 12.5 7.8 3.4 1.0 100 1,345

5 9.1 7.9 10.2 12.8 15.0 12.4 12.9 11.7 6.7 1.4 100 1,347

6 7 7.1 7.7 5.7 3.6 8.6 6.3 9.6 7.6 11.5 9.5 13.9 13.3 15.1 14.8 14.2 16.5 11.2 15.3 3.1 5.4 100 100 1,342 1,345

Source: Authors’ calculations from the CHIP.

9 0.3 2.1 2.9 3.6 6.0 6.1 13.0 16.0 23.3 26.9 100 1,316

10 0.3 0.2 0.3 1.8 2.6 4.7 9.7 12.5 19.4 48.5 100 1,323

8 9 7.06 5.0 3.79 4.2 3.87 2.4 6.39 2.9 7.58 5.3 11.15 7.4 12.42 11.1 15.76 16.5 19.55 21.3 12.42 24.0 100 100 1,345 1,345

10 2.4 2.9 0.9 0.9 2.6 3.8 5.6 9.3 19.6 52.1 100 1,342

. 100 100 100 100 100 100 100 100 100 100 12,628 100 100 100 100 100 100 100 100 100 100 13,445

326

Bj¨orn Gustafsson and Ding Sai

VI. Conclusion In this chapter we have analyzed the growth of nonwork among residents in urban China in 1988, 1995, 2002, and 2007 using surveys covering large parts of the country. The period between 2002 and 2007 marks the first phase of the Hu-Wen leadership. We have reported employment rates by age and gender for each year. Furthermore, we have divided nonworkers into five categories: students, unemployed, early retired, homemakers, and a residual category. We have estimated probability models relating the state of the labor market to household variables and the city employment rate. In addition, we have examined the personal income and disposable income of nonworkers and workers. During the two decades under investigation we have seen large changes in terms of the age that a birth cohort begins to work. In 1988 at least 50 percent of the age cohort of young men was working at age seventeen and that of young women at age eighteen. However, the age for entering the labor force rose continuously, reaching as high as age twenty-four in 2007. This means an increase of six or seven years during the two decades under study. We also reported that among those approaching age thirty, as many as about one in five were not working in 2002 and 2007, whereas nothing similar was observed in 1988. These changes have been driven by prolonged educations as well as by the unemployment that school-leavers face before finding a job. Our results have confirmed that children with parents with high levels of education study longer. Furthermore, they show that in cities with low employment, young adults are not only more likely to be unemployed, but they are also more likely to pursue their studies. Nonwork has also increased in urban China among people aged thirty and older. Compared to 1988, a larger proportion of persons in this age group were not working for pay in 2007. The restructuring of the Chinese economy that took place beginning in the mid-1990s shows up in 2002 in the lower real retirement ages, the age one leaves the labor force for good. However, as the process of reducing employment in state-owned and urban collective units slowed down and the creation of jobs in other ownership sectors accelerated in 2007, the real retirement age bounced back to about the same as that in 1988. The changed employment prospects among middle-aged and older workers have affected women and persons with lower levels of education more than others. In 2002, nonwork had the clearest gender and education profile, a profile that was considerably weaker in 2007. Parallel to this, the dispersion

Unemployment and the Rising Number of Nonworkers in Urban China

327

in employment rates at the city level increased from 1995 to 2002, and continued to increase thereafter. Our results show that a low city employment rate leads not only to a higher probability of open unemployment, but also to a higher probability that middle-aged and upper-middle-aged persons will retire early, will be homemakers, or will belong to the residual category of nonworkers. Our results are mixed regarding the issue of changed gender roles. On the one hand, among young adults we did not find any indication that young women were less likely than young men to study. On the other hand, there were visible, but far from massive indications of a reappearance of traditional housewives. This is one part of the story of Chinese perceptions of women contributing less and men contributing more to a couple’s earnings. Another part of this story is the increased wage gap among employed workers (for more on this, see Chapter 11). A major finding in this study is that much of the income loss due to nonwork in urban China is absorbed within the household. This is consistent with findings reported in Chapter 7 that reports that impulses from the labor market were not the main contributors to the increase in income inequality at the household level during the first years of the Hu-Wen era. We have shown that in 1995, 2002, and 2007, the relationship between personal income and household disposable per capita income was weak in urban China, and it became even weaker over time. Incomes forgone by an increasingly large number of students and unemployed youth are, to a large extent, provided by parents who typically have above-average personal incomes. We have also reported that those who have left the labor force as early retired are relatively evenly spaced over the distribution of household per capita income. However, it should be noted that in 2002 and 2007, unemployed people, who were practically nonexistent in 1988 and were rather few in 1995, fared less well than did other urban residents. For policy makers who regard increased inequality in household income in urban China as a problem, the bad news is that the greater number of unemployed persons in 2002 clearly contributed to a more unequal distribution. The good news is that thereafter the employment problem was reduced in size, thereby counteracting those forces leading toward increasing inequality. However, in other respects, the trend toward more inequality persists. The dispersion in employment rates across cities has continued to increase. The wife’s share of a couple’s income has continued to decrease whereas the contribution by the husband has increased.

328

Bj¨orn Gustafsson and Ding Sai APPENDIX

Table 8A.1. Descriptive statistics for the sample of young adults, 1995, 2002, and 2007 Employed 1995 Age of individual Female (%) Employment rate in the city Average years of parents’ education Observations 2002 Age of individual Female (%) Employment rate in the city Average years of parents’ education Observations 2007 Age of individual Female (%) Employment rate in the city Average years of parents’ education Observations

Unemployed

Student

Others

23.11 46.89 84.16

21.67 46.09 83.23

19.18 48.70 84.09

22.93 51.16 82.25

9.61

8.79

10.69

8.77

1,382

243

538

48

24.04 48.52 70.67

22.83 48.92 69.05

19.80 47.63 70.17

22.93 49.44 65.20

9.64

8.86

10.5

9.54

948

325

865

95

25.42 46.66 62.01

24.4 50.0 58.04

20.25 47.91 60.21

22.82 54.55 57.15

10.71

10.44

11.84

9.91

688

Source: Authors’ calculations from the CHIP.

150

716

22

Unemployment and the Rising Number of Nonworkers in Urban China

329

Table 8A.2. Descriptive statistics for the sample of middle-aged and older workers, 1995, 2002, and 2007 Employed Unemployed Early retired Homemaker Others 1995 Age (years) Female (%) Years of education of the individual Dummy for children in the household Dummy for elderly in the household Education of spouse City employment rate Observations 2002 Age (years) Female (%) Years of education of the individual Dummy for children in the household Dummy for elderly in the household Education of spouse City employment rate Observations

41.54 46.65 10.62

38.83 69.39 9.0

48.22 66.67 8.65

47.95 96.15 6.18

53.76 64.24 8.55

58.54

78.57

13.33

31.54

17.78

6.83

13.27

7.11

4.62

7.68

10.47 85.12 8,945

9.89 83.76 98

9.74 83.19 225

8.72 83.26 130

9.57 83.56 495

42.74 42.69 11.23

42.23 61.40 9.60

49.23 56.41 9.55

44.80 94.02 7.31

53.50 71.12 9.20

39.90

39.27

9.38

40.24

12.88

6.77

9.82

7.07

5.98

7.46

10.94 71.77 8,171

9.98 69.09 764

9.80 67.96 608

9.61 69.59 251

9.89 69.52 831

45.8 25.88 10.38

49.82 26.40 10.68

38.44 34.13 11.15

55.34 45.55 10.21

27.06

13.20

17.46

10.68

4.71

5.20

3.97

4.98

10.58 56.00 85

9.94 56.93 250

10.46 54.91 252

10.19 58.44 281

2007 Age (years) 44.85 Female (%) 25.74 Years of education of the 12.22 individual At least one child in the household 30.55 (%) At least one elderly person in the 3.64 household (%) Education of spouse 11.85 City employment rate 61.29 Observations 3,077 Source: Authors’ calculations from the CHIP.

330

Bj¨orn Gustafsson and Ding Sai References

Appleton, S., J. Knight, L. Song, and Q. Xia (2002), “Labor Retrenchment in China: Determinants and Consequences,” China Economic Review, 13(2–3), 252–275. Cai, F., A. Park, and Y. Zhao (2008), “The Chinese Labor Market in the Reform Era,” in L. Brandt and T G. Rawski, eds., China’s Great Economic Transformation, 167–214, New York: Cambridge University Press. Cai, H., Y. Chen, and L-A. Zhou (2010), “Income and Consumption Inequality in Urban China: 1992–2003,” Economic Development and Cultural Change, 58(3), 385–413. Chi, W. and B. Li (2008), “Glass Ceiling or Sticky Floor? Examining the Gender Earnings Differential Across the Earnings Distribution in Urban China, 1987–2004,” Journal of Comparative Economics, 36(2), 243–263. Connelly, R. and Z. Zheng (2003), “Determinants of School Enrollment and Completion of 10 to 18 Year Olds in China,” Economics of Education Review, 22(4), 379–388. D´emurger, S., M. Fournier, and Y. Chen (2007), “The Evolution of Gender Earnings Gaps and Discrimination in Urban China, 1988–95,” Developing Economics, 45(1), 97–121. Du, F. and X. Dong (2009), “Why do Women have Longer Durations of Unemployment than Men in Post-restructuring Urban China?” Cambridge Journal of Economics, 33(2), 233–252. Duckett, J. and A. Hussain (2008), “Tackling Unemployment in China: State Capacity and Governance Issues,” Pacific Review, 21(2), 211–229. Eichen, M. and M. Zhang (1993), “Annex: The 1988 Household Sample Survey: Data Description and Availability,” in K. Griffin and R. Zhao, eds., The Distribution of Income in China, 341–346, Basingstoke: Macmillan. Gao, Q., I. Garfinkel, and F. Zhai (2009), “Anti-Poverty Effectiveness of the Minimum Living Standard Assistance Policy in Urban China,” Review of Income and Wealth, 55 (Supplement 1), 630–655. Giles, J., A. Park, and F. Cai (2006), “Reemployment of Dislocated Workers in Urban China: The Roles of Information and Incentives,” Journal of Comparative Economics, 34(3), 582–607. Giles, J., A. Park, and J. Zhang (2005), “What is China’s True Unemployment Rate?” China Economic Review, 16(2), 149–170. Gustafsson, B. and Q. Deng (2011), “Dibao Receipt and Its Importance for Combating Poverty in Urban China,” Poverty & Public Policy, 3(19), Article 10. Gustafsson, B. and S. Li (2000), “Economic Transformation and the Gender Earnings Gap in Urban China,” Journal of Population Economics, 13(2), 305–329. Hannum, E., J. Behrman, M. Wang, and J. Liu (2008), “Education in the Reform Era,” in L. Brandt and T.G. Rawski, eds., China’s Great Economic Transformation, 215–249, New York: Cambridge University Press. Hannum, E. and M. Wang (2006), “Geography and Educational Inequality in China,” China Economic Review, 17(3), 253–265. Knight, J. and L. Song (2008), “China’s Emerging Wage Structure 1995–2002,” in B.A. Gustafsson, S. Li, and T. Sicular, eds., Inequality and Public Policy in China, 221–242, New York: Cambridge University Press. Leung, J.C.B. (2006), “The Emergence of Social Assistance in China,” International Journal of Social Welfare, 15(3), 188–198.

Unemployment and the Rising Number of Nonworkers in Urban China

331

Li, H., J. Zhang, L.T. Sin, and Y. Zhao (2006), “Relative Earnings of Husbands and Wives in Urban China,” China Economic Review, 17(4), 412–431. Li, S. and B. Gustafsson (2008), “Unemployment, Earlier Retirement, and Changes in the Gender Income Gap in Urban China, 1955–2002,” in B.A. Gustafsson, S. Li, and T. Sicular, eds., Inequality and Public Policy in China, 243–286, New York: Cambridge University Press Li, S., C. Luo, Z. Wei, and X. Yue (2008), “Appendix: The 1995 and 2002 Household Surveys: Sampling Methods and Data Description,” in B.A. Gustafsson, S. Li, and T. Sicular, eds., Inequality and Public Policy in China, 337–353, New York: Cambridge University Press. Liu, Q. (2012), “Unemployment and Labor Force Participation in Urban China,” China Economic Review, 23(1), 18–33. Maurer-Fazio, M., R. Connelly, C. Lan, and L. Tang (2009), “Child Care, Eldercare, and Labour Force Participation of Married Women in Urban China: 1982–2000,” IZA Discussion Paper No. 4204, Institute for the Study of Labor, Bonn, Germany. Meng, X. (2004), “Economic Restructuring and Income Inequality in Urban China,” Review of Income and Wealth, 50(3), 357–379. Mok, K.H., Y.C. Wong, and X. Zhang (2009), “When Marketisation and Privatisation Clash with Socialist Ideals: Educational Inequality in Urban China,” International Journal of Educational Development, 29(5), 505–512. National Bureau of Statistics (NBS) (various years), Zhongguo tongji nianjian (China Statistical Yearbook), Beijing: Zhongguo tongji chubanshe. National Bureau of Statistics (NBS) (2011), Zhongguo tongji zhaiyao 2011 (China Statistical Abstract 2011), Beijing: Zhongguo tongji chubanshe. Ryan, P. (2001), “The School-to-Work Transition: A Cross-National Perspective,” Journal of Economic Literature, 39(1), 34–92. Solinger, D. (2001), “Why We Cannot Count the ‘Unemployed,’” China Quarterly, no. 167, 671–688. State Statistical Bureau (SSB) (2006), Zhongguo laodong tongji nianjian 2006 (China Labor Statistical Yearbook 2006), Beijing: Zhongguo tongji chubanshe. Tsang, M.C. and Y. Ding (2005), “Resource Utilization and Disparities in Compulsory Education in China,” China Review, 5(1), 1–31. Wong, L. and K. Ngok (2006), “Social Policy between Plan and Market: Xiagang (Offduty Employment) and the Policy of Re-employment Service Centres in China,” Social Policy and Administration, 40(2), 158–173. Zhang, J., Y. Zhao, A. Park, and X. Song (2005), “Economic Returns to Schooling in Urban China 1988 to 2001,” Journal of Comparative Economics, 33(4), 730–752. Zhang, Y., E. Hannum, and M. Wang (2008), “Gender-Based Employment and Income Differences in Urban China: Considering the Contributions of Marriage and Parenthood,” Social Forces, 86(4), 1529–1560. Zhao, S. and D. Wen (2008), “Sanshi nianlai gaoxiao biyesheng jiuye zhidu biange de huigu yu xianxing zhidu de fenxi” (Review of Changes in the College Graduates’ Employment System in the Past Thirty Years and Analysis of the Current System), Zhongguo gaojiao yanjiu, no. 8, 2–5.

NINE

Do Employees in the Public Sector Still Enjoy Earnings Advantages? Yang Juan, Sylvie D´emurger, and Li Shi

I. Introduction Three decades of economic reform have brought tremendous changes in every sector of the Chinese economy. The labor market is no exception, and it was particularly affected by important policy and institutional changes at the turn of the century. On the one hand, the state-sector reform was accelerated after the Chinese Communist Party’s Fifteenth National Congress (September 1997), which encouraged both the corporatization of large state-owned enterprises (SOEs) and the restructuring of small SOEs. On the other hand, the Congress also recognized private enterprises as an important component of the economy and placed an emphasis on rule of law. As a direct consequence, the urban labor market was reshaped due to the unprecedented growth in unemployment and the reallocation of labor from the public to the private sector. At the same time, competition among workers in the urban labor market increased sharply due to the massive rural laborforce exodus, which led to an estimated 140 million rural workers in the cities by 2008. In the context of a transitional economy, these dramatic changes raise a number of issues about the direction of the urban labor market. A key aspect to be explored is whether the labor market has become market oriented and whether enterprises with different ownerships operate competitively. Academic research using data collected from the mid-1990s to the early twenty-first century highlights the incompleteness of the reforms and the “unfinished economic revolution” (Lardy 1998), as well as the remaining rigidities in a segmented labor market with distinct rules for wage determination and limited labor mobility between segments (e.g., Chen, D´emurger, and Fournier 2005; D´emurger et al. 2006; Dong and Bowles 2002; Knight and Song 2003; Wang 2005). Evidence from mid-1990s micro data shows 332

Do Employees in the Public Sector Still Enjoy Earnings Advantages?

333

that workers in the public sector had very few incentives to move and one of the main reasons for this immobility was the higher-than-market-clearinglevel earnings premium provided to workers in state-owned units (Chen et al. 2005; Zhao 2002). Moreover, for the period from 1995 to 2002, D´emurger et al. (2006) find strong evidence of increasing segmentation across ownership, with the gap between the privileged segments of the labor market and the most competitive segments widening over time. Policy-related rationales for studying labor-market segmentation issues are related to both efficiency, as illustrated in the literature on the publicprivate sector earnings gap in developing countries or economies in transition (Adamchik and Bedi 2000; Boeri and Terrell 2002; Falaris 2004; Lokshin and Jovanovic 2003) and income inequalities (Meng and Zhang 2001). A multitiered labor market in which wages are determined not only by skill differentials but also by different institutional arrangements may have strong implications in terms of both labor allocation across sectors and income distribution among workers. In China, where the so-called iron rice bowl (tiefanwan) of lifetime employment and the associated welfare state dominated for years before it was completely dismantled in 1994 (Knight and Song 2005), the issue of public-sector efficiency appears to have special importance. Moreover, the question of income distribution is essential to any government concerned about smooth economic development and social safety. With the growth of the Chinese economy and rising average wages, the earnings gap triggered vigorous debate. In this context, ownership is also a fairly important issue because it is linked to whether the government can provide an equal and efficient business environment for all sorts of companies to develop and maximize social welfare. Given that the number of enterprises in the public sector decreased from about 99 percent of all companies in 1978 to merely 10 percent in 2007, it is also interesting to explore whether the remaining public-sector enterprises still enjoy a privileged position in the labor market due to specific government policies. Macroeconomic data on the average wages of staff and workers in urban China show an increasing trend since the mid-1990s (Figure 9.1). The average wage in 2007 was 2,060 yuan per month, 5.7 times higher than the average wage in 1995 (in constant 2007 prices). Although the increase was rapid for every type of ownership, some discrepancies emerged over time, the most remarkable being a narrowing gap between the public and private sectors. Indeed, whereas average wages in 1995 were the highest in the “other ownership” (private sector) category, they were lower than those in the SOEs in 2007. Similarly, Peking University’s College Students

334

Yang Juan, Sylvie D´emurger, and Li Shi

35,000 30,000

Total SOEs

25,000

UCEs

Yuan

Other 20,000 15,000 10,000 5,000 0 1995

1996

1997

1998

1999

2000

2001

2002

2003

2004

2005

2006

2007

Figure 9.1. Average Annual Real Wage Trend for Public and Private Sectors, 1995–2007. Source: National Bureau of Statistics (2008). Note: In the national statistics, wages refer to the “total remuneration for labor paid by all organizations directly to all staff and workers of those entities.” The reported classifications by ownership do not distinguish foreign-invested enterprises and private enterprises, which are both included in the category “other.” Average annual wages of staff and workers are deflated by the urban consumer price index (1995 = 100).

Employment Survey of more than 100 universities in 2002 and 2007 shows that whereas the first employment intention of students in 2002 was to work in foreign-invested enterprises, in 2007 it was in the SOE sector. This change calls for further research to investigate whether this was due to any discriminatory behavior or to specific powers of certain types of enterprises. The 2008 release of 2007 data from the China Household Income Project (CHIP) project makes it possible to analyze whether China’s labor market is still segmented by ownership in terms of earnings differentials. The comprehensive information on personal characteristics provided by the available micro data sets enables us to investigate wage compensation by controlling for the individuals’ most important characteristics. Previous research on China’s labor-market segmentation utilized data for 2002 or earlier. However, during the 2002–2007 period, China’s economic growth averaged 10.8 percent in real terms and China became increasingly integrated into global markets, especially after joining the World Trade Organization (WTO) in 2001. In addition, private companies that were allowed to enter the previously state-controlled areas, such as steel, aluminum, and automobiles, have been immensely successful and many have gone public

Do Employees in the Public Sector Still Enjoy Earnings Advantages?

335

and become among the top-500 companies in China. This implies that old-style companies like state-owned enterprises and urban collective companies have to compete much more intensively with market-oriented companies, including private companies, joint-venture companies, and foreign companies. Against the backdrop of the accelerating economic reforms between 2002 and 2007, we propose to investigate the trends and determinants of the earnings gap across ownership types during this period. We first analyze the average gaps by using the Oaxaca-Blinder decomposition technique. We then account for the different patterns in the various percentiles for different ownership types by applying the Juhn-Murphy-Pierce decomposition method. The remainder of this chapter is organized as follows: Section II describes the development of various types of enterprises in China. Section III introduces the data set and some descriptive statistics. Section IV discusses the econometric results of the earnings equations by enterprise ownership. Section V and Section VI describe the decomposition results of the earnings gaps across ownership types during the 2002–2007 period. Section VII presents our conclusions.

II. Economic Reforms and the Evolution of Ownership After the People’s Republic of China was founded in 1949, the means of production were gradually transferred to the state, and by 1956 private and individual economic activities had become illegal (Naughton 2007). In urban China, within the period of the First Five-Year Plan, the share of public ownership increased from 21.3 percent in 1952 to 92.9 percent in 1956 in urban China (Su 1999). When the economic reforms began in 1978, the national economy was strongly dominated by public ownership, which consisted of state-owned and collective enterprises. State-owned and collective enterprises (including township and village enterprises [TVEs]) accounted for 24 percent and 76 percent, respectively, of the total number of industrial companies (Su 1999), and produced 77 percent and 23 percent of the total industrial output (Naughton 2007). One major feature of the economic reforms was to encourage the development of the nonstate sector of the economy while reforming the organization of state-owned enterprises. By introducing a series of laws and regulations, the government gradually allowed private and foreign companies to coexist with state-owned and collective companies. In 1988, the State Council issued the “Tentative Stipulations on Private Enterprises” to govern the

336

Yang Juan, Sylvie D´emurger, and Li Shi

registration and management of private firms, and in 1993, the Company Law was promulgated to provide a legal framework for the development of limited liability companies and shareholding companies (D´emurger et al. 2006). Hence, various forms of nonpublic ownership, such as privately owned, foreign-invested, joint-venture, shareholding, stock, and selfemployed companies, became alternatives to the state-owned companies. More recently, efforts have been made to ensure fairer competition between the public and the private sectors and to open more industries to the private sector. In 2003, new regulations allowed nonstate enterprises to enter the steel, aluminum, and even some parts of the national defense industries. In February 2005, the State Council issued its “Thirty-six Suggestions to Encourage and Support Nonstate-Owned Economic Development” in order to reduce the barriers to market entry and to stimulate private investment. In addition to helping promote competition among companies, the development of the nonstate sector helped allocate resources more efficiently. Before the reforms, because resources were allocated according to the plan and the economy was dominated by public ownership, there was no competition among enterprises or employees. Allowing private and foreign companies to enter the labor market led to improvements in the national economy as a whole and consequently to the promotion of prosperity. The other advantage of allowing the existence of private and foreign companies was the alleviation of employment pressures. With the baby boom and soldiers transferred to nonmilitary sectors, the labor force grew by more than 10 million per year, and the nonstate sector became a major channel to absorb the new labor force. Hence, whereas employment in the public sector rose continuously until the mid-1990s, it began to decrease in 1995, with a huge deceleration in 1998 (–18 percent), the pivotal year in the SOEs reforms. Since then, the number of workers in SOEs and urban collective enterprises (UCEs) decreased from 144.6 million in 1995 to 71.4 million in 2007 (National Bureau of Statistics [NBS] 2008), a total decrease of 50 percent. As a consequence, the public-sector share of urban employment dropped from 76 percent to 24 percent during the period (Naughton 2007). In contrast to the downsizing of the public sector, the private-sector share of urban employment increased from 16 percent in 1995 to 42 percent in 2007. The remaining 34 percent in 2007 was made of “other” employment, which “picks up most of the migrants and unregistered businesses” (Naughton 2007: 190). The dramatic increase in the private-sector share of urban employment can be attributed to the development of both private or individual enterprises and

Do Employees in the Public Sector Still Enjoy Earnings Advantages?

337

foreign-invested enterprises beginning in the mid-1990s, as well as to the emergence of new forms of ownership, including limited liability corporations and shareholding corporations. From 1995 to 2007, employment in foreign-invested enterprises tripled (from 5.1 million to 25.8 million) and employment in private and individual enterprises almost quadrupled (from 20.4 million to 78.9 million). Moreover, the number of people employed in the new ownership forms increased tenfold, from 3.2 million in 1995 to 30.8 million in 2007. These figures clearly indicate a significant shift in the employment structure by the turn of the century as China experienced a situation somewhat similar to that in the Eastern European countries when the labor force moved from the public to the private sector.

III. Data and Descriptive Analysis A. Summary Statistics by Ownership The data used in this chapter come from two sources: the CHIP carried out in 2003 for 2002 and the CHIP carried out in 2008 for 2007. For both surveys, the questionnaire was designed by Chinese and foreign researchers and implemented by China’s National Bureau of Statistics (NBS).1 The two data sets include three separate surveys: urban, rural, and migrant. In this analysis, we employ the urban survey that covers only urban residents.2 The 2002 CHIP urban survey covers a sample of 20,632 individuals residing in 6,835 households in twelve provinces, and the 2007 CHIP urban survey covers a sample of 14,699 individuals residing in 5,003 households in nine provinces.3 For the sake of comparison, we keep in our sample only observations of the jointly surveyed seven provinces. The seven provinces are Jiangsu, Anhui, Henan, Hubei, Guangdong, Chongqing, and Sichuan. In view of 1

2

3

Although the sampling design for both surveys was based on that of the annual urban household survey conducted by the NBS, there is one discrepancy between the two data sets that should be noted. Indeed, the 2007 CHIP data were collected from a new NBS sample. The sampled households, which joined the survey in 2008, unlike the households in the 2002 survey that had recorded income, reported their income by recalling. According to the NBS, recalled income might be less accurate than recorded income. Unfortunately, it is not possible to provide robustness checks for this, but we believe that given the scope of the identified effects in our analysis, the bias, if any, should not be too strong. Urban residents are people who live in cities and who hold an urban household registration (hukou). Unregistered urban workers such as rural migrants are not included in this data set. For 2007 we use the additional CHIP 2007 urban sample of 5,000 households that were interviewed only using the CHIP questionnaire, as discussed in Appendix I.

338

Yang Juan, Sylvie D´emurger, and Li Shi

the fact that our analysis is restricted to households in the urban surveys and in the same sample provinces in the two years, we do not use weights in our analysis. In addition, we further restrict the sample to individuals between the ages of sixteen and sixty for men and between the ages of sixteen and fifty-five for women who were earning positive wages with full-time employment.4 The final sample size totaled 5,430 workers in 2002 and 5,029 workers in 2007. Enterprise ownership analyzed in the chapter is divided into five categories (see Table 9.1): state-owned enterprises (SOEs), government agencies or institutions (GAIs), urban collective enterprises (UCEs), private or individual enterprises (PIEs), and foreign-invested enterprises (FIEs). A comparison between 2002 and 2007 shows opposite trends in the public and the private sectors: the share of SOEs decreased from 35 percent to 19 percent, whereas the share of PIEs increased from 24 percent to 33 percent (Table 9.2). This raises the issue of how to classify enterprises according to ownership. In each survey, respondents were required to provide the ownership of their company. In the 2002 CHIP survey, the ownership was divided into thirteen types, and in 2007 it was divided into sixteen types. In order to simplify the analysis, we reduce these different types to five categories. The SOE category thus contains state-owned enterprises, state-controlled enterprises, and state-owned joint ventures. In other words, as long as the state share is dominant, no matter who owns the other shares (whether foreigners or private Chinese investors), in our analysis the enterprise will still fall into the SOEs category. However, we classify the solely foreign-invested companies and foreign-owned joint-venture companies as FIEs. This classification choice may explain why, despite the substantial increase in foreign direct investment from 2002 to 2007, the share of FIEs in our analysis does not change significantly. Descriptive statistics on the individual characteristics of different ownerships are shown in Table 9.2. The gender distribution does not change much across years, with males representing 56 to 57 percent of the urban workers and concentrated particularly in SOEs where they account for 59.5 percent and 61.5 percent of the total in 2002 and 2007, respectively. To some 4

After restricting the sample to full-time employment, the minimum age of the sample increased to eighteen. One may argue that with the expansion of higher education, most individuals between the ages of eighteen and twenty-two are still in school, therefore possibly resulting in a bias in the sample selection. However, in the 2007 CHIP, the percentage of individuals between the ages of eighteen and twenty-two who were still in school only accounted for 3.6 percent of this age group, and the percentage was even lower in 2002. Hence, such a bias, if any, should not seriously affect our estimation results.

Do Employees in the Public Sector Still Enjoy Earnings Advantages?

339

Table 9.1. Definition of ownership categories Public versus private sector Public sector

Ownership categories

Types included

State-owned enterprises (SOEs)

Solely state-owned enterprises; State holding enterprises; State holding joint ventures. Government agencies and party agencies (including the party committee, Government, People’s Congress, the Chinese People’s Political Consultative Conference (CPPCC), public security organs & procurator’s offices & courts, the military); State and collective institutions; Civilian-run enterprises and public service units. Solely collective owned enterprises; Collective holding enterprises; Collective holding joint ventures. Solely private-owned enterprises; Private holding enterprises; Private holding joint ventures; Self-employed individuals. Solely foreign-owned enterprises; Foreign holding joint ventures.

Government agencies and institutions (GAIs)

Semipublic sector

Urban collective enterprises (UCEs)

Private sector

Private and individual enterprises (PIEs)

Foreign-invested enterprises (FIEs)

Note: If the answer given by the respondent is “Other enterprise,” then it is not attributed to any of the above categories and is simply dropped. Source: 2007 CHIP urban survey questionnaire.

extent, this distribution suggests that males may enjoy some recruitment and income from the public sector. Yet a noteworthy change between 2002 and 2007 occurred in the UCEs where females were traditionally overrepresented (D´emurger, Fournier, and Chen 2007; Maurer-Fazio, Rawski, and Zhang 1999). In 2007, males accounted for 53.7 percent of workers in UCEs, against only 44 percent in 2002. Although it still was the lowest share of males across ownership, the difference was not significantly different from the other categories (with the exception of the SOEs). As documented later, this change reflects the improving situation of the UCEs, where increased competition may have boosted productivity and attracted more talented workers. A comparison between 2002 and 2007 shows a slight decrease in the average age of the workforce, but it was more marked in the UCEs and

Table 9.2. Descriptive statistics on individual characteristics by ownership 2002

SOEs

GAIs

UCEs

PIEs

FIEs

All

Male

Observations % of total

0.595 (0.491) 40.86 (8.505) 11.17 (2.702) 17.71 (9.383) 0.283 (0.451) 0.308 (0.462) 0.348 (0.476) 2.912 (1.129) 1,896 34.92

0.551 (0.498) 40.46 (8.914) 12.69 (2.872) 14.43 (9.214) 0.327 (0.469) 0.294 (0.456) 0.305 (0.461) 1.793 (1.013) 1,698 31.27

0.440 (0.497) 41.44 (8.178) 10.13 (2.471) 16.25 (9.380) 0.239 (0.427) 0.450 (0.498) 0.232 (0.422) 1.919 (0.968) 393 7.24

0.549 (0.498) 39.11 (8.666) 10.14 (2.853) 10.45 (9.335) 0.198 (0.398) 0.363 (0.481) 0.236 (0.425) 1.752 (1.069) 1,316 24.24

0.567 (0.497) 35.85 (8.923) 11.96 (2.665) 10.46 (8.417) 0.276 (0.449) 0.504 (0.502) 0.488 (0.502) 2.709 (1.062) 127 2.34

0.558 (0.497) 40.24 (8.716) 11.34 (2.957) 14.65 (9.722) 0.273 (0.445) 0.332 (0.471) 0.302 (0.459) 2.204 (1.194) 5,430 100.00

2007

SOEs

GAIs

UCEs

PIEs

FIEs

All

Male

0.615 (0.487) 40.56 (9.258) 12.14 (3.032) 16.99 (10.72) 0.442 (0.497) 0.248 (0.432) 0.673 (0.469) 2.531 (1.208) 949 18.87

0.570 (0.495) 40.59 (9.332) 12.99 (3.076) 14.85 (10.63) 0.425 (0.494) 0.321 (0.467) 0.664 (0.473) 1.858 (1.082) 1,968 39.13

0.537 (0.500) 39.52 (9.118) 11.78 (3.116) 12.48 (10.19) 0.326 (0.470) 0.389 (0.488) 0.646 (0.479) 1.800 (1.013) 285 5.67

0.555 (0.497) 37.93 (9.236) 11.30 (3.232) 8.606 (8.115) 0.275 (0.447) 0.398 (0.490) 0.606 (0.489) 1.358 (0.783) 1,655 32.91

0.558 (0.498) 34.17 (7.748) 13.39 (3.211) 8.628 (7.126) 0.407 (0.493) 0.628 (0.485) 0.512 (0.501) 2.145 (1.227) 172 3.42

0.571 (0.495) 39.43 (9.353) 12.22 (3.215) 12.85 (10.32) 0.372 (0.484) 0.347 (0.476) 0.640 (0.480) 1.827 (1.102) 5,029 100.00

Age Education Experience in current job Training Coast Capital city Company size

Age Education Experience in current job current job Training Coast Capital city Company size Observations % of total

Note: Ownership categories are state-owned enterprises (SOEs), government agencies or institutions (GAIs), urban collective enterprises (UCEs), private or individual enterprises (PIEs), and foreign-invested enterprises (FIEs). “Male,” “Training,” “Coast,” and “Capital city” are dummy variables for being a male, having received training, living in a coastal city, and living in a provincial capital city respectively. “Education” measures the number of years of education received. “Experience in current job” refers to the number of years in the current occupation. “Company size” measures the number of employees in the company and is grouped by 4 ranks (following the 2002 CHIP): 1 represents 1–100 employees, 2 represents 101–500 employees, 3 represents 501–1,000 employees, and 4 represents 1,000 employees or more. Source: Authors’ calculations using the 2002 CHIP and the 2007 CHIP survey data, urban sample, 7 provinces, with 16 ≤ age ≤ 60 for men and 16 ≤ age ≤ 55 for women, full-time employment, and earning positive wages.

Do Employees in the Public Sector Still Enjoy Earnings Advantages?

341

in the private sector than in the public sector (SOEs and GAIs). In both years, the public sector employed more older workers than did the private sector. As expected, with the expansion of higher education after 1999, the average educational attainment of the workforce, measured in years of schooling, substantially increased over time, by almost one year during our five-year period (from 11.34 years to 12.22 years). Except for the GAIs, which employed the most-educated workers in 2002,5 each ownership category benefited from the increase in the education level so that the absolute gap in the educational attainment of workers across ownership declined from 2.56 years to 2.09 years. This evolution indicates that in addition to public administration, FIEs were increasingly able to attract talented youth in 2007. The average experience in the current job (expressed in years) was much shorter in 2007 than in 2002 for all the sectors except the GAIs. The sharpest decreases occurred in the semipublic sector (UCEs) and the private sector (both PIEs and FIEs). This evolution probably signals increased job mobility in these sectors, whereas jobs in the public sector (SOEs and GAIs) were still the most stable, and hence individuals did not readily leave their positions there. Finally, the average size of companies experienced a decreasing trend between 2002 and 2007, with the SOEs and FIEs among the largest enterprises.

B. The Evolution of Earnings and their Distribution by Ownership Table 9.3 reports the summary statistics on earnings by ownership. Total annual earnings are composed of reported wages, bonuses, in-kind earnings, subsidies, pension income, and so forth. Hourly earnings are calculated by dividing the total annual earnings by the number of declared hours worked in a year. In addition, earnings are adjusted for provincial purchasing power differences by using an updated set of the Brandt and Holz (2006) urban provincial-level spatial price deflators in order to account for differences in living standards across cities. In the five-year period from 2002 to 2007, earnings differentials between enterprises of different ownership changed markedly. On average, real earnings almost doubled, but at a different pace across enterprises. The state sector experienced the slowest growth in annual and hourly earnings (88–92 percent for the SOEs and 62–63 percent for the GAIs); in contrast, both the UCEs and the private sector experienced earnings increases of more than 5

In 2002, the average education level of workers in GAIs was more than twelve years, much higher than that in any other type of enterprise.

342

Yang Juan, Sylvie D´emurger, and Li Shi Table 9.3. Descriptive statistics on individual earnings by ownership

2002

SOEs

Total year income

11,261.6 (7,352.8) Gap to average earnings 0.98 Gini coefficient 0.307 Working hours/week 42.30 (7.972) Hourly wage 5.380 (4.375) Gap to average earnings 0.99 Gini coefficient 0.334 Observations 1,896 2007

SOEs

Total year income Gap to average earnings Gini coefficient 2002–2007 growth rate Working hours/week Hourly wage Gap to average earnings Gini coefficient 2002–2007 growth rate Observations

GAIs

UCEs

PIEs

FIEs

All

14,221.1 (7,992.0) 1.24 0.290 41.23 (8.060) 7.086 (6.096) 1.30 0.328 1,698

8,108.8 (4,880.5) 0.70 0.293 44.38 (10.39) 3.710 (2.444) 0.68 0.322 393

9,270.9 (9,157.6) 0.81 0.386 51.94 (15.63) 3.851 (4.819) 0.71 0.430 1,316

12,907.7 (9,617.8) 1.12 0.324 45.34 (11.61) 5.877 (4.836) 1.08 0.361 127

11,514.9 (8,211.9)

GAIs

UCEs

PIEs

FIEs

All

0.336 44.52 (11.45) 5.434 (5.155) 0.377 5,430

21,614.6 23,096.0 18,897.0 20,492.2 27,455.7 21,870.7 (18,204.8) (16,235.1) (12,956.8) (27,264.2) (19,755.7) (20,872.9) 0.99 1.06 0.86 0.94 1.26 0.341 0.338 0.337 0.408 0.366 0.367 92% 62% 133% 121% 113% 90% 43.24 42.19 44.65 49.87 42.61 45.07 (9.682) (19.21) (10.24) (22.32) (7.716) (18.59) 10.13 11.58 8.826 8.947 12.81 10.33 (9.031) (11.01) (6.914) (11.56) (9.277) (10.68) 0.98 1.12 0.85 0.87 1.24 0.364 0.378 0.375 0.449 0.375 0.405 88% 63% 138% 132% 118% 90% 949 1,968 285 1,655 172 5,029

Notes: Earnings are deflated using the urban provincial-level spatial price deflators calculated by Brandt and Holz (2006), and updated for 2007. Base: Nationwide prices in 2002. Source: See Table 9.2.

110 percent (up to a maximum of 138 percent for hourly earnings in the UCEs). As opposed to what occurred between 1995 and 2002 (D´emurger et al. 2006), the differentials across enterprises somehow readjusted in the direction of more equality due to the dramatic increase in earnings in both the UCEs and the PIEs. On the one hand, although total earnings were the highest in the GAIs in 2002, the much-slower increase in earnings in the GAIs between 2002 and 2007 moved them down to the second rank, below the FIEs.6 On the other hand, the UCEs as well as the PIEs saw their relative 6

Interestingly, this is a complete reversal compared to the 1995–2002 period (see D´emurger et al. 2006).

Do Employees in the Public Sector Still Enjoy Earnings Advantages?

343

positions improve dramatically (again unlike what occurred between 1995 and 2002), and the gap to average total earnings was reduced from 0.7 to 0.86 for the UCEs and from 0.81 to 0.94 for the PIEs. Last, SOEs stood at the middle and the almost doubling of earnings in this part of the state sector allowed workers to maintain their intermediate position, with a gap to average earnings very close to one during the two years.7 Another interesting point focuses on the ongoing convergence of working time between the public and private sectors. From 1995 to 2002, the number of hours worked per week continued to decrease in both PIEs and FIEs. However, in 2007 the working time increased slightly in the public sector, although it remained less than that in the private sector. One possible reason for this convergence is that the competition in the SOEs and GAIs sectors became more intensive and employees had to work harder to maintain their positions. Furthermore, the PIEs and FIEs began to pay more attention to employee rights. The Gini coefficients highlight a general trend of increasing inequality in annual and hourly earnings. For the entire sample, the Gini coefficient for hourly earnings increased from 0.377 to 0.405 between 2002 and 2007. Although PIEs continuously exhibited the largest earnings dispersion over time,8 the increase in earnings inequality was more pronounced in the public sector (including the UCEs), which resulted in a convergence of the earnings distributions across sectors between 2002 and 2007. Nonparametric kernel density estimations for the distribution of the logarithm of hourly earnings by ownership category and by year are presented in Figure 9.2.9 For each year, the figure shows the distribution for the entire sample as well as for the ownership category subsamples. The upper panel of Figure 9.2 displays the kernel density estimates for the year 2002. Hourly earnings in GAIs on average were higher than those in other sectors, which can be seen in both the position of the curve most 7

8

9

One should note, however, that reported earnings may not fully reflect individuals’ actual income in the state sector and may result in an underestimation of earnings. Indeed, the welfare system in the SOEs and GAIs is still much better than that in the FIEs and PIEs, but it is difficult to collect complete information on this, especially with respect to nonpecuniary welfare. Given the comparatively high wages in these two sectors, plus the nonobservable income, jobs in SOEs and GAIs can still be as attractive as, or even more attractive than, jobs in FIEs. This trend confirms the more unequal distribution of hourly wages in the private sector as compared to the public sector that was observed in the 1990s (see Chen et al. 2005; Xing 2008). Nonparametric kernel density estimation is a way to estimate the probability density function of a random variable. In our case, the variable of interest is the logarithm of hourly earnings.

344

Yang Juan, Sylvie D´emurger, and Li Shi Kernel Density of Hourly Earnings by Ownership, 2002

0,8

0,6

0,4

0,2

0 -2

0

Total

SOEs

2

GAIs

4

UCEs

6

PIEs

FIEs

Kernel Density of Hourly Earnings by Ownership, 2007 0,8

0,6

0,4

0,2

0 -2

0

Total

SOEs

2

GAIs

4

UCEs

6

PIEs

FIEs

Figure 9.2. Kernel Density Estimations for the Distribution of Income by Ownership Category, 2002 and 2007. Source: Authors’ calculations using the 2002 and 2007 CHIP survey data. Note: See Table 9.2. Earnings are deflated using the urban provincial-level spatial price deflators calculated by Brandt and Holz (2006), and updated for 2007. Base: Nationwide prices in 2002.

Do Employees in the Public Sector Still Enjoy Earnings Advantages?

345

to the right and the higher kurtosis. In addition, the spread is very narrow and highly concentrated around the mean. FIEs came second, with average earnings only slightly higher than those in SOEs but with a larger width, thereby illustrating a wider distribution. The hourly wages in PIEs were the lowest among the five sectors, with the distribution skewed to the right, indicating that some earnings in PIEs were fairly low. As illustrated in the bottom panel of Figure 9.2, the patterns did not change much over time, except that the five lines seem closer in 2007 than they were in 2002. This further illustrates the converging trend of the hourly earnings gap among the five sectors. In 2007 the FIEs exhibit better hourly wage distributions than do the GAIs. Together with the higher average hourly earnings, FIEs also exhibit a very flat tail in the left part of the distribution, indicating that there were not many low-wage earners in this sector.10 Moreover, the distributions for the GAIs and UCEs are quite similar, except that the hourly wage distribution of the GAIs is on the right of the hourly wage distribution of the UCEs. Finally, the kurtosis is the highest in the SOEs, suggesting a sharper peak and fatter tails for the hourly earnings distribution in the state sector.

IV. The Determinants of Hourly Earnings Tables 9.4 and 9.5 present ordinary least squares (OLS) estimations of an augmented Mincerian hourly earnings function (Mincer 1974) run separately by enterprise ownership and by year.11 The Mincerian earnings equation takes the following form: w ir = βir X ir + uir ,

(1)

where subscript r ∈ [1, 5] represents the five different ownership categories defined earlier. wir is the natural logarithm of hourly earnings (adjusted for provincial purchasing power differences) for individual i in enterprise r. Xir is a vector of his or her individual characteristics, and β gives the set of returns to each observed socio-demographic characteristic. X includes 10

11

This may reflect our ownership classifications. Indeed, FIEs include only foreign-owned and foreign-controlled enterprises, which are mainly concentrated in the higher-end industries. Card (1999) provides a brief introduction to all kinds of estimation methods and their respective advantages and disadvantages in terms of returns to education. He suggests that the OLS estimation method is the most robust technique.

346

Yang Juan, Sylvie D´emurger, and Li Shi Table 9.4. Hourly wage functions by ownership, 2002 (1) SOEs

Male Education Experience Experience2 Experience in current job Training Coast Capital city Company size Constant N R2

0.104*** (0.000) 0.0621*** (0.000) 0.0370*** (0.000) −0.000591*** (0.000) 0.00316 (0.147) 0.200*** (0.000) 0.215*** (0.000) 0.200*** (0.000) 0.0354*** (0.003) −0.108 (0.315) 1,896 0.194

(2) GAIs

(3) UCEs

(4) PIEs

(5) FIEs

0.0700** 0.122** 0.219*** 0.145 (0.012) (0.041) (0.000) (0.201) 0.0657*** 0.0564*** 0.0759*** 0.0857*** (0.000) (0.000) (0.000) (0.000) 0.0422*** 0.0226 0.0380*** 0.0207 (0.000) (0.116) (0.000) (0.365) −0.000705*** −0.000391 −0.000631*** −0.000223 (0.000) (0.158) (0.001) (0.679) 0.0128*** 0.00107 0.0127*** 0.00898 (0.000) (0.761) (0.000) (0.204) 0.0390 0.0914 0.0759 0.195 (0.191) (0.164) (0.252) (0.179) 0.338*** 0.161*** 0.423*** 0.159 (0.000) (0.007) (0.000) (0.225) 0.0362 0.314*** 0.264*** 0.0702 (0.225) (0.000) (0.000) (0.600) 0.0430*** 0.0287 0.128*** 0.116** (0.001) (0.262) (0.000) (0.021) −0.00987 −0.0149 −0.982*** −0.413 (0.929) (0.949) (0.000) (0.263) 1,698 393 1,316 127 0.256 0.156 0.205 0.219

Notes: See Table 9.2. p-values in parentheses. Earnings are deflated using the urban provincial-level spatial price deflators calculated by Brandt and Holz (2006), and updated for 2007. Base: Nationwide prices in 2002. * p < 0.10, ** p < 0.05, and ***p < 0.01. Source: See Table 9.2.

gender, education (measured in years of schooling, as reported in the surveys), work experience12 and its square, work experience in the current occupation, on-the-job training (dummy variable), regional dummies for coastal provinces and for capital cities, and company size. The residual uir stands for all the unobservable factors that may affect individual hourly earnings w. Returns to education are significant in all sectors for both years. They are much higher in GAIs and FIEs than in any other sector. A comparison over time reveals interesting changes. Indeed, returns to education exhibit 12

The actual work experience is not reported in the 2007 survey. As a consequence, we use the potential work experience, defined as age minus number of years in school minus six.

Do Employees in the Public Sector Still Enjoy Earnings Advantages?

347

Table 9.5. Hourly wage functions by ownership, 2007 (1) SOEs Male Education Experience Experience2 Experience in current job Training Coast Capital city Company size Constant N R2

(2) GAIs

(3) UCEs

0.195*** 0.168*** 0.212*** (0.000) (0.000) (0.005) 0.0381*** 0.0717*** 0.0653*** (0.000) (0.000) (0.000) 0.0168** 0.0129** −0.00463 (0.047) (0.049) (0.748) −0.000452*** −0.000352** −0.00000205 (0.009) (0.014) (0.995) 0.00788*** 0.0214*** 0.00599 (0.003) (0.000) (0.187) 0.197*** 0.0478 0.118 (0.000) (0.119) (0.152) 0.341*** 0.337*** 0.271*** (0.000) (0.000) (0.000) 0.0770* 0.0629** 0.183** (0.079) (0.036) (0.017) 0.00609 0.0359** 0.0294 (0.743) (0.012) (0.415) 1.028*** 0.540*** 0.776** (0.000) (0.000) (0.011) 948 1,964 285 0.136 0.243 0.212

(4) PIEs

(5) FIEs

0.250*** 0.0771 (0.000) (0.407) 0.0500*** 0.0904*** (0.000) (0.000) 0.0137** 0.0174 (0.031) (0.474) −0.000458*** −0.000617 (0.002) (0.318) 0.0126*** 0.0336*** (0.000) (0.000) 0.100*** 0.102 (0.006) (0.275) 0.384*** 0.213* (0.000) (0.063) 0.184*** −0.129 (0.000) (0.206) 0.0433** −0.0482 (0.030) (0.201) 0.650*** 0.689 (0.000) (0.125) 1,652 172 0.198 0.336

Note: See Table 9.4. Source: See Table 9.2.

an increasing trend in UCEs, FIEs, and GAIs between 2002 and 2007, but a decreasing trend in both SOEs and PIEs, resulting in a growing gap across sectors. Hence, the range of returns to education depending on enterprise ownership moved from 5.64 percent to 8.57 percent in 2002 to 3.81 percent to 9.04 percent in 2007. Linear and quadratic terms in experience are significant in the public sector (SOEs and GAIs) as well as in the private sector (PIEs), but not significant in UCEs and FIEs in either 2002 or 2007. As discussed in Chen et al. (2005), the observed difference in experience earnings profiles between the public sector and the other sectors suggests that in SOEs and GAIs seniority remains an important component in the determination of wages. Interestingly, however, experience is also important in the private sector. A comparison between 2002 and 2007 shows much

348

Yang Juan, Sylvie D´emurger, and Li Shi

earlier earnings peaks in 2007, suggesting that older people saw their relative position deteriorating over time. Indeed, whatever the enterprise, in 2007 the experience profile began to decrease after twenty years of experience, whereas in 2002 it decreased after thirty years of experience. The introduction of another experience indicator that measures the number of years in the company adds some interesting results for the foreigninvested firms. Indeed, the associated coefficient turns out to be significant in 2007, suggesting that the experience that counts for FIEs is experience accumulated in the enterprise rather than overall experience (which may have also been accumulated in the less efficient public or semi-public sectors). Returns to gender also exhibit noteworthy differences across ownership and over time. In 2002, being a male in a PIE increased log hourly wages by about 21.9 percent, whereas the increase was only 7 percent in GAIs. The “male premium” increased dramatically over time, especially in the public and the semipublic sectors, and reached levels between 16.5 percent (in GAIs) and 25 percent (in PIEs). This partly reflects findings by Li and Song (2010) and in Chapter 11 of this volume that show that gender-wage inequality increased during the 2002–2007 period. Interestingly, FIEs do not appear to favor males over females since the coefficient for the gender dummy variable is never significant. Finally, the coefficient estimates for being located in a coastal province (Beijing, Jiangsu, or Guangdong) show a premium for living along the coast in all enterprises except the FIEs in both years. A comparison of the coefficients over time suggests an increasing “coastal premium” for the SOEs and UCEs, and a slightly decreasing premium for the PIEs.

V. Oaxaca-Blinder Decompositions In order to analyze the earnings differentials between individuals belonging to different enterprises, we first use the Oaxaca-Blinder method (Blinder 1973; Oaxaca 1973) to decompose the mean differences in log earnings into two components: one attributable to the differences in the mean endowments of workers across ownership, and one attributable to the differences in returns to these endowments. The observed difference in average log earnings between two enterprises of different ownership, r1 and r2 , can be defined as w r1 r2 = w r1 − w r2 ,

(2)

Do Employees in the Public Sector Still Enjoy Earnings Advantages?

349

where the bars indicate the mean values. Substituting Equation (1) for Equation (2) yields w r1 r 2 = X r1  βˆr1 − X r2  βˆr2 ,

(3)

where the hats denote the estimated coefficients from the separate earnings equations. Assuming that a nondiscriminatory wage structure β∗ is known, the log wage differential can be decomposed in the following way (Neumark 1988):   w r1 r2 = (X r1 − X r2 ) β∗ + [X r1 (βˆ r1 − β∗ ) − X r2 (βˆ r2 − β∗ )].

(4)

Equation (4) shows that the earnings gap between ownership r1 and ownership r2 can be decomposed into two parts. The first term can be interpreted as the part of the log earnings differential due to differences in average individual characteristics between different ownerships. This measures how much individuals in ownership r1 would earn if they had the same characteristics as those in ownership r2 . The second term represents the amount by which earnings in two different ownerships differ from the assumed nondiscriminatory wage structure. It is the “unexplained” or residual component of the earnings gap. This effect may be due to either segmentation or different productivity levels. In other words, the fact that individuals with the same characteristics are paid differently in firms of different ownership might be due to different production processes that result in a different individual productivity across ownership, or to particular institutional factors, such as monopolistic power that leads to the return gap. There are several ways of constructing the assumed nondiscriminatory wage structure β∗ (Jann 2008). In the following, we present decompositions using the method proposed by Neumark (1988), which assumes a pooled wage structure (including a group indicator as suggested by Jann [2008]) as the nondiscriminatory wage structure. Table 9.6 reports the changes in relative remuneration across enterprises of different ownership in urban China by applying the Oaxaca-Blinder decomposition method separately for 2002 and for 2007. The table presents the mean predictions by ownership group, their difference, and the decomposition of the difference into explained and unexplained parts (expressed in both mean value and percentage of the mean difference). The top panel in Table 9.6 shows the log hourly earnings decomposition results by ownership for 2002. The earnings gaps are rather large, especially between the public sector and the PIEs, as well as between the FIEs and the PIEs. In addition, except for the gap between the SOEs and the FIEs, all

Table 9.6. Oaxaca-Blinder decomposition of log hourly wages by ownership Year 2002 Average log earnings

SOEs-GAIs SOEs-UCEs SOEs-PIEs 350

SOEs-FIEs GAIs-UCEs GAIs-PIEs GAIs-FIEs UCEs-PIEs UCEs-FIEs PIEs-FIEs

Decomposition

Group A

Group B

Difference (A – B)

Explained

Percentage

Unexplained

Percentage

N

1.490*** (0.0141) 1.490*** (0.0141) 1.490*** (0.0141) 1.490*** (0.0141) 1.762*** (0.0154) 1.762*** (0.0154) 1.762*** (0.0154) 1.139*** (0.0293) 1.139*** (0.0293) 0.973*** (0.0266)

1.762*** (0.0154) 1.139*** (0.0293) 0.973*** (0.0266) 1.547*** (0.0587) 1.139*** (0.0293) 0.973*** (0.0266) 1.547*** (0.0587) 0.973*** (0.0266) 1.547*** (0.0588) 1.547*** (0.0587)

−0.271*** (0.0209) 0.351*** (0.0325) 0.517*** (0.0301) −0.0566 (0.0604) 0.622*** (0.0331) 0.789*** (0.0307) 0.215*** (0.0607) 0.166*** (0.0396) −0.408*** (0.0657) −0.574*** (0.0645)

0.0115 (0.0150) 0.105*** (0.0182) 0.256*** (0.0189) −0.00777 (0.0262) 0.0726*** (0.0200) 0.217*** (0.0194) 0.0323 (0.0308) 0.119*** (0.0242) −0.178*** (0.0428) −0.334*** (0.0439)

−4.2%

−0.283*** (0.0219) 0.246*** (0.0324) 0.261*** (0.0274) −0.0488 (0.0556) 0.550*** (0.0340) 0.572*** (0.0307) 0.183** (0.0589) 0.0473 (0.0369) −0.230*** (0.0665) −0.240*** (0.0607)

104.2%

3,594

70.1%

2,289

50.5%

3,212

86.3%

2,023

88.3%

2,091

72.5%

3,014

85.0%

1,825

28.3%

1,709

56.4%

520

41.8%

1,443

29.9% 49.5% 13.7% 11.7% 27.5% 15.0% 71.7% 43.6% 58.2%

Year 2007 Average log earnings

SOEs-GAIs SOEs-UCEs SOEs-PIEs SOEs-FIEs 351

GAIs-UCEs GAIs-PIEs GAIs-FIEs UCEs-PIEs UCEs-FIEs PIEs-FIEs

Decomposition

Group A

Group B

Difference (A – B)

Explained

Percentage

Unexplained

Percentage

N

2.081*** (0.0233) 2.081*** (0.0233) 2.081*** (0.0233) 2.081*** (0.0233) 2.194*** (0.0162) 2.194*** (0.0162) 2.194*** (0.0162) 1.946*** (0.0394) 1.946*** (0.0394) 1.846*** (0.0191)

2.194*** (0.0162) 1.946*** (0.0394) 1.846*** (0.0191) 2.312*** (0.0532) 1.946*** (0.0394) 1.846*** (0.0191) 2.312*** (0.0532) 1.846*** (0.0191) 2.312*** (0.0532) 2.312*** (0.0532)

−0.113*** (0.0284) 0.135** (0.0458) 0.235*** (0.0301) −0.231*** (0.0581) 0.248*** (0.0426) 0.348*** (0.0250) −0.118* (0.0556) 0.100* (0.0438) −0.366*** (0.0663) −0.466*** (0.0565)

−0.0121 (0.0157) 0.0496* (0.0235) 0.140*** (0.0237) −0.0976** (0.0333) 0.125*** (0.0223) 0.232*** (0.0162) −0.00764 (0.0308) 0.0886*** (0.0229) −0.180*** (0.0414) −0.254*** (0.0319)

10.7%

−0.101*** (0.0297) 0.0854 (0.0461) 0.0955** (0.0349) −0.134* (0.0598) 0.123** (0.0398) 0.116*** (0.0244) −0.111* (0.0494) 0.0114 (0.0406) −0.186** (0.0647) −0.212*** (0.0528)

89.3%

2,912

63.3%

1,233

40.4%

2,600

57.7%

1,120

49.6%

2,249

33.3%

3,616

93.5%

2,136

11.4%

1,937

50.8%

457

45.5%

1,824

36.7% 59.6% 42.3% 50.4% 66.7% 6.5% 88.6% 49.2% 54.5%

Notes: See Table 9.2. Standard errors in parentheses. Decompositions based on regression results are presented in Tables 9.4 and 9.5. Earnings are deflated using the urban provincial-level spatial price deflators calculated by Brandt and Holz (2006), and updated for 2007. Base: Nationwide prices in 2002. * p < 0.05, ** p < 0.01, and *** p < 0.001. Source: See Table 9.2.

352

Yang Juan, Sylvie D´emurger, and Li Shi

ownership differences are significantly different from zero. The results of the decomposition reveal that differences in endowment account for a rather small share of the earnings gap for all pairs of sectors, except for the UCE-PIE and PIE-FIE pairs. Hence, in 2002 the unexplained part accounted for most of the observed difference, thereby corroborating the findings by D´emurger et al. (2006) that show the segmentation effect across ownership is fairly serious in urban China. The most striking example of such segmentation can be observed within the public sector: differences in endowments between SOEs and GAIs are negligible, and the 27 percent earnings gap is entirely due to the “unexplained” component, which probably reflects the very strong institutional protection of workers in GAIs at the turn of the century (D´emurger et al. 2006). The same applies between the GAIs and FIEs, the former clearly appearing to be a protected sector as compared to the foreign sector. Compared to 2002, the log hourly earnings gaps across ownership in 2007 were substantially reduced for all pairs of sectors, except for between SOEs and FIEs and for between GAIs and FIEs, for which the gap turned significantly in favor of FIEs. The evolution was generally in favor of both the private and semipublic sectors (PIEs, FIEs, and UCEs), and at the expense of the public sector, mostly the GAIs that had gained substantially during the 1995–2002 period. As already observed in the descriptive part of this chapter, the trend during the 2002–2007 period was toward a rebalancing between the different ownerships. The pattern of decomposition across ownership also changed remarkably between 2002 and 2007, with a striking reversal in the contributions of the explained and unexplained parts in the earnings differentials. Differences in endowment gained importance in accounting for the earnings gaps in 2007 as well as for the general decreasing trend in the earnings differences across ownership, whereas segmentation began to be less important. The decomposition analysis presented in Table 9.6 highlights three main phenomena on the ownership dimension that are important to understand the recent evolution of the labor market in urban China. First, urban collectives and private enterprises, as compared to the public sector, saw their relative position improve dramatically. Indeed, compared to both SOEs and GAIs, the huge decrease in the earnings gap came from two concomitant forces: a convergence in endowments, on the one hand, and a sharp reduction of segmentation against UCEs and PIEs, on the other. This change is important in the sense that it signals an unprecedented better integration of the domestic sectors – public, semipublic, and private.

Do Employees in the Public Sector Still Enjoy Earnings Advantages?

353

Second, although the dominant position of GAIs declined between 2002 and 2007, the still comparatively higher wages in GAIs may be attributed to the employees’ better endowments as compared to those in other sectors. This is especially the case when compared to UCEs and PIEs: differences in endowments account for 50 percent and 67 percent of the earnings gap with GAIs in 2007, whereas these shares were only 12 percent and 28 percent, respectively, in 2002. That is, the strong increase in segmentation in favor of GAIs that was observed in the early 2000s (D´emurger et al. 2006) vanished in the more recent period, both in absolute terms and as a share of the log earnings differences, which may indicate a trend toward less protection of earnings in the public sector. Third, the foreign sector continued to reinforce its position through both better characteristics and more pronounced segmentation, especially compared to the public sector. Interestingly, the sharp increase in the earnings gap between SOEs and FIEs (in favor of the latter) between 2002 and 2007 was due to both diverging characteristics (that explain almost half the gap in 2007) and increasing segmentation. In 2007, if there were no differences in characteristics between SOEs and FIEs, the premium for FIEs would be 13 percent. A premium of a similar magnitude due to the “unexplained” part applies to the difference with GAIs, which explains the entire gap since the characteristics of workers in FIEs and GAIs are very similar. Finally, compared with UCEs and PIEs, the position of FIEs did not change considerably: both the better characteristics in FIEs and the rather strong segmentation contributed almost equally to the still important earnings gaps of 37 percent with UCEs and 47 percent with PIEs.

VI. Juhn-Murphy-Pierce Decomposition The Oaxaca-Blinder decomposition approach deals only with the mean of the distribution and it ignores differences along the distribution, for instance its dispersion or skewness. However, as shown in Section III, the distribution of hourly earnings differs across sectors. Hence, to complement the OaxacaBlinder decomposition, we use the decomposition technique proposed by Juhn, Murphy, and Pierce (1993) that takes into account the entire earnings distribution. The Juhn-Murphy-Pierce decomposition method extends the OaxacaBlinder approach by accounting for the residual distribution so that the hourly earnings gap can be decomposed into three parts: the individual characteristics effect (resulting from a change in the distribution of the Xs),

354

Yang Juan, Sylvie D´emurger, and Li Shi

the return or “price” effect (resulting from a change in the βs), and the residual effect (or the influence of the unobservable factors). Following Juhn et al. (1993), the residual uir in Equation (1) can be written as follows: uir = F r−1 (θir |X ir ),

(5)

where θir is the percentile of an individual in the residual distribution, and Fr is the cumulative distribution function of the earnings equation residuals (for individuals with characteristics Xir in ownership r). Assuming that F ∗ is a reference residual distribution and β∗ is a reference wage structure,13 two hypothetical hourly earnings distributions can be created as follows: −1

w 1ir1 = β∗ X ir1 + F ∗ (θir1 |X ir1 ) −1

w 2ir1 = βr1 X ir1 + F ∗ (θir1 |X ir1 ).

(6) (7)

The first hypothetical set of wages given in Equation (6) is computed by valuing each worker’s characteristics X ir1 in sector r1 at the reference wage structure β∗ and her position in sector r1 s residual distribution at the corresponding position in the reference residual distribution F ∗ . The second hypothetical distribution for sector r1 given in Equation (7) results from giving each worker his or her own estimated returns to characteristics βr1 but the reference residual distribution F ∗ . A main feature of the Juhn-Murphy-Pierce decomposition approach is that it allows for an analysis over the entire earnings distribution. If the notation w˜ indicates a summary statistic of the distribution of the corresponding variable, one can then write the following decomposition of the log earnings difference between two enterprises of different ownership, r1 and r2 :  2  w r2 =  w 1r1 −  w 1r2 + ( w r1 −  w 2r2 ) − ( w 1r1 −  w 1r2 )  w r1 −    w r2 ) − ( w 2r1 −  w 2r2 ) . (8) + ( w r1 −  Given the preceding definitions, the first right-hand-side term simply reflects the individual characteristics effect, or the difference in observable quantities between the two sectors. The second term (in brackets) represents the return effect, or the difference in observable prices, and the 13

In the Oaxaca-Blinder decomposition, the reference wage structure is estimated from a pooled model over the entire sample. The reference residual distribution is the average distribution over both samples. The decomposition results presented here are generated using the jmpierce.ado program for Stata.

Do Employees in the Public Sector Still Enjoy Earnings Advantages?

355

third term represents the residual effect, expressed by the difference of the two sectors’ residual distribution. The results of Juhn-Murphy-Pierce decompositions for each ownership pair are displayed in Figure 9.3. Each subfigure presents the earnings gap as well as its decomposition for an ownership pair at various percentiles: 5th, 10th, 25th, 50th, 75th, 90th, and 95th. Four main observations can be drawn from these figures. First, the distribution of the earnings gaps varies markedly across ownership pairs. A comparison of any ownership with the domestic private sector (i.e., the following pairs: SOEs-PIEs, GAIs-PIEs, UCEs-PIEs, and FIEs-PIES) shows that the largest gap occurs at the bottom of the distribution, but it almost vanishes at the top of the distribution in both 2002 and 2007. This means that the significant average earnings gaps observed between these categories of ownership are mainly due to individuals in the bottom 5th to 10th percentile, with the private sector paying much less than any other category. However, the pattern is completely reversed when comparing UCEs and FIEs in 2002: the earnings gap for the lowest wage earners is fairly small, but it increases significantly when moving up the income distribution. This trend reflects the patterns observed in Figure 9.2, with “high-wage” earners in the foreign sector receiving much higher remuneration than high-wage earners in the semipublic sector in 2002. Finally, the profile for the earning gaps between SOEs and UCEs, between GAIs and UCEs, as well as between SOEs and GAIs is rather flat in 2002. This indicates comparatively fairly equal distributions of the earnings gaps in the public and semipublic sectors, as the difference between the top and bottom percentiles is not substantial. Second, the decomposition of the earnings gaps confirms that individual endowments explain only a small share of the observed gaps within the public and semipublic sectors (SOEs, GAIs, and UCEs), whereas the segmentation (or price) effect is the largest, with no significant variations across the distribution. When compared with the private sector, the quantity component becomes relatively more important, explaining about half of the earnings difference between FIEs and UCEs and between FIEs and PIEs. Finally, the residual effect (unobserved factors) does not play any clear-cut role in explaining the earnings difference, except at the bottom of the distribution for the SOE-PIE, GAI-PIE, and UCE-PIE pairs. Third, the comparison between SOEs and FIEs merits specific comment because the gap varies greatly over the earnings distribution and important changes occurred over time. In 2002, SOEs were paying comparatively higher average wages to the lowest-wage earners, whereas FIEs were offering

JMP decomposition - GAIs vs PIEs, 2002

JMP decomposition - SOEs vs GAIs, 2002 0,1

1,4

0,05

1,2

0 -0,05

1

p5

p10

p25

p50

p75

p90

p95

0,8

-0,1

0,6

-0,15

0,4

-0,2

0,2

-0,25

0

-0,3

-0,2

-0,35

-0,4

T

Q

P

p5

p10

U

p25

T

JMP decomposition - SOEs vs UCEs, 2002

p50

p75

Q

p90

P

p95

U

JMP decomposition - GAIs vs FIEs, 2002

0,5

0,4 0,35

0,4

0,3

0,3

0,25

0,2

0,15

0,2 0,1

0,1

0,05 0

0 p5

p10

p25

p50

p75

p90

p95

-0,1

-0,05

p5

p10

p25

p50

p75

p90

p95

-0,1

T

Q

P

U

T

JMP decomposition - SOEs vs PIEs, 2002

Q

P

U

JMP decomposition - UCEs vs PIEs, 2002

1,2

1

1

0,8

0,8 0,6

0,6 0,4

0,4

0,2

0,2

0

0 -0,2

p5

p10

p25

p50

p75

p90

p95

-0,4

-0,2

p5

T

Q

P

U

p25

T

JMP decomposition - SOEs vs FIEs, 2002

p50

p75

p90

p95

Q

P

U

JMP decomposition - UCEs vs FIEs, 2002

0,2

0,2

0,15

0,1

0,1

0

0,05

-0,1

0 -0,05

p10

-0,4

p5

p10

p25

p50

p75

p90

p95

-0,2

p5

p10

p25

p50

p75

p90

p95 -0,3

-0,1

-0,4

-0,15

-0,5

-0,2

-0,6

T

Q

P

U

T

JMP decomposition - GAIs vs UCEs, 2002

Q

P

U

JMP decomposition - PIEs vs FIEs, 2002

0,8

0,4

0,7 0,2

0,6 0,5

0

0,4 -0,2

0,3 0,2

p5

p10

p25

p50

p75

p90

p95

-0,4

0,1 -0,6

0 -0,1

p5

p10

p25

p50

p75

p90

p95

-0,2

T

Q

P

U

-0,8 -1

T

Q

P

U

Figure 9.3. Juhn-Murphy-Pierce Decomposition of Log Hourly Wages by Ownership.

JMP decomposition - GAIs vs PIEs, 2007

JMP decomposition - SOEs vs GAIs, 2007 0,1

0,5

0,05

0,4 0,3

0 p5

p10

p25

p50

p75

p90

p95

0,2

-0,05 0,1

-0,1

0

-0,15

-0,1

-0,2

p5

p10

p25

p50

p75

p90

p95

-0,2

T

Q

P

U

T

JMP decomposition - SOEs vs UCEs, 2007

Q

P

U

JMP decomposition - GAIs vs FIEs, 2007

0,25

0,1

0,2

0,05 0

0,15

-0,05

0,1

p5

p10

p25

p50

p75

p90

p95

-0,1

0,05 -0,15

0 -0,05

p5

p10

p25

p50

p75

p90

p95

-0,2 -0,25

-0,1

-0,3

T

Q

P

U

T

JMP decomposition - SOEs vs PIEs, 2007

Q

P

U

JMP decomposition - UCEs vs PIEs, 2007

0,5

0,3

0,4

0,25 0,2

0,3

0,15

0,2

0,1

0,1

0,05

0 -0,1

p5

p10

p25

p50

p75

p90

0

p95

-0,05

-0,2

p5

p10

p25

p50

p75

p90

p95

-0,1

T

Q

P

U

T

JMP decomposition - SOEs vs FIEs, 2007

Q

P

U

JMP decomposition - UCEs vs FIEs, 2007

0,1

0,1

0,05 0

0 -0,05

p5

p10

p25

p50

p75

p90

p5

p95

p10

p25

p50

p75

p90

p95

-0,1

-0,1 -0,15

-0,2

-0,2 -0,3

-0,25 -0,3

-0,4

-0,35 -0,4

-0,5

T

Q

P

U

T

JMP decomposition - GAIs vs UCEs, 2007

Q

P

U

JMP decomposition - PIEs vs FIEs, 2007

0,4

0,2

0,35

0,1

0,3 0,25

0

0,2

-0,1

0,15

p5

p10

p25

p50

p75

p90

p95

-0,2

0,1 -0,3

0,05

-0,4

0 -0,05

p5

p10

p25

p50

p75

p90

p95

-0,1

-0,5 -0,6

T

Q

P

U

Figure 9.3 (continued)

T

Q

P

U

358

Yang Juan, Sylvie D´emurger, and Li Shi

higher wages to the 75th percentile, thereby changing the sign over the distribution (and possibly explaining why the mean difference reported in Table 9.6 is not significant). Interestingly, although the gap in favor of SOEs at the bottom of the distribution appears to be explained equally by differences in quantity, price, and residuals, the gap in favor of FIEs at the top of the distribution is mainly explained by different remuneration characteristics (that more than compensate for the better characteristics of SOE workers). Fourth, as previously discussed, earnings differentials were substantially reduced between 2002 and 2007 for almost all pairs of sectors. Whole distributions provide a more complete view of this average evolution by highlighting some differences along the earnings distribution. Hence, the reduction in the earnings gap tends to be more pronounced at the bottom of the distribution, due to the decreasing segmentation. This is particularly the case for the SOE-GAI and UCE-GAI pairs, suggesting that in the public sector, the wage structure has become more harmonized for low-wage earners. In addition, distribution patterns for different ownerships at various percentiles changed considerably, suggesting that wage-setting mechanisms experienced major changes during this period. In this respect, the foreign sector exhibits particularly interesting changes. Indeed, the position of FIEs clearly improved relative to both SOEs and GAIs, with the gap in 2007 fully favoring the FIEs over the whole distribution, and with very clear differences at the top of the distribution, almost fully explained by segmentation in favor of FIEs. This probably reflects a proactive strategy by FIEs toward high-wage earners (this was already visible in 2002, although it was less clear-cut). Interestingly, the smallest earnings differential for the FIE-SOE and FIE-GAI pairs is around the 25th percentile, which indicates that for individuals below the median, wages across these ownerships are quite similar. Finally, the 2007 figures also indicate that segmentation still played a fairly important role in explaining earnings gaps across ownership, with a particularly pronounced importance at the top of the distribution.

VII. Conclusion This chapter analyzes wage inequality trends across ownership during the 2002–2007 period and investigates the reasons for the gap by decomposing the difference in mean wages using the Oaxaca-Blinder technique and analyzing the wage-gap distribution using the Juhn-Murphy-Pierce decomposition method.

Do Employees in the Public Sector Still Enjoy Earnings Advantages?

359

We find that although average earnings gaps were still fairly large across ownership sectors in 2002, they decreased by 2007. Moreover, the observed earnings convergence took place in favor of the private and semipublic sectors, as opposed to the public sector. In terms of earnings differentials across the distribution, the Juhn-Murphy-Pierce decomposition highlights a comparatively fairly equal distribution within the public sector, whereas most of the gap for the private domestic sector came from the bottom of the distribution. As for foreign-invested enterprises, the clear improvement in their position with regard to the public sector (SOEs and GAIs) between 2002 and 2007 is observable across the entire distribution, implying that workers in foreign-invested firms benefited from the improved position of these enterprises. The Oaxaca-Blinder and Juhn-Murphy-Pierce decompositions both show that differences in endowments gained importance over time in accounting for the earnings gaps as well as for the generally decreasing trend in earnings differences across ownership. However, segmentation was less important in 2007 as compared to 2002. Indeed, our results highlight a better integration of the domestic sectors over time. They also show that segmentation in favor of GAIs, which was fairly strong in 2002, vanished over time, although not throughout the entire distribution. In particular, the Juhn-Murphy-Pierce decompositions indicate that segmentation remained important for high-wage earners, as compared to workers at the bottom of the distribution, suggesting that workers at the top of the distribution were still benefiting from some protection. References Adamchick, V.A. and A.S. Bedi (2000), “Wage Differentials between the Public and the Private Sectors: Evidence from an Economy in Transition,” Labour Economics, 7(2), 203–224. Blinder, A.S. (1973), “Wage Discrimination: Reduced Form and Structural Estimates,” Journal of Human Resources, 8(4), 436–455. Boeri, T. and K. Terrell (2002), “Institutional Determinants of Labor Reallocation in Transition,” Journal of Economic Perspectives, 16(1), 51–76. Brandt, L. and C.A. Holz (2006), “Spatial Price Differences in China: Estimates and Implications,” Economic Development and Cultural Change, 55(1), 43–86. Card, D. (1999), “The Causal Effect of Education on Earnings,” in O. Ashenfelter and D. Card, eds., Handbook of Labor Economics, 1801–1863, Amsterdam: North-Holland. Chen, Y., S. D´emurger, and M. Fournier (2005), “Earnings Differentials and Ownership Structure in Chinese Enterprises,” Economic Development and Cultural Change, 53(4), 933–958.

360

Yang Juan, Sylvie D´emurger, and Li Shi

D´emurger, S., M. Fournier, and Y. Chen (2007), “The Evolution of Gender Earnings Gaps and Discrimination in Urban China, 1988–95,” Developing Economies, 45(1), 97–121. D´emurger, S., M. Fournier, S. Li, and Z. Wei (2006), “Economic Liberalization with Rising Segmentation in China’s Urban Labor Market,” Asian Economic Papers, 5(3), 58–103. Dong, X. and P. Bowles (2002), “Segmentation and Discrimination in China’s Emerging Industrial Labor Market,” China Economic Review, 13(2–3), 170–196. Falaris, E.M. (2004), “Private and Public Sector Wages in Bulgaria,” Journal of Comparative Economics, 32(1), 56–72. Jann, B. (2008), “A Stata Implementation of the Blinder-Oaxaca Decomposition,” ETH Zurich Sociology Working Paper 5, ETH Zurich. Juhn, C., K.M. Murphy, and B. Pierce (1993), “Wage Inequality and the Rise in Returns to Skill,” Journal of Political Economy, 101(3), 410–442. Knight, J. and L. Song (2003), “Increasing Urban Wage Inequality in China: Extent, Elements and Evaluation,” Economics of Transition, 11(4), 597–619. Knight, J. and L. Song (2005), Towards a Labour Market in China, Oxford: Oxford University Press. Lardy, N. R. (1998), China’s Unfinished Economic Revolution, Washington, DC: Brookings Institution. Li, S. and J. Song (2010), “Changes in Gender Wage Gap in Urban China during 1995–2007,” University of Chicago and Renmin University of China International Symposium on Family and Labor Economics, September. Lokshin, M. M. and B. Jovanovic (2003), “Wage Differentials and State-Private Sector Employment Choice in Yugoslavia,” Economics of Transition, 11(3), 463– 491. Maurer-Fazio, M., T. G. Rawski, and W. Zhang (1999), “Inequality in the Rewards for Holding Up Half the Sky: Gender Wage Gaps in China’s Urban Labour Market, 1988–1994,” China Journal, no. 41, 55–88. Meng, X. and J. Zhang (2001), “The Two-Tier Labor Market in Urban China: Occupational Segregation and Wage Differentials between Urban Residents and Rural Migrants in Shanghai,” Journal of Comparative Economics, 29(3), 485–504. Mincer, J. (1974), Schooling, Experience, and Earnings, New York: National Bureau of Economic Research. National Bureau of Statistics (NBS) (2008), Zhongguo tongji nianjian 2008 (China Statistical Yearbook 2008), Beijing: Zhongguo tongji chubanshe. Naughton, B. (2007), The Chinese Economy: Transitions and Growth, Cambridge, MA: MIT Press. Neumark, D. (1988), “Employers’ Discriminatory Behavior and the Estimation of Wage Discrimination,” Journal of Human Resources, 23(3), 279–295. Oaxaca, R. (1973), “Male-Female Wage Differentials in Urban Labor Markets,” International Economic Review, 14(3), 693–709. Su, X. (1999), Xin Zhongguo jingji shi (Economic History of New China), Beijing: Zhonggong zhongyang dangxiao chubanshe. Wang, M. (ed.), (2005), China Human Development Report: Development with Equity, Beijing: China Translation and Publishing Corp.

Do Employees in the Public Sector Still Enjoy Earnings Advantages?

361

Xing, C. (2008), “Human Capital and Wage Determination in Different Ownerships, 1989–97,” in Wan Guanghua, ed., Understanding Inequality and Poverty in China: Methods and Applications, 117–136, New York: Palgrave Macmillan. Zhao, Y. (2002), “Earnings Differentials between State and Non-state Enterprises in Urban China,” Pacific Economic Review, 7(1), 181–197.

TEN

Redistributive Impacts of the Personal Income Tax in Urban China Xu Jing and Yue Ximing

I. Introduction Taxation is a major source of fiscal revenue and has a strong effect on income distribution. Taxation can either reduce inequality or make inequality worse, depending on the type and rate of taxation. Generally, a personal income tax will improve inequality, whereas a general sales tax will exacerbate inequality. This is because with the former, there is a statutory rate that increases with income, and with the latter, tax is collected from everyone based on consumption rather than on earnings. Therefore, a tax system that relies heavily on a general sales tax as opposed to a personal income tax has an adverse impact on income distribution. This is precisely the case in China. A tax is progressive if tax liability relative to income rises as income increases. Typically, the statutory rate of the personal income tax increases with income, thus it is progressive. In fact, it is usually the most progressive element in the tax system. Given that the personal income tax is progressive, the extent to which it contributes to reducing inequality depends on its rate, or the proportion of the personal income tax to the total income. As will be seen later in this chapter, the contribution of the personal income tax to reduce inequality mainly depends on two components, its progressivity and the average tax rate. Additional contributions can be derived by raising the average rate of the personal income tax while holding the progressivity constant, or vice versa. It is clear that the personal income tax in China is progressive because there is a very high statutory marginal rate.1 The extent 1

The personal income tax is administered on the basis of the itemized components of the total taxable income. Income from other sources has different exemptions, deductions, tax rates, and so on.

The authors are grateful to Terry Sicular for comments. All errors are ours.

362

Redistributive Impacts of the Personal Income Tax in Urban China

363

of the contribution of the personal income tax to inequality therefore relies on the average proportion of the tax in the total income. In this chapter, we use the China Household Income Project (CHIP) urban survey data for 2007 to analyze the impact of the personal income tax on income distribution in urban China. The CHIP urban data set contains information on income and taxes at the individual or household level, so in principle it should not be difficult to gauge the contribution of the personal income tax to income distribution. Some difficulty arises, however, due to the quality of the information that is reported by the households on the amount of personal income tax paid by the individual members of the household. On close inspection, we find that the data set substantially understates the amount of income taxes paid, probably due to underreporting and/or nonreporting. This bias, if not corrected, will lead to an underestimation of the contribution of the personal income tax to inequality. Thus, although the main task in this study is to evaluate the redistributive effect of the personal income tax, a second and necessary task is to address the bias in the reported tax data. The remainder of this chapter proceeds as follows. The next section, Section II, introduces the personal income tax in the context of the broader Chinese taxation system. In Section III we describe our methodology for measuring the redistributive effect of the personal income tax; we demonstrate the understatement of the paid taxes in our data set; and we explain how we address this understatement by applying the personal income tax schedule to derive the hypothetical tax that a taxpayer would pay based on his income. This is followed in Section IV by a discussion of estimates of the redistributive effects of the personal income tax. These estimates are calculated using the hypothetical tax levels that we have imputed, which are equal to the amount of taxes implied by the official tax schedule. Section V presents our conclusions.

II. The Personal Income Tax in the Chinese Taxation System The current Chinese tax system was established as part of the 1994 fiscal reform, which is also called the tax-sharing reform. The most notable feature of this fiscal reform is the dominance of indirect taxes as a share of the collected tax revenue. Table 10.1 presents the composition of the tax revenue by major taxes in selected years after the 1994 fiscal reform. The valueadded tax (VAT) accounts for the largest share of the total tax revenue. This share was as high as 44.4 percent in the year after the 1994 fiscal reform but thereafter it declined; however, in more recent years, it still accounted

364

Xu Jing and Yue Ximing Table 10.1. Share of major taxes in total tax revenue in selected years after the 1994 fiscal reform Tax

1995

2000

2005

2007

VAT (value-added tax) Consumption Tax Customs Duty Business Tax Corporate Income Tax Personal Income Tax Other Taxes Total

44.4 9.3 6.6 14.6 13.8 2.2 9.1 100.0

36.9 6.8 11.8 14.9 14.0 5.2 10.4 100.0

34.7 5.3 13.7 13.7 17.9 6.8 8.0 100.0

31.6 4.5 12.4 13.3 19.6 6.4 12.2 100.0

for nearly one-third of total tax revenue. The VAT is levied on the value added generated from all activities taking place in primary and secondary industries. Wholesale and retail and repairs and replacements in the tertiary sector are also subject to the VAT. The business tax is complementary to the VAT in terms of its sectoral coverage. All activities carried out in tertiary industries, with the exception of the subsectors subject to the VAT mentioned earlier and government services, are subject to the business tax. The business tax is levied on the total sales of tax-paying units and therefore it is a turnover tax. This type of tax is known to have a cascade effect, which is the main reason for implementation of the VAT. The business tax, as can be seen in Table 10.1, is the third largest tax in terms of share of tax revenue, accounting for a constant proportion of more than 13 percent of the total tax revenue. The consumption tax is a type of excise tax charged on selected goods such as tobacco, alcohol, jewelry, motorcycles, motor vehicles, and so on.2 Customs duties, which consist of value-added and consumption taxes levied on imported goods as opposed to goods produced domestically, have provided about 12 percent of the total tax revenue since 2000. The preceding taxes are the four major indirect taxes in China. Together, in 1995 they accounted for about 75 percent of the total tax revenue. Although this proportion has been declining, in 2007 it still exceeded 60 percent. Excluding customs duties, the three other major indirect taxes charged on domestic goods and services accounted for more than 68 percent in 1995 and close to 50 percent (49.3 percent) in 2007 of the total tax revenue. 2

For a complete list of goods subject to the consumption tax, see China Master Tax Guide (2007).

Redistributive Impacts of the Personal Income Tax in Urban China

365

Clearly, then, since the 1994 fiscal reform, Chinese taxation has been mainly dependent on indirect taxes. This feature has important implications for income distribution. Unlike direct taxes such as a personal income tax, indirect taxes are charged on expenditures. Because the poor tend to spend a larger proportion of their income than do the rich, paid indirect taxes as a proportion of income tends to decline with household income, a well-known regressive feature of an indirect tax. Work in progress by a coauthor of this chapter calculates the traditional tax incidence for China and concludes that, as a whole, the tax system is regressive and has adverse effects on inequality. This occurs despite the fact that the personal income tax together with the corporate income tax, which is progressive under certain assumptions, offsets some of the adverse redistributive effects of the indirect taxes, but not sufficiently to counter all of their adverse effects.3 As in most countries, China has two types of income tax; one is a corporate income tax and the other is a personal income tax (PIT). As can be seen in Table 10.1, the corporate income tax shows an upward trend in proportion to total tax revenue, accounting for close to 20 percent in recent years.4 In the literature on the incidence of taxation, the implications of the corporate income tax on income distribution remain uncertain, because it is difficult to predict who ultimately bears the costs of corporate taxes. The corporate tax can be shifted to consumers who buy goods and services produced by enterprises that are subject to the corporate tax. It can also be shifted to workers, if the owners of enterprises that are subject to the tax lower wages to reduce their corporate tax liability. After such shifts in the tax burden, the remaining liability rests with the owners of the enterprises that are subject to the tax. To the extent that some of the corporate tax is shifted to consumers and/or workers, the corporate tax becomes less progressive than otherwise. Unlike the taxes discussed earlier, in general the personal income tax is progressive. Except for limited groups of special-skilled persons who are able to bargain over their after-tax pay, and for workers who are members of strong labor unions, the personal income tax is borne by the person subject to the tax. Such is the case in China, where labor unions are not strong. Because the statutory rate of the personal income tax is typically higher for 3

4

Personal communication that summarizes findings reported in an unpublished 2010 draft paper by Fukao et al. “Who Bears the Tax in China? An Applied Study of the Incidence of Tax.” The corporate tax as a share of total tax revenue is based on the total revenue of the corporate tax paid by both domestic enterprises and FDI enterprises. Before 2007, the year under study, the corporate tax differed for domestic enterprises and for FDI enterprises, with preferential treatment for the latter. However, the two taxes have been integrated since 2008.

366

Xu Jing and Yue Ximing

taxpayers with higher incomes, the average rate of the tax will rise with the income of the taxpayers. This is the case in China, as discussed more fully later. The current personal income tax in China is administered based on itemized income. More specifically, income subject to the PIT is split into different categories, each of which has a different exemption, deduction, tax rate, and so on. The personal tax liability is the sum of the taxes paid in the different categories of taxable income. Table 10A.1 in the Appendix to this chapter summarizes the main elements in the personal income tax in 2007.

III. Methodology and Data Issues The purpose of this study is to measure how and to what extent the personal income tax affects inequality in urban China. The most frequently used measure of the redistributive effects of taxes is an index proposed by Musgrave and Thin (1948), called the MT index. The MT index is defined as the difference between the Gini coefficients of before-tax and after-tax income, which is expressed by the following formula: MT = G − G ∗ ,

(1)

where MT stands for the MT index. G and G ∗ represent the Gini coefficients for before- and after-tax income, respectively. If the tax reduces inequality, then after-tax inequality as measured by the after-tax Gini coefficient should be lower than before-tax inequality, and the MT index should have a positive value. Conversely, if the tax has dis-equalizing effects, then the MT index will take a negative value. The sign and size of the MT index is used as an indicator of how and to what degree taxes influence the income distribution. The progressivity of a tax is another important indicator used to assess redistributive effects. A tax is proportional if its rate remains constant with income, and it is progressive (regressive) if the rate goes up (or down) with income. The most common measure of progressivity, called a P index, was proposed by Kakwani (1977). A P index is defined as the concentration ratio of the tax minus the Gini coefficient of before-tax income. That is, P = C − G,

(2)

where P denotes the P index of the tax and C denotes the concentration ratio of the tax. The concentration ratio of a tax is an indicator of the distribution of the tax liability among the population in association with

Redistributive Impacts of the Personal Income Tax in Urban China

367

income.5 If the concentration of a tax equals the Gini coefficient of the before-tax income (i.e., the P index takes a value of 0), it means that the tax has the same distribution among the population as that of income in the sense that the proportion of tax liability to income is constant for any individual in the population, or that each individual’s share of the total tax liability equals that individual’s share of the total income. This implies that the tax is proportional. If the concentration ratio of the tax is larger than the Gini coefficient of before-tax income (i.e., the P index takes a positive value), then the tax share is higher than the income share for individuals with higher incomes and lower for individuals with lower incomes.6 Thus, it is a progressive tax. A regressive tax can be defined similarly. Kakwani (1984) makes a distinction between the measure of the redistributive effects of the tax and the measure of the progressivity of the tax by decomposing the MT index into two components as follows: MT = (C d − G ∗ ) +

tP , 1−t

(3)

where C d stands for the concentration ratio of after-tax income7 and t is the average rate of the tax. An explanation of the two terms on the right-hand side of Equation (3) is linked to the principle of two types of tax equity: horizontal equity and vertical equity. In fact, the two terms measure the effects of the two types of tax equity on income distribution. The principle of horizontal tax equity requires that equals should be treated equally. The equals are defined in terms of ability to pay (the tax); observed income is commonly used as a measurable indicator of the ability to pay. In other words, the principle of horizontal tax equity requires that individuals with identical incomes should pay the same amount of tax. Whether, and to what extent, the principle of horizontal tax equity is violated is examined 5

6

7

A rigorous definition of the tax concentration ratio requires a concentration curve of the tax. The concentration curve is a curve with the accumulated share of the population sorted by income in ascending order measured on the horizontal axis, and the share of tax liability of the corresponding population measured on the vertical axis. The concentration curve is exactly the same as the Lorenz income curve, except that in the former the population is sorted by income, not by tax. The concentration ratio is computed on exactly the same basis on the concentration curve as the Gini coefficient is computed on the Lorenz curve. In a diagram of the Lorenz curve, the concentration curve lies farther away than the Lorenz curve from the diagonal, and therefore, the area between the diagonal and the concentration curve is larger than the area between the diagonal and the Lorenz curve; in this case, the P index is positive. Note that the concentration ratio of after-tax income is derived from the concentration curve of after-tax income, and the concentration curve of after-tax income is plotted with the income units sorted according to before-tax income, not after-tax income.

368

Xu Jing and Yue Ximing

empirically by comparing the rankings of individuals in a posttax distribution of income to their rankings in a pretax distribution of income. Rosen (1978), following a suggestion by Feldstein (1976), uses the rank correlation coefficients between the pre- and posttax ordering of the estimated utilities to measure departures from horizontal equity. The measure, as pointed out by Kakwani (1984), has no clear link to any indicator of inequality, thus the effect of horizontal inequity on the distribution of income cannot be measured. Atkinson (1980) employs the concentration ratio and Gini coefficient of after-tax income to examine the effect on equality of a change in the ranking of an individual by taxation, but does not propose any summary indicator. Plotnick (1981) uses a similar measure as the first term on the left-hand side of Equation (3) to measure the horizontal inequity, but he does not link the measure to indices of inequality. Horizontal equity, examined by a reranking of individuals based on aftertax income, requires that the after-tax income should not change the beforetax ranking of individuals. If horizontal equity is violated, the redistributive effects of taxation on inequality are affected. In other words, given the progressivity of the taxation, a reduction in the Gini coefficient of after-tax income would be moderate if there were a change in the pre- and posttax income rankings of individuals. The first term on the right-hand side of Equation (3) captures this point. Where C d is the concentration ratio of the after-tax income, with the ranking by the before-tax income, G ∗ is the Gini coefficient of the after-tax income, with the ranking of the after-tax income. C d equals G ∗ if the rankings of the pretax and posttax income are identical. Otherwise, as demonstrated by Kakwani (1980), Atkinson (1980), and Plotnick (1981), the former is smaller than the latter. This means that C d − G ∗ , the first term on the right-hand side of the equation, takes a maximum value of 0 when the tax does not change the ranking of individuals, and takes a negative value whenever there is a change in ranking from the pretax to the posttax income. It is clear that, given the progressivity of the taxation (the second term on the right-hand side of Equation [3]), the redistributive effect of taxation on inequality is reduced (the MT index will take a small value) when there is a change in the ranking from the pretax and posttax income (i.e., C d − G ∗ takes negative values). The second term on the right-hand side of Equation (3) measures the contribution of the progressivity of the tax to inequality. This term is related to the principle of vertical tax equity, which means that people with different incomes are taxed differently. P is a widely used indicator proposed by Kakwani to measure the tax progressivity. As discussed earlier, P takes a value of zero if the tax rate is constant among people with different incomes;

Redistributive Impacts of the Personal Income Tax in Urban China

369

it takes a negative value if the tax rate decreases with income, and it takes a positive value if the tax rate increases with income. A progressive tax, that is, a tax whose rate increases with income, will contribute to a reduction in inequality, but inequality will remain unchanged if the tax is proportional, and inequality may become worse if the tax is regressive. Note that, with C d − G ∗ given, the extent to which the tax contributes to inequality depends not only on how progressive the tax is (captured by P in the second term on the right-hand side of Equation [3]), but also on how high the tax rate is, which is measured by t/(1 + t) in the term. It is clear that given the progressivity of the tax, a higher tax rate will result in a higher contribution of tax to inequality. Like other chapters in this book, the data used in this study come from the 2007 China Household Income Project (CHIP). We limit our analysis to the CHIP urban sample because the current PIT in China is limited to residents in urban areas. The migrant sample is not included in this study due to data availability.8 Thus, our sample of 10,000 households includes only urban residents. In addition to information on the total income of individuals and the major components of their income, the CHIP urban data set also contains information on taxes paid by individuals. Whereas the personal income tax is administered based on itemized income, which means taxpayers pay their income tax by the category of the income they earn, taxes in the data set are not listed separately according to the income components. Only the aggregate amount of the income tax is provided. On inspection, based on two checks of the data, we find that the available information on taxes in the data set understates the taxes paid by individuals. First, we compare the average PIT rate found in our data set with that derived from tax data published by the State Administration of Taxation (SAT), the government agency responsible for collecting the PIT and other taxes, which publishes annual total tax revenue and revenue by tax, including revenue from the PIT. Information on the total PIT collected each year published by the SAT can be considered to be accurate.9 We combine this tax information with the 8

9

According to law, migrant workers in cities also have to pay personal income tax; however, income data from the migrant sample of the 2007 CHIP are given for the family as a whole, not for each of the family members. This presents an obstacle to estimations of the tax liabilities for wages and salaries, which are levied on an individual basis. For this reason, we do not incorporate migrants from the CHIP migrant survey in our analysis. It is worth noting that the accuracy of the tax data published by the SAT differs from tax evasion. It is widely believed that there is huge PIT evasion in China. This tax evasion, however, is irrelevant to the accuracy of the information on tax revenue actually collected and published by the tax authorities.

370

Xu Jing and Yue Ximing Table 10.2. Comparison of household data average tax rates and alternative data average tax rates Data source Year

2002

2007

Household data Alternative data

0.33 2.95

0.85 3.60

total gross income of urban households published by the National Bureau of Statistics (NBS), based on the NBS household survey,10 to provide an estimate of the average rate of the PIT for urban households as a whole. The resulting tax rates, as shown in Table 10.2, are 2.95 percent in 2002 and 3.60 percent in 2007. If the available information on taxes in the CHIP urban data set is accurate and free of underreporting or nonreporting, it should yield an average PIT rate close to the average rate obtained using the SAT and NBS information, on the assumption that our urban sample is representative of all urban households in terms of income. The average rate of the PIT based on our data set, however, is much lower than that calculated from the SAT and NBS data: 0.33 percent in 2002 and 0.85 percent in 2007,11 far below the rates reported in Table 10.2. Our second check is to look at the information on income and taxes in our data set to see whether all individuals with positive taxable income report taxes.12 We limit this check to the two major income components subject to the PIT. One is wages and salary, and the other is business-operating income. Under the current Chinese personal income tax law, any person who earned wages and salary of more than 800 yuan in any month of 2002 (1,600 yuan in 2007), which were the allowable deductions in those two years, must pay an income tax. Of 5,137 individuals whose average monthly income was above 800 yuan in 2002, 64.6 percent, or 3,319 individuals, reported no tax. This is a fairly large portion. The nonreporting of tax might be more likely for individuals with low wages than for individuals with high wages, probably 10 11

12

The total gross income of urban households is derived from the gross urban income per capita times the total urban population. The average rate of the PIT is derived from the ratio of the total tax to the total gross household income (times 100). It is actually the weighted average of the PIT rate of each individual, with the income of each individual as the weight. One might argue that zero tax for a taxpayer with a positive taxable income could be due to tax evasion rather than to nonreporting as we claim here. If this were the case, the average tax rate derived from our data set would be close to that based on the macro data. However, this is not the case.

Redistributive Impacts of the Personal Income Tax in Urban China

371

Table 10.3. Mean income and proportion of individuals (non)reporting the personal income tax Proportion reporting tax (%)

Mean taxable income (yuan/monthly)*

Number of observations (persons)

No

Yes

Total

2002 1 2 3 4 5 Total

70 215 389 657 1,551 576

1,029 1,026 1,028 1,028 1,026 5,137

84.7 75.0 63.3 54.0 46.2 64.6

15.3 25.0 36.7 46.0 53.8 35.4

100 100 100 100 100 100

2007 1 2 3 4 5 Total

143 461 894 1,623 3,938 1,411

1,345 1,325 1,334 1,335 1,334 6,673

79.3 67.7 49.6 37.6 28.0 52.5

20.7 32.3 50.4 62.4 72.0 47.5

100 100 100 100 100 100

Quintile*

*

The mean taxable income is the gross monthly income minus the allowable deduction of 800 yuan in 2002 and 1,600 yuan in 2007. The quintiles are obtained based on the gross monthly income. The gross monthly income for each individual is derived from the annual income in our data set divided by twelve (months). Ideally, the monthly gross income actually earned by individuals in a certain month should be used. This is impossible, however, because in our data set we only have information on the annual wage and salary. But the mean monthly income still meets our requirements. This is because if the mean monthly income of an individual exceeds 1,600 yuan, then during at least one month in the year under review this individual must have earned more than 1,600 yuan, and this individual should have reported and paid the PIT.

due to the fact that the small amount of tax owed by the former is easier to overlook than the large amount owed by the latter. In order to determine whether this is the case, we look at the proportion of individuals failing to report any tax by quintiles of taxable wages. The results are summarized in Table 10.3 for 2002 and 2007. The table confirms our point. In 2002, the proportion of individuals with positive taxable income who did not report paying taxes dropped significantly by quintile of the income distribution, from 84.7 percent in the lowest quintile to 46.2 percent in the highest quintile. The picture does not change for 2007, with the exception that the proportion of individuals reporting no tax decreased both for the entire sample and for each quintile, reflecting an improvement in the administration of the personal income tax.

372

Xu Jing and Yue Ximing

Table 10.4. Mean business operating income and the proportion of individuals (non)reporting the personal income tax Proportion of whether reporting tax (%)

Mean taxable income (yuan/annual)

Number of observations (persons)

No

Yes

Total

2002 1 2 3 4 5 Total

1,153 3,548 5,691 8,902 19,539 7,760

145 141 143 143 143 715

95.9 100.0 96.5 93.7 88.8 95.0

4.1 0.0 3.5 6.3 11.2 5.0

100 100 100 100 100 100

2007 1 2 3 4 5 Total

154 486 1,046 2,089 7,069 2,156

119 111 121 109 115 575

5.88 6.31 10.74 8.26 15.65 9.39

100 100 100 100 100 100

Quintile

94.12 93.69 89.26 91.74 84.35 90.61

The nonreporting of tax also occurs among individuals who earn business-operating income. Under the personal income tax law, any person must pay the PIT on her/his positive net operational income. Table 10.4 presents information on the nonreporting of tax by quintile of net operating business income. As can be seen in the table, 715 individuals in our data set in 2002 earned a positive net operating income, of which only 5.0 percent reported a positive tax for this category of income. In 2007, the proportion was higher at 9.39 percent, another sign of an improvement in the administration of taxation. The portion reporting no tax among earners of business operating income is much lower than that of wage earners. The proportion of individuals who report no tax on net operating income is lower in the lower and higher quintiles and higher in the middle quintiles in both 2002 and 2007. In view of the very large discrepancy between average tax rates implied by the SAT and NBS data and that found in the CHIP urban data sets, and also in view of the high incidence of no reported tax payments in the CHIP urban data sets, we conclude that the tax information in our data is understated and inadequate for an assessment of the redistributive effects of the PIT. The assessment can be affected by the inadequacy of information in many ways. First, the reported data from our household

Redistributive Impacts of the Personal Income Tax in Urban China

373

survey underestimate the amount of paid personal income tax and thus underestimate the redistributive effects of the tax on inequality, all else being equal, because the degree of the distributional impact of taxes on inequality depends on the tax rate, as shown in Equation (3). Second, there is a tendency that a higher proportion of individuals with lower taxable income do not report their taxes and that this will overestimate the progressivity of the tax, thus overevaluating the impact of the personal income tax on inequality. Finally, the inaccurate data in our household survey could give a different ordering of individuals based on the after-tax income, thus leading to erroneous estimates of the effects of horizontal equity, the first term on the right-hand side of Equation (3). The understatement of taxation in our data set must be corrected in order to obtain a reliable assessment of the redistributive effects of the PIT in China. One way to deal with the inaccurate data is to apply the personal income tax schedule to impute the theoretical amount of tax liability, on the assumption that taxpayers paid all the taxes for which they were liable according to the composition of their income. Even though the imputed tax is unlikely to be identical to the taxes actually paid by each taxpayer, we believe that this is still better than the reported tax information in our data set and can be used to evaluate the impact of the PIT on income distribution. The use of the imputed tax to evaluate the redistributive effect of taxes has the advantage that it provides estimates of the distributional impact of the PIT when it is fully implemented. The personal income tax in China is currently implemented on an itemized basis. That is, the total income of individuals is subdivided into many components, to each of which different deductions and tax rates apply. In order to derive the tax liability from the available information on income in our data set, the income information in our data set must be consistent with the Personal Income Tax Law (PITL) in several respects, such as the subdivisions of the total income, the definitions of each income component, and the period for which each income component is measured. We discuss in the following whether these requirements are met. Regarding the subdivision of the total income and the definitions of its components, fortunately, the components of the total income available from our household data are broadly consistent with the income categories in the PITL. Wages and salary, based on the PITL, had a monthly allowable deduction of 800 yuan in 2002 and 1,600 yuan in 2007, and taxable income, derived from the monthly total of wages and salary minus the monthly deduction, is subdivided into nine income ranges. Increasing tax rates are applied to the nine income ranges, with the lowest at 5 percent and the

374

Xu Jing and Yue Ximing

highest at 45 percent. The wage and salary component of income is available in our data as a distinct component of the total income, which can be used to calculate the tax liability for this category of income. Other labor income is further divided into two parts in the PITL. One is the so-called individual service income, including income from a wide range of activities, such as designing, drafting, testing, medical practices, lecturing, and so on,13 and the other is remuneration from manuscripts. In our data set, however, the two parts are combined into one category. This does not matter, however, because the two parts have the same deduction, 20 percent of the amount received (for amounts of 4,000 yuan and more) or 800 yuan (for amounts of less than 4,000 yuan), and also the same tax rate is applied to each of them (20 percent of taxable income).14 Production and business income in the PITL broadly correspond to the operational income from business in our data set. Both define income as net income, that is, gross income less the costs spent for making the gross income. The category of income in the PITL called income from contracted or leased operations of enterprises/institutions does not correspond to the information in our data set and thus is ignored in our calculations. Property income is divided into four categories in the PITL: (1) royalties; (2) interest, dividends, or bonuses; (3) rental income from leasing; and (4) income from the sale of property. Each of these four categories is available separately in our data set. The definitions of property income in the PITL and in the data set show that the coverage of each category of property income is broadly consistent. Although the division of total income and the definition of each component of total income in our data set are consistent with the PITL, the period for which the income is reported in our data set differs from the period for which the taxable income is measured and taxed for all components of income, except for the operating income (production and business income in the PITL). This presents a substantial obstacle in calculating the tax liability using our household survey data. The PIT on wages and salary is charged on a monthly basis, whereas income in our data set is annual income. In calculating the tax on wages and salary, we use the average monthly income from wages and salary, derived from the annual income divided by 12 (months), to estimate the amount of tax paid monthly. The monthly amount of tax times 12 (months) is used as the annual total of paid tax for this income component. The use of the 13 14

For the difference between wages and salary and individual service income, see China Master Tax Guide (2007: 83). Refer to the Appendix to this chapter for details.

Redistributive Impacts of the Personal Income Tax in Urban China

375

average monthly wages and salary will underestimate the tax liability for this component of income, as it tends to overestimate the (monthly) allowable deduction and reduce the portion of income applicable to a higher marginal tax rate for this component of income. This is always the case whenever individual income varies across months in a year. The following example is illustrative. Assume a person earns 6,000 yuan in wages in two months, 1,000 yuan in the first month and 5,000 yuan in the second month. The total individual tax liability according to the PITL is 360 yuan. The tax distribution between the two months is zero in the first month and 360 yuan in the second month. However, the estimated tax liability based on the average monthly income will be 130 yuan, that is, 65 yuan for each month. The imputed tax liability is much lower than the true tax liability. This is the case whenever an individual’s wages vary across months. For the other components of labor income (inclusive of the personal service income and the income from manuscripts), all the components of the property income (inclusive of royalties, interest and dividends, rental income from the leasing of property, and income from the sale of property), and contingency income and other income in the PITL, according to the PITL taxes are imposed on each single receipt, whereas only the annual totals of these income components are available in our data set.15 In order to calculate tax liability on these income components we need to determine the number of times that individuals receive payments of each of these income components. The number of receipts for a given annual total of these income components will affect the deductions for these income components, but will not affect the tax rates applied to the income components, except for personal service income of more than 20,000 yuan in one receipt, because a uniform tax rate of 20 percent is applied to all these income components (see the Appendix to this chapter). Given the annual total of each of these income components, an increase in the number of receipts will increase the total amount of deductions, thus reducing the amount of paid taxes. This indicates that the current PITL provides taxpayers with an incentive to increase the number of times, whenever possible, that they receive a given amount of income. By doing this, their tax burdens can be reduced. How do we determine the number of times individuals receive the given annual total of each of these income components in our data set? Because 15

The tax on income from contracted or leased operations of enterprises/institutions is charged on an annual basis, with a monthly deduction of 800 yuan in 2002 and 1,600 yuan in 2007. As noted in the text, this component is ignored in our calculations because it does not have a separate income component in our data set.

376

Xu Jing and Yue Ximing

housing rents dominate property leasing by households and payments of housing rent are normally made monthly, it is reasonable to assume twelve receipts per year for rental income from property leasing. With respect to interest from savings, based on current banking practice in China, interest payments are made quarterly for demand deposits, but only after the contracted period for fixed deposits. Dividend payments, if any (many corporations do not pay dividends for years), are normally made once per year, and there are very few cases where dividend payments are made twice per year. Therefore, it is reasonable to assume that individuals receive only interest and dividends once per year if they receive any of this type of income. For the remainder of the income components for which taxes are charged based on receipts, we assume twelve times if the individuals receive any amount of each of these income components. The reason for twelve times per year (or once a month), rather than once a year, is that monthly is the commonly used time span in accounting of economic activities, and the PITL provides an incentive, whenever possible, for individuals to make monthly rather than annual payments. Given the annual totals of each of these income components, the number of receipts differs among individuals. We ignore the differences due to a lack of relevant information. In order to test the sensitivity of the estimates of the redistributive effects to the number of receipts, we provide alternative estimates based on the assumption of payments once per year, instead of twelve times per year, for all income components that are charged on a per receipt basis. This excludes two income components, that is, rental income from the leasing of property and interest and dividends, for which we assume twelve times per year for the former and once per year for the latter. The operational business income available in our data is broadly in line with the production and business income subject to a personal income tax in the PITL, based on both the time and definition of the income component. The annual income of the former is reported in our data, which is consistent with the time span for which this income component is measured and taxed in the PITL. The operational income reported in our data is net income; that is, gross income minus costs incurred in the process of generating the gross income. Similarly, taxable production and business income is calculated as the total income minus the deductions for the cost of production and sales, which actually is net income. Because the two measures are consistent, the imputed tax reliability of this category of income should be more accurate than that of the other categories of income that are liable to taxes.

Redistributive Impacts of the Personal Income Tax in Urban China

377

Table 10.5. Average personal income tax rate by decile Tax rate (%)

Relative tax rate (lowest decile = 1)

Decile

2002

2007

2002

2007

1 2 3 4 5 6 7 8 9 10 All deciles

0.60 0.71 0.70 0.77 0.86 1.21 1.40 1.53 2.34 4.25 2.06

0.15 0.18 0.35 0.52 0.76 1.16 1.65 2.40 3.48 7.74 3.27

1.00 1.18 1.16 1.29 1.43 2.02 2.34 2.56 3.91 7.10 3.44

1.00 1.22 2.33 3.45 5.05 7.71 10.98 15.98 23.23 51.62 21.79

IV. Estimation of the Redistributive Effects of the Personal Income Tax on Income Distribution In this section, we examine the redistributive effects of the personal income tax using the imputed tax estimates derived in the last section.16 Because our purpose is to assess the impact of the personal income tax on the income distribution of the entire population, our analysis that follows covers all family members rather than only those family members who are working and paying an individual income tax. Before presenting the MT index and its decomposition, we provide the average tax rate across deciles of per capita before-tax income in Table 10.5. With very few exceptions, the average tax rate increases as we move up the income distribution in both years under review, indicating a progressive individual income tax. This is not surprising since the statutory rate increases with income earned for most income components in the PITL. The average tax rate for the highest deciles was above seven times that for the lowest deciles in 2002. This ratio increased to 51.6 in 2007, suggesting an increase in the progressivity of the PIT between the two years. It is worth noting that the average tax rate does not move up in a straight line along the deciles. It was slightly lower for the third decile than it was for the second decile in 2002, but it was much lower for the second 16

We also compare the estimated distributive effects between the imputed tax and the reported tax to determine how the reported tax information will bias the estimates of the progressivity and redistributive effects of the personal income tax.

378

Xu Jing and Yue Ximing Table 10.6. The MT index and the P index

Gini coefficient of pretax income (G ) Gini coefficient of posttax income (G ∗ ) MT index (MT) Concentration ratio of tax (C) P (Kakwani) index

2002

2007

0.3212 0.3148 0.0064 0.6330 0.3117

0.3459 0.3322 0.0137 0.7574 0.4115

decile than for the lowest decile. This is mainly due to the fact that the tax burden differs based on the income source, and the composition of the income source varies across deciles. In 2007, for instance, operating income constituted 10.5 percent of the total income for the lowest decile, but 7.1 percent for the second-lowest decile.17 The same figures for property income are 0.41 percent and 0.31 percent, respectively. However, the tax burden is much higher for property income and operating income than it is for other income sources, for instance, wages and salary. Defining the tax burden of a certain income component as the proportion of the tax charged on that income component of the total tax, on average the tax burden in 2007 was 21.3 percent for property income, 18.5 percent for operating income, and 3.4 percent for wages and salary. Therefore, the reason why the average rate of the total tax for the lowest decile exceeded that for the second-lowest decile in 2007 is because for the lowest decile income components with high tax burdens constituted a larger proportion of total income. We now report the estimates of the MT index and the P index. The relevant figures are summarized in Table 10.6. Most of the results in this table are expected, given the pattern of the average tax rate across deciles in Table 10.5. The Gini coefficient decreases after taxes in both years and the MT index takes a positive sign, meaning that the tax is equalizing. A rise in the MT index during the period under examination, from 0.0064 in 2002 to 0.0137 in 2007, indicates an increasing redistributive effect of the individual income tax on the inequality of urban households. The P index takes a positive sign in both years, implying that the tax is progressive. This is not surprising given the rise in the average tax rate as we move up the income distribution. The increase in the P index between the two years, from 0.31 in 2002 to 0.41 in 2007, implies that the individual income tax in China became more progressive over this period. 17

Here the total income includes transfer income, which is exempt from tax.

Redistributive Impacts of the Personal Income Tax in Urban China

379

Table 10.7. Decomposition of the MT index into the effects of horizontal equity and vertical equity

MT index (MT) Horizontal equity (C d − G ∗ ) Vertical equity (= P ∗ t/(1 − t))

2002

2007

0.0064 −0.0002 0.0066

0.0137 −0.0002 0.0139

The unexpected message, perhaps the most important message, conveyed by Table 10.6 is that the equalizing effects of the individual income tax on inequality are so small that the after-tax income inequality shows little improvement compared with that of the before-tax income. It is clear from the MT index that the tax reduced the Gini coefficient of before-tax income by only 0.64 percentage points in 2002 and by only 1.37 percentage points in 2007. A decomposition of the MT index into the effects of horizontal equity and vertical equity provides some clues as to why the personal income tax makes only a minor contribution to inequality. The results of the decomposition are presented in Table 10.7. As shown, the measure of the horizontal equity is −0.0002 for both 2002 and 2007, meaning that the current personal income tax altered the ordering of individuals based on the before-tax income and violated the principle of horizontal equity, but not to a great extent. The violation of the principle of horizontal equity was mainly due to the different tax burdens of the different sources of income, as discussed earlier. Given the negative but small values of the measures of horizontal equity, the MT index is very close to a measure of vertical equity. In other words, the degree of the redistributive effect of the personal income tax in China mainly depends on the degree of vertical equity. The effect of vertical equity, as shown in the last section, consists of two parts: the progressivity of the tax, measured by the P index, and the average tax rate. With the P index given in Table 10.6 and the average tax rate given in Table 10.5, the effect of vertical equity, as shown in Table 10.7, was 0.0066 in 2002 and 0.0139 in 2007, respectively. Although raising the progressivity of the tax may increase the redistributive effects of the tax, the small impacts of the current personal income tax on inequality were mainly due to the low tax rate. This point becomes more evident from international comparison. Wagstaff et al. (1999) evaluated the redistributive effects of the personal income tax for twelve Organisation for Economic Co-operation and

380

Xu Jing and Yue Ximing

Development (OECD) countries on a comparable basis, using a similar methodology to that used in our study.18 According to their estimates, the concentration ratio of the personal income tax among the twelve OECD countries takes a minimum value of 0.3895, and a maximum value of 0.6628, with a simple mean of 0.5251. These figures are significantly lower than those in China, where the concentration ratio is 0.7574 in 2007 (Table 10.6). The P index, defined above as the concentration ratio of taxes minus the Gini coefficient of before-tax income, shows a similar pattern. For the twelve OECD countries, it takes a minimum value of 0.0891, a maximum value of 0.2717, and a simple average of 0.1963, but in China in 2007 the P index has very high value of 0.4115. A comparison of the tax rates, however, shows the opposite pattern. The effective tax rate is very high in the OECD countries, with a minimum of 6.2 percent, a maximum of 32.7 percent, and a simple mean of 16.61 percent, but it is very low in China, at 3.27 percent in 2007. Although a comparison between China and other developing countries is difficult due to lack of data, there is still some scattered evidence available for this purpose. Bird and Zolt (2005), for instance, provide information on two ratios: the ratio of revenues from the personal income tax to GDP and the ratio of revenues from the personal income tax to fiscal revenues of the central government. These two ratios in 2005 are, respectively, 0.3 percent and 1.7 percent for China, 0.4 percent and 2.6 percent for Vietnam, 1.4 percent and 16.1 percent for India, 2.1 percent and 17.2 percent for the Philippines, and 2.7 percent and 14.7 percent for Malaysia. Thus, China’s personal income tax is relatively low compared with those in other Asian countries (the ratios for Korea and Japan are even higher). One last task is a sensitivity analysis. In calculating the tax liability based on the preceding individual income tax schedule, we know that assumptions of how many times an individual receives payments of certain income components can affect the tax liability for every individual who has income from such sources. Estimations of the redistributive effects of these taxes are based on the assumption that each of the income components was received once per month. In order to look at the sensitivity of the estimated redistributive effects of the tax to this assumption, we assume that individuals who have these income sources receive each of these income components once per year, with the exception of two of these income 18

The year under investigation in Wagstaff et al. (1999) varies by country, ranging from 1987 to 1993.

Redistributive Impacts of the Personal Income Tax in Urban China

381

sources, that is, income from property rentals and income from interest and dividends.19 The results of these reestimations show that the once-per-year assumption increases the average tax rate but reduces the progressivity of the tax.20 The MT index eventually became smaller, from 0.0064 to 0.0061 in 2002, and from 0.0137 to 0.0133 in 2007. The changes in the estimated MT index can be considered small, suggesting that the assumption regarding the number of times an individual receives an income component for which there are taxes does not greatly change the redistributive effects.

V. Conclusion The purpose of this chapter is to evaluate the redistributive effects of the individual income tax. Because of the substantial understatement of the PIT in the CHIP urban data set, we apply the official tax schedule to the reported components of income to impute the tax liability for individuals in our sample. Using this imputed tax liability, which measures the taxes individuals would pay if their taxes were paid according to the regulations, we calculate the MT index, the most commonly used measure of the redistributive effects of taxes and governmental subsidies, and we decompose the MT index into the effects of horizontal equity and vertical equity. The MT index and its decomposition reveal that the personal income tax does reduce inequality, but the effect is negligible. The low average personal income tax rate is the main reason why the personal income tax fails to contribute more to improving inequality. We note that our results are based on the assumption that individuals pay taxes according to the official income tax rates. In fact, actual tax payments are probably lower than those implied by the regulations. Also, the discrepancy between actual tax payments and the amounts owed according to the regulations is probably greater among higher-income individuals than among lower-income individuals. For these reasons, the redistributive impact of the personal income tax may be even weaker than what is implied by our estimates. 19 20

We assume once per month for the former and once per year for the latter. The once-per-year assumption reduces the deductions, thus increasing the tax liability, compared with the once-per-month assumption. This, combined with the fact that the income components on which the taxes are levied on a per-receipt basis constitute a larger proportion of the total income for low-income households than for high-income households, reduces the progressivity of the tax.

382

Xu Jing and Yue Ximing

APPENDIX Table 10A.1. Main elements of the personal income tax in China: Categories of income subject to the personal income tax by category, the time basis for the tax levied, deductions, and the tax schedule

Category of taxable income

Basis on which the tax is levied

Wage and salary income

Monthly

RMB 1,600

Production and business income

Annual

Costs

Income from contracted or leased operations of enterprises/institutions Income from individual services

Annual

RMB 1,600×12 months

Per receipt

20% of receipts or RMB 800*

Remuneration from manuscripts Royalties

Per receipt

20% of receipts or RMB 800* 20% of receipts or RMB 800* No deductions 20% of receipts or RMB 800*; taxes, levies, and repair costs incurred Original value of the property; reasonable expenses No deductions No deductions

Per receipt

Interest and dividends Rental income from property leasing

Per receipt Per receipt

Income from sale of properties

Per receipt

Contingency income Other income that the MOF specifies as taxable

Per receipt

Deduction

Note: RMB = Renminbi (Chinese currency); MOF = Ministry of Finance. * Whichever is higher.

Tax rate 0–500: 1%; 501–2,000: 10%; 2,001–5,000: 15%; 5,001–20,000: 20%; 20,001–40,000: 25%; 40,001–60,000: 30%; 60,001–80,000: 35%; 80,001–100,000: 40%; 100,001– : 45% 0–5,000: 5%; 5,001–10,000: 10%; 10,001–30,000: 20%; 30,001–50,000: 30%; 50,001–: 35% 0–5,000: 5%; 5,001–10,000: 10%; 10,001–30,000: 20%; 30,001–50,000: 30%; 50,001– : 35% 0–20,000: 20%; 20,001–50,000: 30%; 50,001– : 40% 20% 20% 20% 20%

20%

20% 20%

Redistributive Impacts of the Personal Income Tax in Urban China

383

References Atkinson, A.B. (1980), “Horizontal Equity and the Distribution of the Tax Burden,” in H.J. Aaron and M J. Boskin, eds., The Economics of Taxation, 3–18, Washington, DC: Brookings Institution. Bird, R.M. and E.M. Zolt (2005), “The Limited Role of the Personal Income Tax in Developing Countries,” Journal of Asian Economics, 16(6), 928–946. China Master Tax Guide (2007), 5th ed., Hong Kong: CCH Hong Kong Ltd. Feldstein, M. (1976), “On the Theory of Tax Reform,” Journal of Public Economics, 6(1–2), 77–104. Kakwani, N.C. (1977), “Measurement of Tax Progressivity: An International Comparison,” Economic Journal, 87(345), 71–80. Kakwani, N. (1980), Income Inequality and Poverty: Methods of Estimation and Policy Applications, New York: Oxford University Press for the World Bank. Kakwani, N. (1984), “On the Measurement of Tax Progressivity and Redistribution Effect of Taxes with Applications to Horizontal and Vertical Equity,” in R.L. Basmann and G.F. Rhodes, Jr., eds., Advances in Econometrics: Economic Inequality Measurement and Policy, 149–168, New York: JAI Press. Musgrave, R.A. and T. Thin (1948), “Income Tax Progression 1929–48,” Journal of Political Economy, 56(6), 498–514. Plotnick, R. (1981), “A Measure of Horizontal Equity,” Review of Economics and Statistics, 63(2), 283–288. Rosen, H.S. (1978), “An Approach to the Study of Income, Utility, and Horizontal Equity,” Quarterly Journal of Economics, 92(2), 307–322. Wagstaff, A. et al. (1999), “Redistributive Effect, Progressivity and Differential Tax Treatment: Personal Income Taxes in Twelve OECD Countries,” Journal of Public Economics, 72(1), 73–98.

ELEVEN

Changes in the Gender-Wage Gap in Urban China, 1995–2007 Li Shi and Song Jin

I. Introduction China has been transitioning from a planned economy to a market economy for three decades. Since the mid-1990s when the urban reforms were implemented in earnest, wage inequality has widened (Appleton et al. 2002; Knight and Song 2008). During the period of the planned economy, one of the objectives of the Chinese government that was supported ideologically by Mao was to narrow the gender-wage gap. As a result, urban China boasted a smaller wage gap compared to other countries (Gustafsson and Li 2000). The economic transition has had an effect on the gender-wage gap through development of the private sector and the granting of more autonomy to state-owned enterprises to hire and fire employees and to determine wages. Given the pre–labor-market discrimination against women in terms of educational attainment, female workers were more concentrated in occupations requiring unskilled workers with low human capital. With the great flow of rural migrant workers, particularly unskilled female workers, into the cities and female workers facing much more severe competition in the urban labor market, wages have been depressed and the gender-wage gap has increased. This chapter investigates changes in the gender-wage gap since the mid1990s using the household survey data collected for 1995, 2002, and 2007. It should be noted that there were two shocks to the labor market in urban China during this period. One shock was the economic restructuring of The first draft of this chapter was presented at the Workshop on Income Inequality in China, May 21–22, 2010, at Beijing Normal University. The authors are grateful for comments from the participants at the workshop and for constructive comments and suggestions from Terry Sicular.

384

Changes in the Gender-Wage Gap in Urban China, 1995–2007

385

urban enterprises, during which time a majority of the state-owned and collective enterprises were privatized or restructured. As a consequence, millions of urban workers in the state-owned and collective sectors were laid off. The number of employees in state-owned enterprises (SOEs) decreased from 112.6 million in 1995 to 71.6 million in 2002, and this number declined further to 64.2 million in 2007. At the same time, the number of employees in urban collective enterprises decreased from 31.5 million in 1995 to 8.1 million in 2002, and 7.2 million in 2007 (National Bureau of Statistics [NBS] 2008). The shock of being laid off affected male and female employment in different ways. At the end of 2002, compared with male laborers, female laborers had higher unemployment rates. Furthermore, there was a higher proportion of early retirees among female laborers (Li and Gustafsson 2008). This shock no doubt had different effects on male and female wage growth in urban China. The second shock to the urban labor market was a significant increase in rural migrant workers moving to the cities because of changes in government policy with respect to rural migration. In the 1990s the government discouraged rural migration, but since 2000 obstacles to rural migration have been gradually reduced. Of course, many institutional barriers associated with the household registration (hukou) system remain and make it difficult for rural migrants to settle down in urban areas. Nevertheless, with the change in migration policy, the number of rural-urban migrant workers increased significantly, from an estimated less than 80 million in 2001 to 132 million in 2006 (Li and Luo 2007). Moreover, more urban jobs and occupations opened up to rural migrant workers. But, as a result, competition increased between local workers and migrant workers in the urban labor market. This competition also affected male and female laborers differently. Because, on average, migrant workers are less educated and less skilled, they are more likely to compete for unskilled jobs, which are usually held by local female workers. We hypothesize that local female workers face more competition from migrant workers than do their male counterparts. Both the privatization of state-owned enterprises and the rapid growth of the private sector have contributed to the rising wage gap between male and female workers. Dong et al. (2004) provide evidence that in township and village enterprises in Shandong and Jiangsu provinces, female workers have not received a return for their work experience and that they are discriminated against in terms of wage determination. Due to the reform and restructuring of the labor market in urban China, the average wage of urban workers has steadily increased since the

386

Li Shi and Song Jin

mid-1990s. The nominal annual wage of urban workers increased from 5,500 yuan in 1995 to 25,000 yuan in 2007, with an 11.4 percent per annum real growth rate (NBS 2008). Because the National Bureau of Statistics (NBS) does not provide information on average wages by gender, we do not know from the official data whether there has been a gender difference in the wage growth rates. However, our survey data show that wages increased by 10.4 percent per year for male workers and 9.2 percent per year for female workers. It should be noted that at the beginning of the twenty-first century, the Chinese government implemented a number of laws and regulations in an attempt to safeguard the legitimate rights and interests of women in the labor market, to promote gender equality, and to enable women to play an active role in society. These laws and regulations, however, have been only loosely implemented. The Law on the Protection of Women’s Rights and Interests was revised in 2005 in order to further guarantee gender equality. The revised law provides that gender should not constitute a pretext for refusing to hire a woman. Meanwhile, gender equality is stipulated in the National Program for Women’s Development (2001–2010). To a certain degree, these new regulations have reduced the extent of discrimination against women, but their impact on the gender-wage gap remains unclear. This chapter examines whether gender-wage differences of urban local workers continued to widen during the period under study. Rural-urban migrants are not covered in our analysis. We utilize the Blinder/Oaxaca decomposition methodology to decompose the wage differences between male and female workers into explained and unexplained components (Blinder 1973; Oaxaca 1973). The decomposition analysis is based on a general wage function and a quantile regression analysis. The results from the decomposition in 1995, 2002, and 2007 indicate that the gender-wage gap increased significantly, particularly from 2002 to 2007, and that a growing part of this gap was due to unexplained components, thereby implying rising discrimination against female workers. The decomposition results based on a quantile regression analysis indicate that the gender-wage gap was greater for the low-income groups and the share of unexplained components in the gap was also greater for the low-income groups. The chapter is structured as follows: the next section summarizes the main findings in previous studies. Section III describes the data and the differences in the basic statistics on the personal characteristics and employment structure for men and women. Section IV discusses the methodology and interprets the results of the decomposition analysis. The final section presents our conclusions.

Changes in the Gender-Wage Gap in Urban China, 1995–2007

387

II. Literature Review During the period of the planned economy, the gender-wage gap was not an important issue in urban China and therefore there was relatively little research on the topic. This situation continued through the early stage of the economic transition in the 1980s. In the international context, the observed wage gap between male and female workers at the time was relatively small. For instance, Gustafsson and Li (2000) found that in 1988 the wage of female workers was 84 percent that of their male counterparts. This contrasts with 82 percent in Sweden and 78 percent in Canada in the early 1990s. Compared to other explanatory variables, such as the ownership structure and the economic sector and location, the gender-wage gap among urban workers was considered less important (Knight and Song 1993). However, along with the reform of public enterprises and the development of the private sector, the gender-wage gap began to widen in favor of male workers. Between 1988 and 1995 the gap increased by 2 percent (Gustafsson and Li 2000). After the acceleration of the urban economic reforms in the mid-1990s, both the employment system and the system of remodeling social security underwent major changes. As a result, many urban workers were laid off and entered the ranks of the unemployed, or were laid off but kept their ties with the work unit (xiagang) (Appleton et al. 2002; Knight and Li 2006; Li and Hong 2004). A major impact of the increasing number of unemployed/xiagang workers in urban China was a decline in the female participation rate in the labor market. By the turn of the century, many female workers had left the labor market to become housewives. The data from the 1995 and 2002 surveys indicate that the participation rate of urban females between the ages of sixteen and sixty fell from 76 percent in 1995 to 67 percent in 2002, whereas the participation rate for males during the same period fell from 86 percent to 82 percent.1 The female participation rate in the labor market in 2007 remained unchanged from 2002, but during the same period, the male participation rate increased by 1.4 percentage points (see Table 11.1). In particular, there was a significant decrease in labormarket participation among less-educated female workers. As an example, the participation rate for females with a junior middle-school education decreased from 78.34 percent in 1995 to 61.90 percent in 2002, and it decreased further to 51.90 percent in 2007. 1

Housewives as a percentage of urban female adults ages sixteen through sixty increased from 2.9 percent in 1995 to 4.3 percent in 2002 and more or less remained at that level (4.2 percent) in 2007.

388

Li Shi and Song Jin Table 11.1. Labor-force participation and unemployment 1995 Total

Male

2002 Female

Total

Male

2007 Female

Total

Male

Female

Proportion of those between the ages of 16 and 60 71.6 70.58 72.56 74.61 73.91

75.28

74.77 73.55

75.95

Labor-force participation (%) 80.93 86.28

74.32 81.89

67.06

74.97 83.21

67.17

10.58

8.32

13.22

66.46 75.08 7.86 6.82 9.65 9.76 12.33 7.01 2.3 0.24 0.92 0.69 95.32 95.1

Unemployment rate (%) 3.26

2.93

75.8 3.62

Composition of the labor force (%) Employed 78.29 83.75 73.05 Unemployed 2.64 2.53 2.75 Students 7.69 7.9 7.49 Retired 8.94 5.02 12.7 Housewives 1.55 0.11 2.94 Others 0.89 0.7 1.07 Types of employment (%) Wage 98.42 98.37 employment Self-employment 1.58 1.63 and others

98.48 1.52

4.68

4.9

5.23

9.42

58.19 8.87 9.55 17.43 4.29 1.13

69.61 78.86 5.36 4.35 9.7 10.36 11.97 5.7 2.27 0.21 1.01 0.44

60.84 6.33 9.08 17.91 4.23 1.54

95.59

93.37 93.08

93.72

4.41

7.15

6.63

6.92

6.28

Source: CHIP urban household data, 1995, 2002, and 2007.

In light of the remarkable changes in the labor market and the reforms in urban enterprises that resulted in a rising gender wage-earnings gap in urban China, several studies have examined the magnitude of the genderearnings (wage) gap in the 1980s and 1990s and its changes over time (Gustafsson and Li 2000; Kidd and Meng 2001; Liu, Meng, and Zhang 2000; Maurer-Fazio and Hughes 2002; Meng 1998; Meng and Miller 1995; Rozelle et al. 2002). Entering the new millennium, the Chinese government attempted to implement a new development strategy that stressed balanced development between the urban and rural areas to reduce regional disparities and to narrow income/wage inequalities. The rising gender-wage gap, however, was not a policy priority. During the last decade, there has been a notable widening in the gender gap (D´emurger, Fournier, and Chen 2007). Li and Gustafsson (2008) indicate that between 1995 and 2002, as more women were laid off, the gender-income gap increased significantly. Chi and Li (2008) note that the average earnings gap between male and female workers increased considerably from 1996 to 2004.

Changes in the Gender-Wage Gap in Urban China, 1995–2007

389

Studies on the gender-wage gap in China apply decomposition analysis to the Chinese data. Wang and Cai (2008) found that a large part of the genderwage gap was due to discrimination. These findings are based on survey data from five large cities in 2001. Chi and Li (2008), using 1987, 1996, and 2004 data, show that the gender-earnings differential in urban China increased across the earnings distribution and this increase was greater in the lower quantiles. Work by Zhang et al. (2008) supports these findings. Most studies that investigate the gender-wage/earnings gap during the pre-2000 period utilize data from relatively few cities. The few studies using data collected after 2000 indicate a growing gender-wage/earnings gap in urban China, with discrimination playing an increasing role in this gap. However, many questions about the gender-wage/earnings gap remain unanswered.

III. Research Questions and Data Description We focus on the following research questions. Against the backdrop of economic transition and changes in the government development strategy, how has the gender-wage gap changed in urban China during the period of our study? Has the gap become wider or narrower over time? If the former, what were the driving forces behind such a change? Did rising discrimination against women play a role? If so, which group of female workers faced the most discrimination? Do female workers at the low end of the wage scale suffer more discrimination than do others? To answer these questions, we compare the gender-wage gap in urban China during two periods, 1995–2002 and 2002–2007. We also decompose the gap into two parts, explained and unexplained, for the three survey years and for changes in the gap between the two periods. The data come from three series of the urban household income survey conducted by the China Household Income Project (CHIP) in 1995, 2002, and 2007. To be comparable, the household samples in the three surveys were selected from the same provinces. The common provinces covered by the surveys are Beijing, Shanxi, Liaoning, Jiangsu, Anhui, Guangdong, Henan, Hubei, Sichuan, Yunnan, and Gansu. These provinces were initially selected to represent five regions (Li and Luo 2007), that is, Beijing to represent the provincial-level metropolitan cities;2 Jiangsu and Guangdong to represent 2

In the CHIP analyses, Chongqing is treated as a western city rather than as a provincial-level city. The reason is that Chongqing is qualitatively different from the three provincial-level cities (Tianjin, Beijing, and Shanghai) in terms of its level of development and urbanization. It is more similar to a western province rather than to the three other provincial-level cities

390

Li Shi and Song Jin

the eastern region; Liaoning to represent the northeastern region; Shanxi, Anhui, Henan, and Hubei to represent the central region; Sichuan and Yunnan to represent the southwestern region; and Gansu to represent the northwestern region.3 The observations in our analysis are individual urban local workers (employees); thus, we exclude the unemployed and self-employed, rural migrant workers, and owners of private enterprises. The wage variable is defined to include the wage and salary, bonuses, and cash subsidies received by workers. The wages in 2002 and 2007 are deflated using 1995 prices. Table 11.1 provides basic information about labor-market participation and employment and unemployment during the three survey years. As indicated in the table, the share of males and females between the ages of sixteen and sixty differed by about 3 percentage points during the first period, but remained relatively stable during the second period.4 There was also a rising supply of urban laborers between the first and second periods. In response, the participation rate in the urban labor force decreased considerably, resulting in a rising unemployment rate. For all laborers, the participation rate dropped by 6.6 percentage points. Although the participation of males decreased by less than 5 percentage points, there was a sharp fall of over 7 percentage points for females. Female participation went up slightly in 2007 compared to 2002. Due to the policy of laying off workers and the rise in the supply of labor, the real unemployment rate of urban workers increased dramatically, from 3 percent in 1995 to 8 percent in 2002 for male workers and from 4 percent to 13 percent for female workers, as shown in Table 11.1. The

3

4

in terms of the proportions of the urban and rural populations and the level of economic development. Also, Chongqing was not established as a province separate from Sichuan until 1997, so in earlier rounds of the CHIP survey Chongqing is included in the Sichuan sample. We did not use weights to adjust for any bias in our data. There are two reasons. First, we could weight the 2002 and 2007 data based on the regional distribution of urban workers in the 2000 census data and the 2005 mini census data, as other chapters in this volume. But there are no reference data for weighting the samples in the 1995 survey. Therefore, we prefer to keep the same provinces for all three surveys. Second, we compared the results with respect to provincial weights for the 2002 and 2007 data and found that after introducing a province dummy into the wage equations, the regression results did not change significantly. Consequently, we believe that our findings are quite robust. Because the official retirement age for the majority of female workers is fifty or fifty-five, the percentage of female workers in the age groups over fifty is comparatively lower (see Table 11A.1). Female workers over age fifty remaining economically active (not retired) are officials or professionals who have relatively higher wage income. Therefore, the earlier retirement regulation for female workers is more likely to lead to an underestimation of the gender-wage gap. In other words, the gender-wage gap would be larger if female workers retired at the same age as male workers.

Changes in the Gender-Wage Gap in Urban China, 1995–2007

391

unemployment rate declined from 2002 to 2007, but the rate was still higher in 2007 compared to 1995.5 Beginning in the mid-1990s, gender differences in unemployment became more significant. For example, the unemployment rate was 59 percent higher for female workers than for male workers in 2002 but 80 percent higher in 2007. Table 11.1 also provides information about the composition of the wage earners and the self-employed. Along with the development of the nonpublic sector and the privatization and restructuring of public enterprises, the number of self-employed increased during the period of our analysis. As indicated by the data, the proportion of self-employed among all employees increased over time, from less than 2 percent in 1995, to 4.7 percent in 2002, and further to 6.6 percent in 2007. It should be noted that the data may underestimate the share of self-employed due to a sampling bias. Table 11.2 shows the monthly wages of male and female workers and the gender-wage ratios for all the observations and for the various worker groups in 1995, 2002, and 2007. On average, female wages as a percentage of male wages decreased over time, from 84 percent in 1995 to 82 percent in 2002 and then dropped further to 74 percent in 2007. The data thus indicate a sharper drop in the relative wages of female workers during the second period than during the first period. To a certain extent, this drop can be attributed to competition with rural migrant workers, whose number increased rapidly during the second period. Examining the gender-wage ratio by type of worker, we find that it increased among almost all groups. Comparing the gender-wage gap in 2007 to that in 1995, for instance, among the nine age groups, there was a decline in the ratio of female to male wages in seven of the groups (see Table 11.2); among twelve sectors, ten sectors showed a decline in the ratio, and among the eleven provinces, there was a decline in nine provinces. Because we are interested in changes in the relationship between educational attainment and the gender-wage gap, we also provide the gender-wage gap for worker groups with different levels of education. The gender-wage ratios (females/males) decreased for all education groups during the period under study. However, during the first period we find a sharper decline among those with a lower level of education. For example, between 1995 and 2002 the gender-wage ratio declined by 13 percentage points for those who had less than a primary school education, but only by 1 percentage point for those with four years of college. The opposite occurred between 2002 and 2007. The corresponding figures are 1 and 11 percentage points. Thus, 5

Our unemployment rates for the three years are much higher than the official rates. The official rates are 2.9 percent, 4 percent, and 4 percent in 1995, 2002, and 2007 respectively (NBS 1996, 2003, 2008).

392

Li Shi and Song Jin

Table 11.2. Wage structure and the gender-wage gap in urban China, 1995, 2002, and 2007 1995 Male Real monthly wage Age group 16–20 21–25 26–30 31–35 36–40 41–45 46–50 51–55 56–60

2002

Female F/M

Male

Female

2007 F/M

Male

Female

F/M

520.12 436.0

0.84

953.73

266.39 360.24 402.89 478.06 516.08 563.47 585.93 624.11 661.89

252.83 317.57 382.53 410.82 470.37 483.59 487.68 503.77 442.11

0.95 0.88 0.95 0.86 0.91 0.86 0.83 0.81 0.67

327.98 655.01 796.87 889.15 962.11 1,006.65 1,009.4 1,035.93 1,068.27

Minority groups Han 521.85 437.53 Minority 480.7 402.18

0.84 0.84

954.46 956.4

776.37 0.81 879.5 0.92

1,712.86 1,262.6 0.74 1,491.58 1,179.24 0.79

Marital status Married Single Others

0.83 0.89 1.05

985.31 706.53 720.09

797.97 0.81 634.78 0.9 807.94 1.12

1,767.03 1,268.85 0.72 1,232.47 1,180.11 0.96 972.42 1,275.98 1.31

376.83

0.8

720.3

481.63 0.67

394.3 416.4 480.4

0.81 0.84 0.91

768.75 869.48 918.73

561.51 0.73 727.36 0.84 843.63 0.92

1,169.87 776.87 0.66 1,445.03 1,047 0.72 1,536.21 1,179.12 0.77

514.17 557.87

0.93 0.88

1,088.85 938.33 0.86 1,337.57 1,163.47 0.87

1,914.41 1,461.52 0.76 2,512.74 1,904.54 0.76

530.06 456.7

0.86

1,018.53

879.9

0.86

1,887.58 1,456.06 0.77

421.55 335.35 679.96 642.16

0.8 0.94

681.36 1,230.82

550.97 0.81 930.11 0.76

1,365.12 1,183.36 0.87 1,701.03 1,264.13 0.74

525.03 520.57

0.99

668.33

461.21 0.69

1,137.77

827.65 0.73

532.91 447.81

0.84

853.6

637.57 0.75

902.5

676.21 0.75

496.17 440.93 619.31 585.83

0.89 0.95

975.95 1237.25

861.41 0.88 1175.85 0.95

542.57 449.71 360.93 322.85 426 448.4

Educational attainment Primary and 469.46 less Junior middle 486.28 Senior middle 494.11 Professional 526.5 school 2-year college 552.63 4-year college 631.3 Ownership State-owned sector Collective Joint-venture foreign firm Private selfemployed Other Occupation Office worker Office manager

779.33 0.82

1,705.84 1,259.59 0.74

340.95 616.73 663.12 785.41 794.99 810.82 859.37 866.86 619.37

723.29 1,106.41 1,501.02 1,752.67 1,849.38 1,815.88 1,753.83 1,696 1,628.9

1.04 0.94 0.83 0.88 0.83 0.81 0.85 0.84 0.58

989.8

494.32 1,076.06 1,283.6 1,334.44 1,260.78 1,189.04 1,277.87 1,586.58 868.74

0.68 0.97 0.86 0.76 0.68 0.65 0.73 0.94 0.53

657.93 0.66

1,845.5 1,382.54 0.75 2,380.25 1,693.08 0.71

Changes in the Gender-Wage Gap in Urban China, 1995–2007

1995 Male Professional technician Manual worker Other Industry Manufacturing Agriculture Mining Construction Transportation and communication Commerce and trade Public utilities Finance and insurance Education and culture Health and social welfare Scientific research and technology Government social organizations Province Beijing Shanxi Liaoning Jiangsu Anhui Henan Hubei Guangdong Sichuan Yunnan Gansu

2002

Female F/M

Male

Female

393

2007 F/M

Male

Female

F/M

526.63 466.65

0.89

961.08

907.69 0.94

2,091.32 1,711.63 0.82

414.14 356.52 443.42 376.52

0.86 0.85

666.94 677.46

520.11 0.78 462.11 0.68

1,270.73 1,239.69

491.78 543.37 533.54 561.65 581.34

403.54 432.14 443.64 418.13 463.53

0.82 0.8 0.83 0.74 0.8

808.58 898.59 701.72 945.35 977.43

680.58 820.32 580.4 810.26 864.42

0.84 0.91 0.83 0.86 0.88

1,460.33 1,400.95 1,708.53 1,670.13 1,622.09

1,092.62 1,342.19 1,090.42 1,152.55 1,290.74

0.75 0.96 0.64 0.69 0.80

477.05 387.84

0.81

730.76

612.27 0.84

1,281.06

918.1

0.72

560.78 428.74 572.2 521.78

0.76 0.91

931.78 1,172.92

658.06 0.71 894.74 0.76

1,486.74 1,041.53 0.70 2,006.57 1,632.8 0.81

582.12 496.28

0.85

1,227.12

963.29 0.78

2,013.93 1,673.35 0.83

564.47 520.61

0.92

1,142.75 1,025.38 0.9

1,787.62 1,487.99 0.83

619.34 519.54

0.84

1,298.38 1,308.8

1.01

2,321.07 1,566.73 0.68

514.47 491.19

0.95

1,102.25

0.87

2,166.65 1,510.34 0.7

662.94 432.13 478.47 566.76 429.2 400.95 479.26 921.55 480.93 485.65 392.23

0.87 0.74 0.82 0.85 0.78 0.8 0.9 0.85 0.85 0.88 0.82

1,329.38 1,031.53 0.78 806.14 651.47 0.81 887.18 643.5 0.73 1,011.31 798.93 0.79 826.13 619.13 0.75 756.23 575.22 0.76 814.96 674.47 0.83 1,564.54 1,332.93 0.85 804.65 681.84 0.85 911.68 807.28 0.89 805.04 627.83 0.78

574.34 320.8 393.83 480.92 335.63 321.34 430.86 781.49 408.4 426.4 320.41

Source: CHIP urban household data, 1995, 2002, and 2007.

954.5

2,341.99 1,394.83 1,310.72 2,174.11 1,427.15 1,227.6 1,617.76 2,649.59 1,352.98 1,186.28 1,175.41

899.47 0.71 779.64 0.63

1,888.47 1,063.91 833.61 1,544.7 953.75 999.79 1,076.59 1,822.7 1,142.82 1,045.16 828.53

0.81 0.76 0.64 0.71 0.67 0.81 0.67 0.69 0.84 0.88 0.7

394

Li Shi and Song Jin

assuming that manual workers have less education than do professionals, the gender-wage ratio declined more among manual workers from 1995 to 2002 and more among professionals from 2002 to 2007. Moreover, on average, among those with more education in state-owned firms as opposed to private firms, the drop in the gender-wage ratio was more significant in private firms, as also indicated in Table 11.2.

IV. Methodology We use the method originally proposed by Blinder (1973) and Oaxaca (1973) to decompose the gender-wage gap in each survey year. The decomposition method is presented as follows, given average wage/earnings for males as Ym,t and for females as Yf,t . Ym,t − Yf,t = βm,t X m,t − βf,t X f,t ,

(1)

where X m,t and X f,t is the vector of the explanatory variables adopted in the income function and βm,t and βf,t is the vector of the estimates of the coefficients of the explanatory variables. The subscripts m and f here refer to the male and female groups respectively. The difference (Ym,t − Yf,t ) can be decomposed into two components; that is, Ym,t − Yf,t = βm,t (X m,t − X f,t ) + X f,t (βm,t − βf,t ).

(2)

In Equation (2), the component βm,t (Xm,t − Xf,t ), attributed to differences in personal and employment characteristics between males and females, is usually interpreted as explained, whereas the component Xf,t (βm,t − βf,t ), attributed to differences in the coefficients of wage/earnings functions between males and females, is commonly regarded as unexplained, that is, attributable to discrimination.6 To decompose changes in the gender-wage/earnings gap over time, the following formula can be used: Yt+1 − Yt = βm,t ( X t+1 − X t ) + X t+1 (βm,t+1 − βm,t ) + X f,t ( βt+1 − βt ) + βt + 1(X f,t+1 − X f,t ), (3) where Yt+1 and Yt refer to the gender income gaps in time t+1 and time t, and the gender gaps for the characteristics and coefficients in times t 6

The gender-wage gap can also be decomposed as Ym,t − Yf,t = βf,t (Xm,t − Xf,t ) + Xm,t (βm,t − βf,t ). Here we adopt Equation (2) instead of the alternative formula. The reason is that we suppose that females would be treated as males if there were no discrimination in the labor market.

Changes in the Gender-Wage Gap in Urban China, 1995–2007

395

and t + 1 are given by X t = X m,t − X f,t , X t+1 = X m,t+1 − X f,t+1 , βt = βm,t − βf,t , and βt+1 = βm,t+1 − βf,t+1 .

(4)

It is clear that Equation (3) has four parts. The first part βm,t ( Xt+1 − Xt ) picks up the change in the gap due to changes in endowments of the gender differences between two time points. The second part, Xt+1 (βm,t+1 − βm,t ), picks up the changes in the gap due to changes in the coefficients of the male wages/earnings functions between two time points. The third part, Xf,t ( βt+1 − βt ), captures the changes in the gap due to changes in the gender differences of the coefficients of the wages/earnings functions between two time points. The fourth part, βt+1 (Xf,t+1 − Xf,t ), captures the changes in the gap due to changes in the endowments of females between two time points. The weakness of this method is that it is not able to capture the effects of differences in the wage distribution between males and females on the gender-income gap. In a segmented labor market, like that in China, it is more likely that female workers with low levels of education, who are unskilled and who are employed in the informal sectors face, more serious discrimination. To address this issue, we decompose the gender-wage gap based on a quantile regression analysis, decomposing the gender gap at each quantile point.

V. Results and Interpretations A. Changes in the Gender-Wage Gap over Time To determine whether the gender-wage gap was due purely to gender, we first conducted a regression analysis in which gender was treated as a dummy variable and with a number of control variables, such as marital status, education attainment, ownership of the work unit, region, and so on. Because occupation and industry segregation between males and females has been discussed extensively in recent years, we needed to consider whether the occupation and industry variables should be included in the wage equations. As discussed in other studies, the inclusion of occupation and industry may underestimate the gender-wage gap if occupational and industrial segregation exist between male and female workers.7 Therefore, we provide 7

See Table 11A.1 for the sample distribution among different occupations and industries. From the table, we see that male workers have a higher probability of being employed

396

Li Shi and Song Jin

the results for the regression analyses both controlling for occupation and industry and not. Table 11.3 summarizes the regression results for 1995, 2002, and 2007 (see Table 11A.2 for detailed regression results). It is obvious that the genderwage gap increased over time. As shown in the table, compared to the wages of female workers, the wages of male workers were 10.5 percent higher in 1995, 17.4 percent higher in 2002, and almost 29.7 percent higher in 2007.8 The results from the regression using the pooled data show that on average the wages of males were 18.5 percent higher than those of females during the 1995–2007 period. It is also apparent that the gender gap widened considerably during the second period. Moreover, excluding the occupation and industry variables increases the contribution of the male dummy variable to the gender-wage gap. This increase was especially significant in 2002. However, in 2007 the difference between the coefficients with and without controlling for the occupation and industry variables declined, which may indicate a reduction in segregation. As seen in Table 11.A3, the contribution of the age variable changed dramatically over time. Figures 11.1a and 11.1b illustrate the changes in the age profile of ln-wage for male and female workers during the three years. In 1995, the age-wage profile for female workers displays a pattern that is similar to that in most market economies. Wages increase with age, reach a peak between ages forty-one and forty-five, and decline thereafter until retirement age. However, the age-wage profile for male workers displays wages rising increasingly with age and an even sharper increase with age in the years leading up to retirement. This is why the age variable contributed positively to the gender-wage gap in 1995. Empirically, age is commonly regarded as indicative of work experience. The age-wage profile of female workers became more similar to that of male workers in 2002 (see Figure 11.1b), implying there was a convergence of returns to work experience from 1995 to 2002. Moreover, the return-to-work experience increases more significantly for female workers compared to male workers during the 2002–2007 period (see Figure 11.1c). This is the major reason for the declining role of the age variable in contributing to the genderwage gap.

8

as “office workers,” “officials or managers,” or “professionals or technicians,” positions that are better paid, whereas more female workers are manual workers. The segregation in terms of occupation seems to be worse in 1995 and 2002 but is more likely to be improved in 2007. However, the disproportional distribution among industries for male and female workers clearly did not change during the period under investigation. Male wages as a higher percentage than female wages are computed using the following formula: P = exp (C) − 1, where P = percentage, and C = coefficients.

Table 11.3. Regression analysis on the gender-wage gap in urban China1 1995 2

Ln-wage

IO

Male

0.10 (10.15)**

4.47 (98.39)** 0.37

2002

NIO

2007

Pooled sample

IO

NIO

IO

NIO

IO

NIO

0.12 (12.84)**

0.16 (12.03)**

0.20 (15.16)**

0.26 (20.87)**

0.27 (22.18)**

4.52 (104.95)**

4.80 (54.38)** 0.35

4.85 (54.14)**

5.14 (51.88)** 0.43

5.13 (51.44)**

0.17 (24.36)** 0.47 (53.06)** −0.52 (60.37)** 4.93 (125.03)** 0.54

0.19 (28.16)** 0.42 (49.03)** −0.52 (61.71)** 4.96 (128.94)**

2007 397

2002 Constant Observations R-squared

Notes: 1. The table is derived from the wage functions with a number of control variables. Detailed results can be found in the Appendix Table 11A.2. 2. “IO” and “NIO” refer to the regressions with and without controlling for the industry and occupation dummies respectively. Source: CHIP urban household data, 1995, 2002, and 2007.

398

Li Shi and Song Jin

6.6

7

6.4

6.8 6.6

6.2

6.4

Ln-wage

Ln-wage

6 5.8 5.6

6.2 6 5.8

Male 2002

5.6

Female 2002

Male 1995

5.4

Female 1995

5.2

5.4

5

5.2 16-20 21-25 26-30 31-35 36-40 41-45 46-50 51-55 56-60

Age

16-20 21-25 26-30 31-35 36-40 41-45 46-50 51-55 56-60

Age

(a) 1995

(b) 2002

7 6.8 6.6

Ln-wage

6.4 6.2 6 5.8 5.6

Male 2007

5.4

Female 2007

5.2 16-20 21-25 26-30 31-35 36-40 41-45 46-50 51-55 56-60

Age

(c) 2007

Figure 11.1. Wage-Age Profile for Male and Female Workers, 1995, 2002, and 2007. Note: The figures are drawn based on the predicted ln-wages for an “average” man and “average” woman, that is, with the most common characteristics for each of the variables except age.

Another important variable contributing to the rising gender-wage gap is the ownership status of the employer. Figure 11.2 shows the gender-wage gap among workers employed in sectors of different ownership during the three years. In 1995 female workers were paid relatively better in state-owned, foreign-invested, joint-venture, and other ownership sectors compared to their male counterparts. Their relative advantage gradually declined thereafter. With the exception of the urban collective sector, male workers in other sectors of ownership, such as state-owned enterprises (SOEs), experienced more rapid wage growth than did their female counterparts, especially in 2007. As a result, gender differences based on employment sector increased over time, contributing to the rising share of the gender-wage gap due to discrimination against female workers.

Changes in the Gender-Wage Gap in Urban China, 1995–2007

399

6.4 7 Male 1995

6.3

Male 2002 6.8

Female 1995

Female 2002

6.1

6.6

6

6.4

Ln-wage

Ln-wage

6.2

5.9

6.2

5.8 6 5.7 5.8

5.6 5.5

5.6 other collective state-owned joint-venture private or selfenterprise or foreign firm employed ownership sector

(a) 1995

collective state-owned joint-venture private or self- other sector enterprise or foreign firm employed ownership

(b) 2002

7.2 7

Male 2007 Female 2007

6.8

Ln-wage

6.6 6.4 6.2 6 5.8 5.6 collective state-owned joint-venture private or self- other sector enterprise or foreign firm employed ownership

(c) 2007

Figure 11.2. Ln-Wage Levels for Male and Female Workers by Ownership Sector, 1995, 2002, and 2007. Note: The figures are drawn based on the predicted ln-wages for an “average” man and “average” woman, that is, with the most common characteristics for each of the variables except the ownership of their work unit.

B. Results from the Decomposition Analysis To examine which personal and employment characteristics are important determinants of the gender-wage gap, we carried out estimations of the wage equations for male and female workers separately. The results from the regression analysis can be found in Table 11A.3. With the results from the regression analysis and following the approach developed by Blinder (1973) and Oaxaca (1973), we decompose the gender-wage gap into two components, part due to gender differences in endowments and part due to gender differences in the coefficients. Table 11.4 provides the decomposition results for 1995, 2002, and 2007. The results indicate that the raw gender-wage gap increased from 1995 to 2002 and even more sharply from 2002 to 2007. It is clear that the increase in the gender-wage gap was not due to greater differences in endowments since the contribution of the endowments declined over time. As shown in

400

Li Shi and Song Jin Table 11.4. Oaxaca’s decomposition analysis for the gender-wage gap, 1995, 2002, and 2007

Raw differential (R) (E+C) due to endowments (E) due to coefficients (C) Endowments as % total (E/R) Discrimination as % total (D/R)

1995

2002

2007

19.5 9.4 10.1 48 52

24.1 7.5 16.7 31 69

32 7.1 24.9 22.3 77.7

Note: The numbers in the first three rows indicate by what percentage the wages of male workers are higher than are those of female workers. Source: CHIP urban household data, 1995, 2002, and 2007.

Table 11.4, the share of the wage gap explained by differences in endowments decreased from 48 percent to 31 percent between 1995 and 2002, and further to 22 percent in 2007. Meanwhile, the share of the wage gap due to differences in the coefficients increased from 52 percent in 1995 to 69 percent in 2002 and to 78 percent in 2007. Clearly, the importance of the unexplained portion of the gender-wage gap increased markedly. If the gender-wage gap due to differences in the coefficients is interpreted to be the outcome of discrimination, Table 11.4 clearly shows that the part of the gender-wage gap explained by discrimination as a percentage of the total gap increased significantly over time.

C. Results from the Quantile Regression Analysis A large number of rural migrant workers flowing into cities, particularly during the second period under study, would increase competition for jobs in the urban labor market, making it difficult for local workers to find jobs and leading to a decline in the wages of urban workers with the same endowments as the migrant workers. However, it is not clear that the shock of rural migration had the same impact on male workers as it did on female workers. To determine gender differences in terms of the impact of rural migration on the urban labor market, we conducted a quantile regression analysis using the data from the three years. Table 11.5 presents the results from a quantile regression analysis for 1995, 2002, and 2007. It is apparent that the raw wage gap between male and female workers rose for all wage groups from 1995 to 2002 and even further to 2007. At the same time, the gap became larger for the lower wage groups compared to the higher wage groups over the three years, as shown in

Changes in the Gender-Wage Gap in Urban China, 1995–2007

401

Table 11.5. Decomposition results from the quantile regression analysis Quantile 1995 Raw differential (R) (E+C) due to endowments (E) due to coefficients (C) Endowments as % total (E/R): Discrimination as % total (D/R):

10th

25th

50th

75th

90th

22.4 12.4 10.1 55.4 45.1

18.7 9.8 8.9 52.4 47.6

17.5 8.7 8.7 49.7 49.7

16.7 8.3 8.4 49.7 50.3

17.2 7.7 9.5 44.8 55.2

2002 Raw differential (R) (E+C) due to endowments (E) due to coefficients (C) Endowments as % total (E/R) Discrimination as % total (D/R)

29.2 9.8 19.4 33.6 66.4

26.6 8.5 18.1 32.0 68.0

23.4 8.0 15.5 34.2 66.2

22 7.4 14.6 33.6 66.4

18.1 6.6 11.4 36.5 63.5

2007 Raw differential (R) (E+C+U) due to endowments (E) due to coefficients (C) Endowments as % total (E/R) Discrimination as % total (D/R)

35.2 7.4 27.9 20.7 79.3

34 6.7 27.3 19.7 80.3

33 7.2 25.8 21.8 78.2

31 8.3 22.7 26.8 73.2

28.2 7.6 20.5 27.3 72.7

Note: The numbers in the first five rows for each year panel are the percentages of which the wages of male workers are higher than those of female workers. Source: CHIP urban household data, 1995, 2002, and 2007.

Figure 11.3a. The gender-wage gap due to discrimination (or the unexplained part) as a percentage of the raw gap is larger for the lowest wage group than it is for the highest wage group in all three years. Moreover, the discrimination share increases continuously over time, especially for the 40

90

35

80 70

25

Percent

Percent

30 20 15 1995 2002 2007

10 5 0

10th

25th

50th

75th

Quantile (a) Raw Difference

90th

60 50 40 30 20

1995 2002 2007

10 0

10th

25th

50th

75th

90th

Quantile (b) Share of Discrimination

Figure 11.3. Gender-Wage Differential Resulting from the Quantile Analysis, 1995, 2002, and 2007. Note: The figures are drawn with the data from Table 11.5.

402

Li Shi and Song Jin

lower income group, as shown in Figure 11.3b. Therefore, the results indicate that discrimination against female workers was even more significant among lower-wage earners and that it grew over time. It is obvious that, on the one hand, the share of discrimination augmented over time for all quantile groups, and that, on the other, the share increased more rapidly among the lower-wage groups than among the higher-wage groups.

D. Results of the Dynamic Analysis To compare the changes in the gender-wage gap between the 1995–2002 period and the 2002–2007 period, we used Equation (3) in Section III to decompose the changes in the gender-wage gap during each period into four parts. The results in Table 11.6 present the relative contributions of each part to the changes in the gender-wage gap. Table 11.6 indicates that the change in the gender-wage gap during 1995– 2002 can be attributed primarily to the explanatory variables, such as occupation, ownership, and industry. It is clear that the changes in the endowments of females (part [4]) and the changes in the gender differences of the coefficients of the wage/earnings functions (part [3]) are the two largest shares in explaining the changes in the gender gap during 1995–2002. Comparing the second period with the first period, the changes in the gender differences of the coefficients of the wage/earnings functions (part [3]) became the most important contributor. This implies that the gender differences in the coefficients of the wage functions became greater during the second period than they were during the first period.

V. Conclusion The wage gap between males and females has widened since the mid-1990s. Most previous studies, focusing on the 1990s, provide empirical findings either at one point in time or for a short period. This chapter presents an updated analysis of the changes in the gender-wage gap in urban China since the mid-1990s, using data from the 1995, 2002, and 2007 CHIP surveys. We find that the gender-wage gap increased significantly, particularly during the 2002–2007 period, and this increase was largely due to unexplained components, thereby implying that discrimination against female workers in the urban labor market in China was rising. To determine whether females at the low end of the labor market faced more discrimination, we conducted a decomposition analysis based on a quantile regression analysis. The results indicate that the gender-wage gap is greater for the low-wage groups and that the share of the unexplained

Table 11.6. Decomposition results for changes in the gender-wage gap

403

Total Yt+1 − Y

(1) βm,t ( X t+1 − X t )

(2) X t+1 (βm,t+1 − βm,t )

(3) X f,t ( βt+1 − βt )

(4) βt+1 (X f,t+1 − X f,t )

0.05 100.00

−0.01 −20.26

0.00 −9.75

0.03 60.28

0.03 69.72

0.07 100.00

−0.05 −63.47

0.04 52.30

0.08 104.11

0.01 7.07

1995–2002 % 2002–2007 %

Source: CHIP urban household data, 1995, 2002, and 2007.

404

Li Shi and Song Jin

components in the gap is also greater for the low-wage groups. It is hypothesized that, with the inflow of rural migrant workers, unskilled and lesseducated urban workers have faced greater competitive pressures in the labor market. But this competition has different implications for male and female workers; wage growth is more significantly depressed for female workers than it is for male workers, particularly for female workers who are unskilled and less educated.

APPENDIX Table 11A.1. Proportion of sample in urban China, 1995, 2002, and 2007 1995 Male Sample as a percentage of total Age group 16–20 21–25 26–30 31–35 36–40 41–45 46–50 51–55 56–60

100

2002

Female 100

Male 100

2007

Female 100

Male 100

Female 100

1.97 8.53 9.49 14.07 17.97 18.78 13.44 9.51 6.25

2.63 9.68 11.95 16.22 21.01 21.66 11.03 4.52 1.29

0.65 5.13 6.97 13.05 18.48 16.32 20.67 13.62 5.11

0.94 7.59 9.47 17.19 21.89 19.79 16.78 5.42 0.94

0.48 5.23 7.99 10.68 16.58 20.79 15.52 16.17 6.54

0.45 7.38 11.04 15.83 21.42 24.03 14.04 5.47 0.32

Minority group Han Minority

95.78 4.22

95.72 4.28

96.07 3.93

95.89 4.11

96.83 3.17

96.39 3.61

Marital status Married Single Others

86.9 12.51 0.60

87.65 11.12 1.24

88.99 10.12 0.89

86.68 10.73 2.59

88.85 10.54 0.61

86.73 10.36 2.91

Educational attainment Primary and less Junior middle school Senior middle school Professional school 2–3-year college 4-year college

5.21 32.54 18.89 15.25 17.92 10.20

7.26 37.04 21.79 16.35 12.53 5.04

3.48 26.80 22.85 10.43 23.50 12.94

2.95 25.11 25.52 14.82 23.61 7.98

2.03 19.45 24.27 10.49 25.63 18.14

1.68 14.86 27.26 12.93 29.10 14.18

86.88 10.97 1.26

78.93 18.39 1.17

73.09 5.71 2.40

66.78 9.13 2.12

63.06 5.22 16.54

55.32 7.63 16.02

Ownership State-owned sector Collective sector Joint-venture or foreign firm

Changes in the Gender-Wage Gap in Urban China, 1995–2007

1995

Private or self-employed Other ownership Occupation Office worker Official or manager Professional or technician Manual worker Others Industry Manufacturing Agriculture Mining Construction Transportation and communication Commerce and trade Public utilities Finance and insurance Education and culture Health and social welfare Scientific research and technology Government and social organizations Province Beijing Shanxi Liaoning Jiangsu Anhui Henan Hubei Guangdong Sichuan Yunnan Gansu

2002

405

2007

Male

Female

Male

Female

Male

Female

0.59

0.81

8.67

8.87

11.02

12.63

0.31

0.70

10.13

13.09

4.16

8.41

19.61 17.04 47.91

22.82 5.83 41.64

18.88 16.30 46.72

24.78 5.10 37.64

32.98 7.42 20.06

37.54 3.20 19.61

12.30 3.14

22.42 7.29

16.53 1.57

30.11 2.37

35.13 4.41

34.56 5.09

43.26 2.02 1.20 3.33 5.92

40.85 1.29 0.94 2.56 4.14

27.73 1.43 2.35 4.24 10.25

25.09 1.26 1.06 2.42 5.44

23.12 1.13 1.85 3.73 11.44

15.67 0.73 0.70 1.93 4.50

12.31

17.21

7.82

12.64

8.39

15.29

3.34 1.84

4.49 2.21

13.26 2.68

18.59 3.35

15.45 2.81

20.60 3.63

6.59

8.69

8.90

10.55

8.14

10.79

3.61

5.77

4.15

7.22

3.12

5.93

2.64

2.13

2.27

1.60

5.50

3.75

13.93

9.74

14.90

10.78

15.33

16.47

7.30 9.62 10.59 11.14 6.80 8.40 10.81 8.28 12.12 9.32 5.62

7.07 9.17 10.36 11.10 7.26 7.99 10.66 8.23 12.71 9.95 5.51

8.10 9.41 12.20 10.44 7.31 9.41 10.87 8.87 8.71 8.63 6.04

8.83 7.97 10.63 10.02 6.43 9.85 11.34 10.35 8.58 10.12 5.87

11.92 8.55 10.41 7.30 8.16 9.07 5.77 14.93 7.85 7.60 8.44

11.00 7.47 9.61 7.02 8.25 9.79 5.54 16.29 8.20 8.13 8.70

Table 11A.2. Wage functions in urban China, 1995, 2002, and 2007 (results of linear regression) Ln-wage Male Age group 21–25 26–30 31–35

406

36–40 41–45 46–50 51–55 56–60 Marital status Married Others

1995

2002

2007

Pooled sample

0.1 (10.15)**

0.16 (12.03)**

0.26 (20.87)**

0.17 (24.36)**

0.12 (12.84)**

0.2 (15.16)**

0.27 (22.18)**

0.19 (28.16)**

0.26 (7.16)** 0.42 (10.23)** 0.54 (12.66)** 0.64 (15.25)** 0.7 (16.58)** 0.69 (15.92)** 0.71 (15.80)** 0.78 (15.97)**

0.44 (5.75)** 0.52 (6.67)** 0.63 (7.78)** 0.71 (8.74)** 0.75 (9.22)** 0.76 (9.38)** 0.7 (8.51)** 0.58 (6.66)**

0.33 (3.84)** 0.56 (6.41)** 0.61 (6.88)** 0.64 (7.23)** 0.65 (7.28)** 0.66 (7.36)** 0.7 (7.74)** 0.58 (6.17)**

0.3 (9.09)** 0.46 (13.26)** 0.55 (15.57)** 0.63 (17.81)** 0.67 (18.70)** 0.66 (18.48)** 0.66 (18.08)** 0.6 (15.36)**

0.29 (8.03)** 0.44 (10.89)** 0.56 (13.70)** 0.68 (16.60)** 0.75 (18.20)** 0.74 (17.71)** 0.76 (17.43)** 0.85 (17.90)**

0.46 (5.83)** 0.55 (6.89)** 0.67 (8.12)** 0.76 (9.19)** 0.8 (9.69)** 0.84 (10.09)** 0.79 (9.33)** 0.69 (7.77)**

0.34 (3.84)** 0.57 (6.49)** 0.62 (6.88)** 0.67 (7.39)** 0.67 (7.47)** 0.68 (7.52)** 0.73 (7.96)** 0.6 (6.38)**

0.32 (9.92)** 0.49 (14.34)** 0.59 (16.82)** 0.68 (19.47)** 0.72 (20.52)** 0.72 (20.45)** 0.73 (20.05)** 0.68 (17.51)**

0.13 (5.29)** 0.06 −1.15

0.15 (4.38)** 0.08 −1.29

0.18 (6.17)** 0.15 (2.83)**

0.12 (7.26)** 0.05 −1.67

0.13 (5.50)** 0.07 −1.25

0.15 (4.25)** 0.08 −1.24

0.17 (5.95)** 0.16 (3.01)**

0.12 (7.14)** 0.06 −1.78

Minority status Minority Education Junior middle school Senior middle school Professional school 2–3-year college 4-year college

407

Ownership State-owned enterprise Joint-venture or foreign firm Private or self-employed Other ownership Occupation Office worker Official or manager

−0.07 (2.99)**

0.07 (2.10)*

−0.04 −1.34

−0.01 −0.62

−0.07 (2.96)**

0.07 (2.16)*

0.12 (5.88)** 0.18 (7.69)** 0.24 (10.26)** 0.27 (10.84)** 0.35 (12.72)**

−0.04 −1.33

−0.01 −0.77

0.04 −0.98 0.17 (4.33)** 0.24 (5.97)** 0.34 (8.56)** 0.47 (10.86)**

0.16 (3.56)** 0.26 (5.96)** 0.39 (8.41)** 0.52 (11.53)** 0.7 (14.86)**

0.09 (4.83)** 0.16 (8.70)** 0.26 (13.04)** 0.34 (17.73)** 0.5 (23.91)**

0.14 (6.76)** 0.22 (9.78)** 0.32 (13.86)** 0.35 (15.16)** 0.45 (17.34)**

0.07 −1.87 0.23 (5.98)** 0.38 (9.21)** 0.51 (12.90)** 0.68 (16.18)**

0.17 (3.87)** 0.29 (6.62)** 0.45 (9.84)** 0.62 (13.81)** 0.83 (18.03)**

0.11 (6.10)** 0.2 (11.14)** 0.34 (17.91)** 0.46 (24.39)** 0.65 (32.29)**

0.25 (17.77)** 0.45 (9.94)** 0.35 (3.20)** 0.2 (2.57)*

0.25 (9.56)** 0.45 (9.31)** −0.11 (3.45)** 0.1 (3.37)**

0.24 (9.73)** 0.13 (4.85)** −0.04 −1.23 −0.3 (8.97)**

0.24 (20.23)** 0.2 (11.68)** −0.1 (5.63)** −0.1 (5.80)**

0.28 (20.06)** 0.45 (10.30)** 0.31 (3.05)** 0.12 −1.65

0.32 (12.42)** 0.45 (9.04)** −0.17 (5.30)** 0.08 (2.46)*

0.28 (11.42)** 0.12 (4.37)** −0.08 (2.95)** −0.36 (10.88)**

0.28 (23.96)** 0.19 (11.00)** −0.15 (8.98)** −0.14 (8.10)**

0.11 (6.72)** 0.23 (11.33)**

0.23 (10.70)** 0.37 (13.72)**

0.08 (4.74)** 0.23 (7.99)**

0.14 (13.82)** 0.27 (18.92)** (continued)

Table 11A.2 (continued) Ln-wage Professional or technician Others Industry Agriculture Mining Construction

408

Transportation and communication Commerce and trade Public utilities Finance and insurance Education and culture Health and social welfare Scientific research and technology Government and social organizations Province Beijing Shanxi

1995

2002

2007

0.16 (11.05)** 0.06 (2.36)*

0.27 (14.45)** −0.15 (3.11)**

0.19 (9.96)** −0.07 (2.26)*

0.2 (20.69)** −0.02 −0.89

−0.01 −0.28 0.09 (2.06)* 0.02 −0.69 0.1 (4.59)** −0.05 (3.58)** −0.07 (2.88)** 0.24 (7.19)** 0.05 (2.52)* 0.06 (2.69)** 0.13 (4.16)** 0 −0.12

0.05 −0.91 0.07 −1.35 −0.01 −0.16 0.15 (5.92)** −0.02 −0.88 0.03 −1.25 0.15 (3.72)** 0.18 (7.07)** 0.23 (7.67)** 0.21 (4.47)** 0.11 (4.63)**

0.09 −1.45 0.33 (6.45)** −0.01 −0.17 0.08 (3.45)** −0.07 (3.08)** −0.01 −0.63 0.16 (4.48)** 0.05 (2.12)* 0.02 −0.49 0.15 (4.94)** 0.07 (3.01)**

0 −0.16 0.14 (4.80)** 0 −0.13 0.1 (7.55)** −0.04 (3.79)** −0.05 (4.01)** 0.16 (7.52)** 0.08 (6.03)** 0.09 (5.47)** 0.16 (7.94)** 0.04 (3.41)**

0.48 (17.65)** 0.05 −1.9

0.46 (13.49)** −0.05 −1.53

0.69 (26.66)** 0.21 (7.54)**

0.58 (34.87)** 0.09 (5.22)**

0.45 (17.14)** 0.02 −0.62

Pooled sample

0.45 (12.81)** −0.04 −1.18

0.71 (27.12)** 0.23 (8.08)**

0.57 (34.29)** 0.08 (4.76)**

Liaoning Jiangsu Anhui Henan Hubei Guangdong

409

Sichuan Yunnan

0.18 (7.16)** 0.43 (17.49)** 0.12 (4.38)** 0.03 −0.98 0.21 (8.47)** 0.84 (32.42)** 0.19 (7.83)** 0.25 (9.90)**

0.22 (6.73)** 0.28 (8.70)** 0.04 −1.21 −0.03 −0.93 0.05 −1.53 0.72 (21.64)** 0.13 (3.90)** 0.17 (5.00)**

0.17 (6.28)** 0.63 (21.85)** 0.24 (8.77)** 0.07 (2.47)* 0.36 (11.76)** 0.81 (33.23)** 0.25 (9.03)** 0.18 (6.29)**

4.47 (98.39)** 10,777 0.37

4.8 (54.38)** 8,657 0.35

5.14 (51.88)** 9,979 0.43

Year 2007 Dummy Year 2002 Dummy Constant Observations R-squared

0.2 (12.48)** 0.46 (27.74)** 0.16 (9.08)** 0.04 (2.46)* 0.22 (13.22)** 0.8 (50.45)** 0.21 (12.60)** 0.22 (13.16)** 0.47 (53.06)** −0.52 (60.37)** 4.93 (125.03)** 29,413 0.54

0.15 (6.19)** 0.4 (16.70)** 0.09 (3.29)** −0.02 −0.66 0.18 (7.29)** 0.81 (31.99)** 0.17 (7.21)** 0.23 (9.34)**

0.19 (5.81)** 0.29 (8.70)** 0.05 −1.47 −0.02 −0.5 0.06 −1.75 0.72 (21.24)** 0.12 (3.49)** 0.2 (5.74)**

0.18 (6.72)** 0.64 (21.77)** 0.25 (8.91)** 0.08 (2.87)** 0.35 (11.28)** 0.82 (33.52)** 0.27 (9.55)** 0.2 (6.80)**

4.52 (104.95)** 11,358 0.35

4.85 (54.14)** 8,719 0.31

5.13 (51.44)** 9,980 0.41

0.18 (11.45)** 0.45 (27.44)** 0.15 (8.66)** 0.03 (2.02)* 0.2 (12.33)** 0.8 (50.21)** 0.2 (12.52)** 0.23 (13.46)** 0.42 (49.03)** −0.52 (61.71)** 4.96 (128.94)** 30,057 0.53

Notes: Standard errors in parentheses. Base group: female, age group 16–20, Single, Han, primary school or less, collective sector, manual worker, manufacturing, Gansu. **p < 0.01 and *p < 0.05.

410

Li Shi and Song Jin Table 11A.3. Wage functions in urban China, 1995, 2002, and 2007 1995

Variables Age group 21–25 26–30 31–35 36–40 41–45 46–50 51–55 56–60 Marital status Married Others Minority status Minority Education Junior middle school Senior middle school Professional school 2–3-year college 4-year college Ownership State-owned enterprise Joint-venture or foreign firm Private or self-employed Other ownership

2002

2007

Male

Female

Male

Female

Male

Female

0.39 (7.73)** 0.48 (8.61)** 0.6 (10.40)** 0.69 (11.83)** 0.77 (13.28)** 0.78 (13.29)** 0.83 (13.81)** 0.89 (14.22)**

0.16 (3.08)** 0.39 (6.43)** 0.5 (8.11)** 0.63 (10.25)** 0.66 (10.67)** 0.62 (9.70)** 0.55 (7.93)** 0.43 (3.88)**

0.42 (3.87)** 0.56 (5.18)** 0.62 (5.56)** 0.72 (6.37)** 0.77 (6.86)** 0.78 (6.92)** 0.75 (6.62)** 0.65 (5.56)**

0.46 (4.26)** 0.52 (4.50)** 0.66 (5.56)** 0.72 (6.13)** 0.75 (6.37)** 0.77 (6.50)** 0.61 (4.94)** 0.25 −1.63

0.17 −1.56 0.45 (4.01)** 0.53 (4.62)** 0.57 (4.92)** 0.59 (5.09)** 0.57 (4.93)** 0.58 (4.99)** 0.5 (4.27)**

0.55 (4.03)** 0.76 (5.48)** 0.8 (5.72)** 0.83 (5.95)** 0.81 (5.82)** 0.85 (6.05)** 0.95 (6.62)** 0.25 −1.2

0.18 (5.58)** −0.03 −0.4

0.08 (2.11)* 0.1 −1.28

0.19 (4.31)** 0.02 −0.25

0.1 −1.75 0.09 −1.08

0.2 (4.91)** −0.04 −0.37

0.11 (2.63)** 0.15 (2.25)*

−0.05 −1.56

−0.09 (2.48)*

−0.02 −0.4

0.16 (3.07)**

−0.05 −1.26

−0.02 −0.52

0.09 (3.20)** 0.15 (4.76)** 0.18 (5.51)** 0.21 (6.37)** 0.3 (8.45)**

0.15 (4.79)** 0.19 (5.84)** 0.31 (8.75)** 0.34 (9.09)** 0.43 (9.55)**

0.01 −0.25 0.13 (2.64)** 0.17 (3.33)** 0.26 (5.27)** 0.41 (7.85)**

0.09 −1.45 0.25 (3.95)** 0.35 (5.30)** 0.46 (7.07)** 0.57 (7.87)**

0.14 (2.56)* 0.24 (4.37)** 0.34 (5.81)** 0.48 (8.42)** 0.65 (10.98)**

0.16 (2.16)* 0.27 (3.77)** 0.41 (5.53)** 0.53 (7.34)** 0.72 (9.41)**

0.23 (11.28)** 0.4 (6.63)** 0.4 (2.61)** 0.08 −0.54

0.25 (12.70)** 0.49 (7.25)** 0.26 −1.7 0.26 (2.61)**

0.24 (6.63)** 0.5 (7.98)** −0.1 (2.35)* 0.16 (3.75)**

0.25 (6.88)** 0.39 (5.10)** −0.12 (2.58)** 0.05 −1.12

0.31 (8.82)** 0.18 (4.85)** 0.01 −0.16 −0.19 (3.88)**

0.15 (4.43)** 0.07 −1.89 −0.08 (1.97)* −0.36 (7.92)**

Changes in the Gender-Wage Gap in Urban China, 1995–2007

1995 Variables Occupation Office worker Official or manager Professional or technician Others Industry Agriculture Mining Construction Transportation and communication Commerce and trade Public utilities Finance and insurance Education and culture Health and social welfare Scientific research and technology Government and social organizations Province Beijing Shanxi Liaoning Jiangsu Anhui Henan Hubei

2002

411

2007

Male

Female

Male

Female

Male

Female

0.08 (3.32)** 0.19 (7.46)** 0.12 (5.82)** 0.05 −1.21

0.12 (5.11)** 0.25 (7.01)** 0.17 (8.32)** 0.06 (1.99)*

0.2 (6.55)** 0.33 (9.97)** 0.22 (8.45)** −0.11 −1.64

0.24 (7.81)** 0.4 (7.81)** 0.31 (10.74)** −0.16 (2.36)*

0.07 (3.39)** 0.22 (6.45)** 0.16 (6.53)** −0.05 −1.2

0.11 (4.20)** 0.24 (4.42)** 0.25 (8.01)** −0.07 −1.72

0.02 −0.35 0.09 −1.66 0.02 −0.49 0.09 (3.22)** −0.08 (4.11)** −0.01 −0.37 0.23 (4.96)** 0.03 −1.08 0.03 −0.8 0.13 (3.43)** −0.02 −0.91

−0.06 −0.9 0.09 −1.23 0.02 −0.54 0.12 (3.24)** −0.03 −1.26 −0.09 (2.46)* 0.28 (5.62)** 0.07 (2.62)** 0.1 (3.00)** 0.13 (2.47)* 0.04 −1.37

0.03 −0.46 0.14 (2.36)* 0.01 −0.23 0.15 (5.07)** −0.04 −1.07 0.11 (3.87)** 0.26 (4.87)** 0.26 (7.76)** 0.23 (5.20)** 0.17 (3.03)** 0.15 (5.09)**

0.07 −0.79 −0.12 −1.18 0.04 −0.6 0.16 (3.43)** −0.02 −0.45 −0.04 −1.37 0.04 −0.76 0.08 (2.07)* 0.22 (5.09)** 0.31 (3.82)** 0.09 (2.07)*

0.06 −0.81 0.33 (5.76)** 0.03 −0.78 0.06 (2.26)* −0.09 (3.08)** 0.01 −0.44 0.11 (2.22)* 0.02 −0.51 −0.03 −0.6 0.17 (4.74)** 0.1 (3.47)**

0.16 −1.57 0.25 (2.35)* −0.11 −1.73 0.11 (2.36)* −0.04 −1.2 −0.03 −1.11 0.2 (3.99)** 0.07 −1.81 0.03 −0.75 0.12 (2.37)* 0.03 −0.78

0.47 (13.34)** 0.09 (2.80)** 0.17 (5.32)** 0.38 (11.78)** 0.1 (3.02)** 0.04 −1.07 0.16 (4.95)**

0.48 (11.60)** 0 −0.09 0.18 (4.69)** 0.49 (12.98)** 0.13 (3.07)** 0.02 −0.46 0.26 (6.88)**

0.49 (11.11)** −0.06 −1.36 0.23 (5.59)** 0.27 (6.49)** 0.05 −1.22 −0.03 −0.77 0.03 −0.75

0.45 (8.48)** −0.03 −0.52 0.23 (4.41)** 0.31 (6.09)** 0.04 −0.72 −0.01 −0.29 0.08 −1.65

0.67 (19.83)** 0.22 (6.11)** 0.2 (5.72)** 0.63 (16.55)** 0.26 (7.16)** 0.02 −0.42 0.38 (9.47)**

0.72 (18.07)** 0.18 (4.25)** 0.12 (2.93)** 0.63 (14.20)** 0.22 (5.28)** 0.13 (3.15)** 0.34 (7.09)** (continued)

412

Li Shi and Song Jin Table 11A.3 (continued) 1995

Variables Guangdong Sichuan Yunnan Constant Observations R-squared

2002

2007

Male

Female

Male

Female

Male

Female

0.81 (24.05)** 0.18 (5.57)** 0.21 (6.43)** 4.56 (72.32)** 5,688 0.36

0.87 (21.70)** 0.21 (5.55)** 0.28 (7.37)** 4.49 (68.52)** 5,089 0.36

0.69 (16.07)** 0.08 −1.74 0.12 (2.62)** 4.98 (40.88)** 4,827 0.33

0.75 (14.66)** 0.23 (4.23)** 0.23 (4.39)** 4.74 (36.68)** 3,830 0.36

0.81 (25.22)** 0.17 (4.73)** 0.12 (3.23)** 5.46 (43.16)** 5,579 0.41

0.8 (21.80)** 0.35 (8.22)** 0.24 (5.61)** 5.04 (32.04)** 4,400 0.41

Notes: Standard errors in parentheses. Base group: age group 16–20, Single, Han, primary school or less, collective sector, manual worker, manufacturing, Gansu. ***p < 0.01, **p < 0.05, and *p < 0.1.

References Appleton, S., J. Knight, L. Song, and Q. Xia (2002), “Labor Retrenchment in China: Determinants and Consequences,” China Economic Review, 13(2–3), 252–275. Blinder, A. S. (1973), “Wage Discrimination: Reduced Form and Structural Estimates,” Journal of Human Resources, 8(4), 436–455. Chi, W. and B. Li (2008), “Glass Ceiling or Sticky Floor? Examining the Gender Earnings Differential across the Earnings Distribution in Urban China, 1987–2004,” Journal of Comparative Economics, 36(2), 243–263. D´emurger, S., M. Fournier, and Y. Chen (2007), “The Evolution of Gender Earnings Gaps and Discrimination in Urban China, 1988–95,” Developing Economics, 45(1), 97–121. Dong, X., F. MacPhail, P. Bowles, and S. P. Ho (2004), “Gender Segmentation at Work in China’s Privatized Rural Industry: Some Evidence from Shandong and Jiangsu,” World Development, 32(6), 979–998. Gustafsson, B. and S. Li (2000), “Economic Transformation and the Gender Earnings Gap in Urban China,” Journal of Population Economics, 13(2), 305–339. Kidd, M.P. and X. Meng (2001), “The Chinese State Enterprise Sector: Labour Market Reform and the Impact on Male-Female Wage Structure,” Asian Economic Journal, 15(4), 405–423. Knight, J. and S. Li (2006), “Unemployment Duration and Earnings of Re-employed Workers in Urban China,” China Economic Review, 17(2), 103–199. Knight, J. and L. Song (1993), “Why Urban Wages Differ in China,” in K. Griffin and R. Zhao, eds., The Distribution of Income in China, 221–242, Basingstoke: Macmillan. Knight, J. and L. Song (2008), “China’s Emerging Urban Wage Structure, 1995–2002,” in B.A. Gustafsson, S. Li, and T. Sicular, eds., Income Inequality and Public Policy in China, 221–242, New York: Cambridge University Press. Li, S. and B. Gustafsson (2008), “Unemployment, Earlier Retirement, and Changes in the Gender Income Gap in Urban China, 1995–2002,” in B.A. Gustafsson, S. Li, and

Changes in the Gender-Wage Gap in Urban China, 1995–2007

413

T. Sicular, eds., Income Inequality and Public Policy in China, 243–266, New York: Cambridge University Press. Li, S. and Z. Hong, eds. (2004), Jingji zhuanxing de daijia: Zhongguo chengshi shiye pinkun, shouru chaju de jingyan fenxi (Economic Policy Analysis: China’s Experience with Urban Unemployment, Poverty, and Income Inequality), Beijing: Zhongguo caizheng jingji chubanshe. Li, S. and C. Luo (2007), “Zhongguo chengxiang jumin shouru chaju de chongxin guji” (Re-estimating the Income Gap between Urban and Rural Households in China), Beijing daxue xue bao (Zhexue shehui kexue ban), no. 2, 111–120. Liu, P.W., X. Meng, and J. Zhang (2000), “Sectoral Gender Wage Differentials and Discrimination in the Transitional Chinese Economy,” Journal of Population Economics, 13(2), 331–352. Maurer-Fazio, M. and J. Hughes (2002), “The Effects of Market Liberalization on the Relative Earnings of Chinese Women,” Journal of Comparative Economics, 30(4), 709– 731. Meng, X. (1998), “Male-Female Wage Determination and Gender Wage Discrimination in China’s Rural Industrial Sector,” Labour Economics, 5(1), 67–89. Meng, X. and P. Miller (1995), “Occupational Segregation and Its Impact on Gender Wage Discrimination in China’s Rural Industrial Sector,” Oxford Economic Papers, 47(1), 136–155. National Bureau of Statistics (NBS) (various years), Zhongguo tongji nianjian (China Statistical Yearbook), Beijing: Zhongguo tongji chubanshe. Oaxaca, R. (1973), “Male-Female Wage Differentials in Urban Labor Markets,” International Economic Review, 14(13), 693–709. Rozelle, S., X. Dong, L. Zhang, and A. Mason (2002), “Gender Wage Gaps in Post-reform Rural China,” Pacific Economic Review, 7(1), 157–179. Wang, M. and F. Cai (2008), “Gender Earnings Differential in Urban China,” Review of Development Economics, 12(2), 442–454. Zhang, J., J. Han, P.W. Liu, and Y. Zhao (2008), “Trends in the Gender Earnings Differential in Urban China, 1988–2004,” Industrial & Labor Relations Review, 61(2), 224–243.

TWELVE

Intertemporal Changes in Ethnic Urban Earnings Disparities in China Ding Sai, Li Shi, and Samuel L. Myers, Jr.

I. Introduction The opening of the Chinese economy in 1978 by Deng Xiaoping ushered in an era of significant economic growth (Chow 1993). During the following thirty years, gross domestic production expanded, the manufacturing sector grew, and exports to the outside world skyrocketed. Much of this dramatic growth has been attributed to capital accumulation and productivity increases (Chow and Li 2002). Yet, the period also represented a significant shift in national policies toward growth tempered by attention to social equity (Friedman 2006). This shift under the Hu Jintao–Wen Jiabao leadership is commonly referred to as promotion of a “harmonious society.” Figure 12.1 shows the significant advances of the Chinese economy over the thirty-year period. Annual average gross domestic product (GDP) growth during the period from 1978 to 2007 was 9.74 percent. By way of comparison, annual GDP growth in the United States during the same period was only 3.3 percent (Myers forthcoming). One clear indicator of the slowing of the Chinese economy occurred during the 1992–1999 period. Figure 12.1 shows a growth rate of almost 15 percent in 1992, followed by a sharp decline in the ensuing years and only a little more than 7 percent in 1999. It is well known that one of the consequences of the overall pattern of sharp economic growth in the postreform era has been a widening of inequality between those at the top and those at the bottom of the income distribution, both overall and regionally (Cai, Wang, and Du 2002). Measures of overall income inequality, as well as of the spatial inequality of income uniformly, show sizable increases from the early reform years to Blanca Monter and Juan Cardenas provided valuable research assistance. Support from the U.S. Fulbright Foundation is gratefully acknowledged.

414

Intertemporal Changes in Ethnic Urban Earnings Disparities in China 415 25

20

Growth Rate

15

10

5

–5

2006

2004

2002

2000

1998

1996

1994

1992

1990

1988

1986

1984

1982

1980

1978

1976

1974

1972

1970

0

Year

Figure 12.1. Real Rate of GDP Growth: China. Source: World Bank, World Development Indicators, http://databank.worldbank.org/ddp/home.do?step=12&id=4&cno=2. Accessed October 2, 2011.

the present (Organisation for Economic Co-operation and Development [OECD] 2010: 140–141). Corrections for measures of imputed rents and public subsidies yield high, but stable, measures of inequality from 1995 to 2002 (Gustafsson, Li, and Sicular 2008). Inequality in disposable household income per capita, as measured by the Gini coefficient, widened in urban areas in China during the period of rapid economic growth from the 1980s to the early twenty-first century. In urban China, the Gini coefficient rose from 0.244 in 1988 to 0.339 in 1995 and 0.322 in 2002 (Gustafsson et al. 2008). Income inequality continued to grow from 2002 to 2007 in urban China but not as rapidly as it had grown prior to the Hu Jintao–Wen Jiabao era (see Chapter 7 in this volume). By the mid-2000s, the overall level of inequality indices placed China ahead of most European nations and the United States, similar to Mexico and Chile, and behind South Africa and Brazil (OECD 2010: 130). One lesser-known consequence of the economic policies leading to the expansion of the Chinese economy has been the narrowing of the earnings gap between the majority Han population and the ethnic minorities in the urban areas. Among rural households, the ratio of minority to Han household per capita income stagnated at 66.3 percent in 1988, 67.14 percent in 1995, and 65.73 percent in 2002. But among urban households, the ratio increased from 92 percent in 1988 to over 100 percent in 2002, leading some

416

Ding Sai, Li Shi, and Samuel L. Myers, Jr.

commentators to conclude that a Han-minority earnings gap no longer existed in urban China. This finding contrasts with the findings of a deterioration in the relative status of minorities in rural areas. Gustafsson and Li (2003), examining survey information from nineteen provinces in 1988 and 1995, find that the per capita income gap of 19.2 percent in 1988 had increased to 35.9 percent by 1995. Gustafsson and Li (2003) decompose the rural income gaps into portions that can be explained by human capital, spatial and political factors, and an unexplained portion. They report that most of the gap in rural incomes between the majority and minority populations can be explained by human capital and related factors. In those rural provinces where the gaps actually diminished, increased educational attainment among minorities stands out as a key explanatory factor. The Gustafsson and Li (2003) findings of widening gaps in rural household per capita incomes between 1988 and 1995 contrast with the narrowing gaps in rural per capita household incomes tentatively found in research by Luo and Sicular (2012) examining changes in per capita income between 1995 and 2002. The gap in 2002, however, is still wider than the gap in 1988. An important insight for understanding changes in the relative economic well-being of minorities in rural and/or urban areas is thus the timing of the changes. The mechanism by which government policies might have contributed to improvements in the relative economic status of minorities in urban areas, but not necessarily in rural areas, stems from an inherent selection effect. In addition to targeted affirmative action policies that provided assistance to minority-group members in admissions to college and exemptions from restrictions on childbearing, the Chinese government initiated investment protocols that boosted incomes in rural areas, which indirectly improved the well-being of minorities, who are largely concentrated in rural areas (Hannun 2002). The out-migration of minority rural workers to urban areas depressed the overall incomes of the remaining minorities in the rural areas and contributed to the widening of the Han-minority income gap observed by Gustaffson and Li (2003). But these policies arguably contributed to the migration of higher-educated minorities to urban areas, further contributing to the perception that there were no longer income disparities between Han and minorities in urban areas. The conventional wisdom is that there are now only small differences in household per capita incomes between Han and minorities in urban areas (Zang and Li 2001). Thus, in addition to a broad expansion of the Chinese economy during the entire thirty-year period, the Chinese government advanced policies

Intertemporal Changes in Ethnic Urban Earnings Disparities in China 417

to assist ethnic minorities that putatively resulted in reduced disparities between Han and minorities. This chapter details the factors that contributed to the historic narrowing of the minority-Han earnings gap during the period from 1995 to 2002, as rates of economic growth were falling slightly. It also explores the heretofore undocumented rise in ethnic disparities between 2002 and 2007 in urban China. An innovative contribution of this analysis is that it provides two different types of decompositions of the changes in income disparities: (a) intertemporal, within-group differences, and (b) intratemporal, betweengroup differences. The chapter is organized as follows: first, we provide background information about the nature of the changes in the conditions facing Han and minority workers during the past several decades. Then we provide an analytical framework for understanding wage and salary income disparities between Han and minorities, wherein we decompose the earnings gaps between periods within groups and between groups within periods. The approach is to construct a measure of minority versus Han wage and salary disparities and to decompose that measure into portions explained by differences in endowments and portions explained by differences in treatment, both between groups and between time periods. In a concluding section, we discuss the implications for the policies aimed at improving access to education for minorities and for the policies designed to promote minorities in state-owned enterprises.

II. Background Conventional wisdom states that the expansion of economic growth through the market reforms in China was accompanied by a widening of overall inequality in per capita incomes. Much of this widening inequality is attributed to rural-urban differences in access to infrastructure as well as the attendant implications of changes in educational attainment and the quality of education. Although literacy rates, attendance rates, and overall educational attainment improved, the gaps between rural and urban areas widened (Hannum 2002). Because ethnic minorities are concentrated in rural and underdeveloped regions of China, the gaps in educational outcomes are attributed to vocational differences (Hannum and Yu 1998; Rong and Shi 2001; Zhang and Kanbur 2005). National statistics show that poverty rates in autonomous ethnic areas are much higher than they are in the rest of rural China. From 2006 to 2009, the poverty rates in autonomous ethnic areas were 18.9, 18.6, 17, and 16.4 percent, respectively. In the same years,

418

Ding Sai, Li Shi, and Samuel L. Myers, Jr. 120 100 80 Mean Median

60 40 20 0 1995

2002 Year

2007

Figure 12.2. The Ratio of Minority-to-Han Mean and Median Family-Household Total Incomes in Urban China (1995, 2002, and 2007 CHIP data). Source: Authors’ computations using the 1995, 2002, and 2007 CHIP data.

the poverty rates in rural China were 6, 4.6, 4.2, and 3.6 percent, respectively (Central People’s Government 2011). Zang and Li (2001), using a small sample of Han and minorities in Beijing, find few demographic differences between Han and minorities, which can be attributed to the selective migration of higher-educated minorities to urban areas (Zang and Li 2001: 41). They also contend that the statesanctioned entitlements provided to ethnic minorities provided a source of upward mobility (Zang and Li 2001: 41). They find no statistically significant ethnic differences in total earnings, including bonuses, investment returns, and wages and salaries. However, they do find wide disparities in the returns to education and returns to state employment. Thus, they argue that minorities benefit more than do nonminorities from improved education and employment in state enterprises. Estimating a simple human capital model using data from 1989 and 1992, with no controls for rural-urban residence or for ethnic minority status, Maurer-Fazio (1999: 27) finds rates of return to education of about 3 to 4 percent, with higher rates for females than for males. This points to the possibility that changes in earnings disparities might be due to differences in returns to schooling between males and females. These stylized facts about Han-minority urban wage differentials conflict with other evidence about disparities in family household incomes, personal incomes, and wage and salary incomes drawn from national samples during different periods of economic growth in China. Figure 12.2 shows that during the period of a downward trend in economic growth, 1995, the ratio

Intertemporal Changes in Ethnic Urban Earnings Disparities in China 419 120.0 109.3 100.0

98.1 91.1 91.9 91.3

93.1

90.9 84.6

87.4

80.0 60.0 40.0 20.0 0.0 1995

2007

2002

Males

Females

Both

Figure 12.3. Ratio of Minority-to-Han Wage and Salary Incomes. Source: Authors’ computations using the 1995, 2002, and 2007 CHIP data.

of minority-to-Han mean and median incomes was lower than it was during the upturn in 2002. The growth rate remained stable thereafter with only minor declines in 2007, the point at which income ratios were again lower (see Figure 12.2). Because the evidence does not point to a constant pattern of income disparities, a more careful look at the underlying labor-market dynamics that might contribute to a narrowing and then a widening of the earnings gap is called for. Figure 12.3 details the ratio of the mean wage and salary incomes for ethnic-minority group members to the mean wage and salary incomes for Han in urban areas. For simplicity, we focus only on wage and salary incomes and not on bonuses, subsidies, or related benefits. The reason for this specific focus on wages and salaries is that other forms of compensation – such as bonuses, housing subsidies, and health care – vary widely from industry to industry and are more loosely related to worker productivity than are wages and salaries. The sample years are 1995, 2002, and 2007. The calculations are based on persons with positive wage and salary incomes who are ages eighteen and over, and they are restricted to persons with an urban household registration (hukou). The sampled provinces are common across the years presented. The ratios are presented for males, females, and both males and females.

420

Ding Sai, Li Shi, and Samuel L. Myers, Jr.

According to these unadjusted estimates, minorities earned less than did Han in 1995, 2002, and 2007. The ratio of minority-to-Han earnings, 91.1 for males in 1995, declined slightly to 90.87 in 2002. It dropped again to 84.5 in 2007. Thus, over the span of a decade, minority-Han earnings ratios declined for males. In 1995, the ratio for females was 91.91. The ratio rose to 109.32 in 2002, but then dropped to 93.06 in 2007. Thus, over the span of a decade, earnings of minority females improved relative to those of Han. Overall, combining males and females, the ratio of minority-to-Han wage and salary incomes rose from 91.29 in 1995 to 98.11 in 2002, a period of improvement propelled largely by the increase in the relative earnings of minority females. By 2007, however, the ratio had declined to 87.39, stemming from declines in the relative earnings of both males and females since 2002. The current chapter proposes to explain these stylized facts. One obvious potential explanation for the changing disparities in earnings between ethnic minorities and Han is differences in age patterns and/or educational attainment. These demographic changes, cast into a conventional human capital framework, can be seen as potential explanatory factors underlying the story conveyed in Figure 12.3. Another potential explanation is the changing treatment of Han versus minorities over the decade, following the logic advanced by Darity (1982). Statistically, this is measured by the differential returns to education, job opportunities, household structures, firm types, or provincial labor markets. The economic interpretation of these differential returns is that they can produce unequal treatment of otherwise identically situated workers. The task for the analysis that follows is to decompose the observed gaps in earnings into portions that can be explained by such factors as age, education, and job markets and into portions that are unexplained and thus can be attributed to differential returns.

III. Law and Policy Regarding Antidiscrimination and the Development of Ethnic Minorities in China Law and public policy providing protections against discrimination and preferences for ethnic minority members have evolved over the years.1 1

The People’s Republic of China currently officially recognizes the Han majority and fiftyfive different ethnic minorities (minzu, meaning ethnic group or nationality). Based on the Rules for Classifying the Nationality of Chinese Citizens, which were enacted in 1990, a person is classified as a minority based freely on the nationality of one of his or her parents. Minority status can be registered by a person’s parents before he or she is eighteen

Intertemporal Changes in Ethnic Urban Earnings Disparities in China 421

The evolution shows a subtle shift from protection of minorities against discrimination prior to the Hu Jintao–Wen Jiabao era to preferential treatment thereafter. The Constitution of the People’s Republic of China bans ethnic minority group discrimination.2 Other examples of antidiscrimination efforts include the 1951 ban against derogatory ethnic names for streets or towns and the 1997 criminalization of discrimination against minorities. More-recent initiatives have taken on the tone of ethnic preferences. These include extra points awarded on the national college admissions examination, which is the primary vehicle for admissions to college, the ability to take examinations in languages other than Mandarin, relaxed population control measures, and guarantees of political representation in the autonomous regions (Gustafsson and Ding 2009). Certain minorities and persons in rural areas are exempt from China’s 1979 one-child policy. Based on the Law of Population and Family Planning, the people’s congresses in the provinces and autonomous regions can enact their own specific rules. In general, a minority family in a rural minority area may have three children. There is no limit to the number of children in families in Tibet. In urban areas, however, the policy is much stricter. In some urban areas, such as Anhui or Shandong, if the husband and wife are both minorities, they may have two children. In other urban areas, such as Xinjiang and Qinghai, if only one member of the couple is a member of a minority, they may have two children. In still other urban areas, such as Guangxi and Hebei, if only one member of the couple is a member of a minority and the city has a population of less than 10 million, the couple may apply to have a second child. There are five autonomous regions for ethnic minorities at the provincial level; 77 cities at the prefectural level, prefectures, autonomous prefectures,

2

years old, or the person can select a nationality when he or she is eighteen or older. One’s nationality cannot be changed after the age of twenty. Article 4: All nationalities in the People’s Republic of China are equal. The state protects the lawful rights and interests of the minority nationalities and upholds and develops a relationship of equality, unity, and mutual assistance among all of China’s nationalities. Discrimination against and oppression of any nationality are prohibited; any act which undermines the unity of the nationalities or instigates division is prohibited. The state assists areas inhabited by minority nationalities in accelerating their economic and cultural development according to the characteristics and needs of the various minority nationalities. Regional autonomy is practiced in areas where people of minority nationalities live in concentrated communities; in these areas organs of self-government are established to exercise the power of autonomy. All national autonomous areas are integral parts of the People’s Republic of China. All nationalities have the freedom to use and develop their own spoken and written languages and to preserve or reform their own folkways and customs.

422

Ding Sai, Li Shi, and Samuel L. Myers, Jr.

and mengs (leagues) at the prefectural level; 698 districts under the jurisdiction of cities, cities at the county level, counties, banners, autonomous counties, and autonomous banners at the county level; and 7,745 administrative units at the township and town government level (National Bureau of Statistics [NBS] 2009a). These areas with high concentrations of ethnic minorities have special political and administrative status. The Law of Regional National Autonomy was enacted in 1984 and updated in 2001. It is one of the three basic political systems in China. Ethnic regional autonomy is under the leadership of China’s central government and is implemented in the ethnic minority autonomous areas. According to the Law of Regional National Autonomy, once autonomous agencies are established, minorities have the right to autonomy, and they can manage their own internal affairs in the ethnic minority autonomous areas. The Regulations on Urban Nationality Work were enacted in 1993. Among its thirty articles, thirteen articles encourage the hiring of more minorities, the generation of minority enterprises, the training and selection of minority cadres, attention to minority education, and the provision of tax rebates. In addition, international conventions about discrimination against minorities affect China as well, such as the International Covenant on Economic, Social, and Cultural Rights, which entered into effect in China in March 2001; the International Convention on the Elimination of All Forms of Racial Discrimination, which entered into effect in December 1981; and the Employment Policy Convention, which was ratified in December 1997.

IV. The Model The conventional human capital perspective posits that (the log of) wage and salary incomes depend on experience and education, proxied by age, age-squared, and educational attainment or years of education. Within the context of China, however, one must also account for the industrial structure. The market reforms have resulted in an occupational class that is related to the educational system as well as to the hierarchical structure of the labor market, which, in turn, influences wage determination. We first consider the determination of wages as a function of human capital, family structure, industry, occupation, and location. We then detail our method for decomposing wages between minorities and Han. Finally, we describe a technique for understanding the changes in the ratio of minority-to-Han incomes over time.

Intertemporal Changes in Ethnic Urban Earnings Disparities in China 423

A. The Effects of Minority Status on Wage and Salary Income Consider a vector of human capital and industry/occupational indicators, X. Denote minority status by M, equal to one if a person is a member of one of the fifty-five officially recognized minority groups and equal to zero otherwise. We estimate the following model separately for males and female for each period t:  ln y t = αt0 + αti x ti + δt Mt + εt , (1) where the random error term, ε, is assumed to be normally distributed, with a zero mean and a constant variance, and is assumed to be uncorrelated with M or X. The test of the hypothesis that there is no adverse impact of minority status on earnings, once one controls for human capital, industry, and occupational characteristics, is δ = 0. An alternative way to test the hypothesis that there is no adverse impact of minority status once one controls for relevant human capital, industry, and occupational factors is to do the following: estimate the log-earnings equation separately for minorities and nonminorities, denoted by the superscripts m and h:  βhit x hit + ωht ln y ht = βht0 + (2)  m m m m m ln y t = βt0 + βit x it + ωt , There is no reason to assume that the error terms in the h and m equations are the same, nor is it necessary to assume that the effects of x’s on y are the same for both minorities and nonminorities. These are restrictions imposed by estimating Equation (1). So, an alternative measure of the adverse impact on earnings of being a minority would be to compute the counterfactual earnings of minorities when they face the same “treatment” as nonminorities:  ˆh (3) βˆhit · x m ln ym t = βt0 + it . An alternative measure of the unexplained gap in earnings, or the portion of the earnings that cannot be attributed to differences in the characteristics of minority and Han, is given by =

m ln ym t − ln y t , ln y ht − ln y m t

(4)

where the numerator is the unexplained residual difference in log earnings and the denominator is the actual gap in earnings. The ratio is the proportion of the total gap in log earnings that cannot be explained by differences

424

Ding Sai, Li Shi, and Samuel L. Myers, Jr.

in the characteristics of Han and minorities. This is the familiar BlinderOaxaca decomposition. We hypothesize that patterns of unexplained residuals will be different between males and females, with minority males facing larger disparities than do minority females, and we hypothesize that the unexplained disparities will differ across years. To know, however, how many of the intertemporal changes in characteristics explain the pattern of changing income disparities requires that we decompose the gaps between periods.

B. The Determinants of Changes in Minority-Han Income Disparities Consider the measure I(t, t + 1), which denotes minority-Han earnings disparities between two periods, t and t + 1. Earnings in periods t and t + 1 for Han and ethnic minorities, h and m, can be given by  m m m βm ln y m t = β0,t + it x it + ωt  m m m ln y m βm t+1 = β0,t+1 + it+1 x it+1 + ωt+1 (5)  h h h h h βit x it + ωt ln y t = β0t +  ln y ht+1 = βh0,t+1 + βhit+1 x hit+1 + ωht+1 . If the ratio of minority-to-Han earnings rises from period t to period t + 1, then earnings disparities are declining. When the numerator of I is larger than the denominator (the earning ratio in period t is greater than the earnings ratio in period t + 1), then the earnings gaps are widening. Thus, Equation (6) provides a means for summarizing the components of the changes in disparities between periods: ⎤ ⎡ m yt h y h m h I (t, t + 1) = ln ⎣ m  t ⎦ = ln y m t − ln y t − ln y t+1 + ln y t+1 . (6) y t+1 h y t+1 Note that changes in any particular factor, say, xj , affect the earnings disparities in the following manner: ∂ ln y m ∂ ln y ht+1 ∂ ln y m ∂ ln y ht ∂I t+1 t = − − + ∂x j ∂x j ∂x j ∂x j ∂x j h m h = βm t − βt − βt+1 + βt+1 .

(7)

A factor xj contributes to the narrowing of an earnings gap when its marginal impact on I is negative. When the sign of the derivative in Equation (7) is positive, the factor contributes to a widening of the earnings gap.

Intertemporal Changes in Ethnic Urban Earnings Disparities in China 425

In particular, this derivation permits us to determine whether particular factors, such as educational achievement or employment in foreign-owned enterprises, have consistent impacts on minority-Han wage disparities. Two key policy instruments available to the central and provincial governments are the expansion of educational opportunities for minorities through preferential treatment in college admissions or differential scoring on entrance examinations, and preferential hiring in state-owned enterprises (SOEs). One would expect, for example, that uniform expansions of education and of employment in SOEs would narrow the gaps in earnings if returns to education and employment in SOEs were increasing for minorities. However, if the returns to education or employment in SOEs were higher for Han than for minorities, then the effect of a uniform increase in education or employment in SOEs would result in a widening of the disparities. The disadvantage of measuring changes in earnings disparities by Equation (7) is that it assumes that there is a constant change in each of the independent variables. An alternative derivation based on Smith and Welch (1975, 1977, 1989) and Darity, Myers, and Chung (1998) considers the decomposition of the disparity into portions due to differences in the coefficients between groups and between periods, and differences in the endowments between groups and between periods. Two different decompositions can be envisioned: an intertemporal decomposition that examines the differences in endowments and coefficients between periods and an intratemporal decomposition that examines the differences in endowments and coefficients between groups within periods.

C. Intratemporal Decomposition This decomposition divides I(t, t + 1) into a portion that is due to differences in the treatment of minorities and Han within each period and a portion that is not due to such differences within a period. The portion that is not due to differences in treatment within a period is due to differences in endowments within the period. Equation (8) shows that the disparity measure, I(t, t + 1), can be rewritten as the sum of the treatment and endowment effects: h m h ln I (t, t + 1) = ln y m t − ln y t − ln y t+1 + ln y t+1

 m  h m h = ln y m y t − ln ym t − ln y t − ln y t+1 + ln y t+1 + ln t   m ym + ln y t+1 − ln t+1   m   ym ym = ln y m t − ln t − ln y t+1 − ln t+1   m   m y t+1 − ln y ht+1 , (8) + ln y t − ln y hy − ln

426

Ding Sai, Li Shi, and Samuel L. Myers, Jr.

where the first bracketed expression is the treatment effect and the second bracketed expression is the endowment effect. The equal treatment value of income in a given period j is given by  ˆh (9) ln ym βˆhi,j · x m j = β0,j + i,j , denoting the income of minorities if they faced the treatment of Han in period j. It is the predicted value of the ln-earnings for minorities if they were treated as nonminorities but had the characteristics of minorities. If the coefficients on all of the betas are the same within a period for both minorities and nonminorities, the left-hand-side value in Equation (9) will be equal to the minority ln-earnings, resulting in the first bracketed term in Equation (8) being equal to zero.

D. Intertemporal Decomposition This decomposition divides I(t, t + 1) into portions that are due to intertemporal treatment effects, wherein the treatment of both minorities and Han in period t + 1 is the same as it is in period t and is due to intertemporal endowment effects, wherein the endowments in period t + 1 are the same as they are in period t : h − ln y ht − ln y m ln I (t, t + 1) = ln y m t+1 + ln y t+1  t m    h = ln y t − ln ym y ht,t+1 t,t+1 − ln y t − ln   h   m y t,t+1 − ln y ht+1 , (10) + ln y t,t+1 − ln y m t+1 − ln

where the intertemporal equal treatment for the kth group is given by  ln y kt,t+1 = βˆk0,t + (11) βˆki,t · x ki,t+1 . Equation (11) denotes the instance in which the kth group’s treatment in period t + 1 is predicted by its treatment in period t but by its characteristics in period t + 1. Thus, it is possible to decompose the disparities measure I(t, t + 1) into portions that can be attributed to (a) differences in endowments within groups between periods and (b) differences in the rates of return on those endowments (or treatment) between periods.

V. Data and Descriptive Statistics This chapter uses data from the 1995, 2002, and 2007 China Household Income Project (CHIP) urban surveys. The CHIP data are part of the data collected through a sample survey of urban households conducted by the National Bureau of Statistics (NBS). To make the data comparable

Intertemporal Changes in Ethnic Urban Earnings Disparities in China 427

across years, we restrict the analysis to the provinces that are common to all three urban surveys. The twelve common provinces for the three years are Beijing, Shanxi, Liaoning, Yunnan, Gansu, Jiangsu, Anhui, Henan, Hubei, Guangdong, Chongqing, and Sichuan. Our samples were collected in cities of various sizes, numbering 69 in 1995, 77 in 2002, and 300 in 2007. The sample cities and towns in the urban areas are selected by using a stratified random sampling method (NBS 2009b), in which stratification is based on province and city size. The sampling of households within cities and towns results in a random population sample. For the purposes of the creation of the CHIP sample, households were selected randomly from provinces organized along the geographic distribution of the national population. Accordingly, the CHIP urban sample is regarded as a selfweighted sample. An important limitation of the CHIP urban sample is that it excludes persons who work in the urban area but whose hukou is elsewhere. This exclusion has important implications for the interpretation of our results, to which we return in a concluding section of the chapter. To facilitate the estimation of Equation (5), we identified variables that are common across all three surveys. They include age, years of education, minority status, and household head. In our analysis, these are identified as human capital variables. The occupational variables include owner or manager of a private enterprise; professional or technical worker; manager of an institution; and workers, including office workers, skilled workers, and unskilled workers; or other occupations not classified elsewhere. The excluded category in the analysis is professional or technical workers. The type of firm includes SOEs, including local publicly owned firms; collectives; privately owned firms or self-employed firms, including partnerships and individual enterprises; and other types of firms such as Sinoforeign joint ventures, foreign-owned firms, township and village enterprises, and jointly owned economic units; limited liability corporations; and shareholding corporations, foreign-funded economic units, and overseas Chinese from Hong Kong-, Macao-, and Taiwan-funded economic units. The comparison group for the purposes of the regression analysis is collectives. Table 12.1 provides the key information that is the source of our inquiry: changes in wage and salary incomes across the years for the common provinces. The table highlights the dramatic increase in wage and salary incomes as well as the changes in the disparities between minorities and Han.

428

Ding Sai, Li Shi, and Samuel L. Myers, Jr.

Table 12.1. Minority and Han wage or salary income in the same twelve provinces Minority

Han

Total

1995 Observations Percentage Average individual wage or salary income

967 4.46 5,243.69

20,729 95.54 5,744.40

21,696 100 5,723.15

2002 Observations Percentage Average individual wage or salary income

902 4.41 10,527.53

19,537 95.59 10,620.19

20,439 100 10,616.34

2007 Observations Percentage Average individual wage or salary income

781 3.50 17,237.08

21,548 96.49 19,658.51

22,333 100 19,577.15

Source: Authors’ computations using the 1995, 2002, and 2007 CHIP data.

Wage and salary earnings nearly doubled between 1995 and 2002 and increased more than threefold between 1995 and 2007. In each year, minority wage and salary incomes lagged behind those of Han. For example, in 2007, the annual average wage and salary income of urban residents in the common provinces was 17,237 yuan for minorities but 19,659 yuan for Han. This gap is wider than it was in 1995, when minorities earned 5,244 yuan and Han earned 5,744 yuan. Table 12.2 reports on the Han-minority disparities in income within age groups and Han-minority disparities between age groups, separately for Han and minority males and females. An important insight gleaned from this table is that within the gender groups there are widening disparities between younger workers (eighteen to thirty years of age) and older workers (thirty-one to sixty years of age). This limitation of persons to ages eighteen to sixty represents a partition of the sample that is not reported in Figure 12.2. There we saw that there was an improvement in relative earnings of minority females from 1995 to 2007. Table 12.2 shows that the ratio of minority-to-Han earnings among females improved from 91.56 percent in 1995 to 108.68 percent in 2002. But the ratio dropped to 89.27 percent in 2007, denoting a slight decline for this restricted age group. Still, the broad year-to-year patterns for both males and females ages eighteen to sixty are the same as those found in Figure 12.2: there was a continuous decline in the ratio of minority-to-Han earnings among males from 1995 to 2002 to 2007. There was also an increase in the ratio among females from 1995 to 2002 and

Table 12.2. Ratio of minority-to-Han income and ratio of the income of those eighteen to thirty years old to the income of those thirty-one to sixty years old 1995 Minority/ Han 429

Total, Males Males, 18–30 Males, 31–60 Between–age-group disparity Total Females, 18–30 Females, 31–61 Between–age-group disparity

Minority 18–30 31–60

2002 Han 18–30 31–60

91.16% 105.53% 89.78%

Minority/ Han

Minority 18–30 31–60

2007 Han 18–30 31–60

90.69% 68.38% 97.78% 78.73%

66.98%

91.56% 82.22% 93.86% 75.04%

Minority 18–30 31–60

Han 18–30 31–60

120.57%

104.03%

120.14%

110.16%

84.65% 86.17% 74.35% 51.49%

73.62%

108.68% 105.46% 111.06% 65.73%

Minority/ Han

89.27% 95.74% 87.78% 77.84%

81.98%

Source: Authors’ computations using the 1995, 2002, and 2007 CHIP data. Between-age-group disparities noted in italics.

430

Ding Sai, Li Shi, and Samuel L. Myers, Jr.

then a decline from 2002 to 2007. For males between the ages of eighteen and sixty, the minority-to-Han ratio was 91.16 percent in 1995, 90.69 percent in 2002, and 84.65 percent in 2007. Within specific age groups, however, the patterns diverge. For example, among males between the ages of eighteen and thirty, the minority-to-Han earnings ratio dropped from 105.53 percent in 1995 to 68.38 percent in 2002; it thereafter rose to 86.17 percent in 2007. By way of contrast, among males between the ages thirty-one and sixty, the ratio was 89.78 percent in 1995, 97.78 percent in 2002, but 74.35 percent in 2007. Among females, the minority-to-Han earnings ratio rose for both age groups from 1995 to 2002 but fell from 2002 to 2007, although the ultimate result of the changes between 1995 and 2007 differs between the eighteen- to thirty-year-olds and the thirty-one- to sixty-year-olds. Among the younger females, the ratio was larger in 2007 than it was in 1995; among the older females, the ratio was smaller in 2007 than it was in 1995. Table 12.2 also reveals information on the changing earnings within ethnic groups between younger and older workers. The rows labeled “between age group disparity” compute the ratio of the earnings of eighteen- to thirty-year-olds to the earnings of thirty-one- to sixty-year-olds within an ethnic group, by year and by gender. Among minority males, the ratio of earnings of eighteen- to thirty-year-olds to the earnings of thirty-one- to sixty-year-olds declined from 78.73 percent in 1995 to 51.49 percent in 2002. Among Han males, the ratio rose from 66.98 percent in 1995 to 73.62 percent in 2002. For both Han and minority males, the ratio jumped to over 100 percent in 2007. For both minority and Han females, the ratio of earnings between the younger age group and the older age group increased continually from 1995 to 2007. In short, there are important age and gender differences in the changing Han-minority patterns in earnings disparities. Controlling for these differences may account for the observed differences in earnings across years.

VI. Results Equation (1), which predicts ln-earnings as a function of M, the minority dichotomous variable, provides the starting point for our analysis. Ordinary least squares estimates of the coefficient δ, the percentage difference in earnings due to minority status, were obtained separately for year and gender, first without controls and then controlling successively for human capital, family structure, occupation, industry, and province. Table 12.3

Intertemporal Changes in Ethnic Urban Earnings Disparities in China 431 Table 12.3. Ordinary least squares estimates of the effects of minority status on ln-earnings

Males Unadjusted Adjusted for human capital Adjusted for human capital, occupation, and type of firm Adjusted for human capital, occupation, type of firm, and province Females Unadjusted Adjusted for human capital Adjusted for human capital, occupation, and type of firm Adjusted for human capital, occupation, type of firm, and province

1995

2002

2007

− 0.086 (2.27)** − 0.055 (1.59) − 0.048 (1.41) − 0.071 (2.17)**

− 0.202 (3.22)*** − 0.102 (1.86)* − 0.062 (1.16) − 0.070 (1.32)

− 0.194 (2.19)** − 0.192 (2.33)** − 0.152 (2.05)** − 0.158 (2.18)**

− 0.090 (1.99)** − 0.076 (1.85)* − 0.087 (2.16)** − 0.115 (2.95)***

0.120 (1.65) 0.138 (2.13)* 0.199 (3.16)** 0.177 (2.78)***

− 0.183 (1.75)* − 0.133 (1.44) − 0.044 (0.53) − 0.043 (0.51)

Notes: Absolute value of t-statistics in parentheses. Full regression results are available at http:// www.hhh.umn.edu/centers/wilkins/pdf/DoesaRisingTideLiftAllShips.pdf. Accessed October 3, 2011. * significant at 10 percent; ** significant at 5 percent; *** significant at 1 percent.

presents these results, providing separately the coefficients on M for males and females in each year for each set of controls. Without controls, the estimate of δ is negative and statistically significant for males in all years and for females in 1995 and 2007, which confirms our earlier results of a Han-minority disparity. The interpretation of the estimated coefficients on δ is the percentage difference in earnings between minorities and Han. A negative coefficient indicates that minorities earn less than Han. The first set of rows for males and females do not control for age, education, or family structure. When one controls for these human capital– related variables, the size of the estimated coefficient on minority status, δ, drops from −0.086 to −0.055 and from −0.202 to −0.102 for males in 1995 and 2002. The 1995 coefficient is not significant; the 2002 coefficient is barely significant. Thus, controlling for human capital factors in 1995 and 2002 “explains” many of the differential earnings among male minorities. In 2007, by way of contrast, controlling for human capital variables leaves the estimated coefficient δ largely unchanged.

432

Ding Sai, Li Shi, and Samuel L. Myers, Jr.

The unadjusted value is −0.194. The adjusted value for human capital factors is −0.192. Controlling for occupation and type of firm further reduces the size and significance of the coefficient on minority status, but adjusting for the province effects produces revised estimates of δ that are statistically significant in 1995 and 2007, although smaller in absolute value than the unadjusted values. In 1995, the estimate of δ is −0.071. In 2007, the estimate is −0.158. In short, using this model, we conclude that there is a negative effect of minority status on earnings in 1995 and 2007 and that effect increases in absolute value. Among females, the unadjusted effect of minority status is negative and statistically significant in 1995 and 2007, but positive and statistically significant in 2002. Controlling for human capital, occupation, type of firm, and province, the estimated coefficients on δ are negative and statistically significant in 1995, positive and statistically significant in 2002, and negative but not statistically significant in 2007. The coefficients in 1995 and 2007 are −0.115 and −0.043, reflecting a decline in the adverse impact of minority status on earnings among females between the two years.

A. Returns to Education and Premium to SOEs It is instructive to isolate two key economic factors that appear to have consistently significant impacts on earnings for males and females and for each ethnicity. Educational attainment in every instance has a positive and significant coefficient across years and across gender and ethnicity. Employment in state-owned enterprises also has positive and significant coefficients. Table 12.4 reports these results. It shows returns to education in 1995 of 2.5 to 3.9 percent, on the same order of magnitude reported by Maurer-Fazio for a similar period. In 2002, these returns explode to 9 percent for males and 13 percent for females. In 2007, the estimated returns to education were 8.3 percent for males and 11 percent for females. As can be seen in Table 12.4, there were marked differences in the rates of return to education in 1995, but virtually no difference between minorities and nonminorities in returns to education in 2002 and 2007. In 1995, the returns to education for minority males were 3.5 percent, but for Han, they were 2.4 percent. For females in 1995, the returns to education were 5.5 percent for minorities, but 3.8 percent for Han. This evidence of higher rates of return to education for minorities disappears in 2002 and 2007, where the estimated coefficients are remarkably similar for minorities and Han. The convergence in the returns to education is an important finding that heretofore has not been recognized.

Intertemporal Changes in Ethnic Urban Earnings Disparities in China 433 Table 12.4. Returns to education and employment in state-owned enterprises 1995 Male Education

Female

2002 Male

Female

2007 Male

Female

All groups

0.025 0.039 0.090 0.133 0.083 0.110 (10.30)*** (12.48)*** (23.04)*** (27.99)*** (16.90)*** (17.95)*** Minorities 0.035 0.055 0.099 0.133 0.083 0.102 (2.55)** (2.82)*** (5.19)*** (6.00)*** (3.01)*** (2.89)*** Han 0.024 0.038 0.089 0.133 0.083 0.110 (9.95)*** (12.26)*** (22.20)*** (27.32)*** (16.52)*** (17.58)***

State-owned enterprises All groups

0.215 0.280 0.068 0.148 0.715 0.702 (11.12)*** (14.51)*** (2.34)** (3.99)*** (23.40)*** (19.46)*** Minorities 0.108 0.295 0.212 − 0.177 0.833 0.772 (1.08) (2.33)** (1.21) (0.64) (4.60)*** (3.58)*** Han 0.220 0.278 0.062 0.153 0.710 0.702 (11.12)*** (14.33)*** (2.12)** (4.08)*** (22.83)*** (19.13)*** Notes: Absolute value of t statistics in parentheses; OLS estimates of coefficients on education and SOEs in ln-wage equation, controlling for age, education, household head, occupation, type of firm, and province. Full regression results are available at http://www.hhh.umn.edu/centers/wilkins/pdf/ DoesaRisingTideLiftAllShips.pdf. Accessed October 3, 2011. * significant at 10 percent; ** significant at 5 percent; *** significant at 1 percent.

Table 12.4 also provides estimates of the premium associated with employment in SOEs. During the expansion of the Chinese economy and the opening of the private sector, the share of jobs in non-SOEs increased substantially. Table 12.5 reports that the share of Han versus minority male workers employed in private enterprises skyrocketed from 1.46 versus 2.19 percent in 1995 to 14.96 versus 15.97 percent in 2007. The share of workers in various forms of foreign-owned or jointly owned partnerships and corporations also increased substantially. In contrast, the share of Han versus minority workers in SOEs dropped from 86.03 percent and 84.01 percent for Han and minority males in 1995 to 73.81 percent and 73.95 percent, respectively, in 2007. For females, the share of workers employed in SOEs dropped from 75.41 and 77.64 percent for Han and minorities in 1995 to 64.26 and 63.37 percent for Han and minorities in 2007. Surprisingly, though the premium associated with employment in SOEs was once higher for Han males than for minority males, from 1995 to 2007 there was a faster rise for minority males than for Han males. By 2007, the SOE employment premium for minority males was higher than that for Han males. The premium for employment in SOEs was 10.8 percent and

434

Ding Sai, Li Shi, and Samuel L. Myers, Jr. Table 12.5. Descriptive statistics from the 1995, 2002, and 2007 CHIP data 1995

MALES Age Education Occupation Owner or manager of private enterprise or self-employed Professional or technical Manager of government department or institution Worker or other

2002

2007

Han

Minority

Han

Minority

Han

Minority

42.98 10.8

41.71 10.46

43.06 12.31

42.29 12.05

39.17 12.51

38.25 12.58

1.65

0.96

1.9

6.48

1.08

0.67

21.2 18.62

21.09 14.06

20.95 15.81

18.52 15.28

18.05 6.77

19.46 6.71

58.53

63.9

61.35

59.72

74.1

73.15

Ownership SOEs Collectives Private or self-employed Other

86.03 10.99 1.46 1.53

84.01 12.54 2.19 1.25

83.67 11.45 3.9 0.98

76.92 17.95 5.13 0

73.81 6.18 14.96 5.05

73.95 5.04 15.97 5.04

FEMALE Age Education

42.35 9.81

42.75 9.58

40.45 12.01

39.42 11.95

37.41 12.51

38.39 12.59

1.49

2.25

1.53

5.75

1.05

3.25

21.24 5.94

24.12 2.57

23.34 5.02

29.31 5.17

17.74 2.84

24.39 3.25

71.34

71.06

70.11

59.77

78.37

69.11

75.41 21.17

77.64 18.94

79.69 16.47

61.11 22.22

64.26 9.39

63.37 7.92

Occupation Owner or manager of private enterprise or self-employed Professional or technical Manager of government department or institution Worker or other Ownership SOEs Collectives

Source: Authors’ computations from the 1995, 2002, and 2007 CHIP data.

22 percent for minority males and Han males, respectively, in 1995. The premium for employment in SOEs was 83.3 percent and 71.0 percent for minority and Han males in 2007. This surprisingly large shift over such a short time is consistent with the hypothesis that changing labor-market structures toward privatization have had a large influence on disparities in ethnic earnings.

Intertemporal Changes in Ethnic Urban Earnings Disparities in China 435 Table 12.6. Residual difference analysis of ethnic minority versus Han wage and salary income

HanHan Minority minority (1) (2) (1) − (2) Males 1995 2002 2007

Minority treated as Han Residual Unexplained percentage (3) (3) − (2) [(2) − (3)]/[(1) − (2)]

8.498 8.709 9.247

8.382 8.511 9.056

0.116 0.198 0.192

8.415 8.610 9.145

0.033 0.099 0.089

28.3% 50.2% 46.7%

Females 1995 8.139 2002 8.268 2007 8.604

7.639 8.550 8.475

0.500 − 0.281 0.129

8.079 8.204 8.511

0.441 − 0.346 0.037

88.1% 122.9% 28.4%

Notes: From ln-earnings regressions controlling for age, education, household head, occupation, type of firm, and a regional dummy variable for the western and central provinces.

Among females, the premium associated with employment in SOEs also increased substantially. Among all females, the premium associated with employment in SOEs rose from 28.0 percent in 1995 to 70.2 percent in 2007. Among minority females, the premium was 29.5 percent in 1995 and 77.2 percent in 2007. Among Han females, the premium was 27.8 percent in 1995 and 70.2 percent in 2007. Thus, the marginal returns to working in SOEs were remarkably similar for minority females and Han females.

B. Residual Difference Analysis The estimates of Equation (1) hinge on the untenable assumption that in every case there are no interactions between minority status and the other variables. Therefore, we have estimated separate regressions for Han and minorities in 1995, 2002, and 2007 and decomposed the gaps in ln-earnings between the explained and unexplained portions, as indicated in Equations (2) through (4). The results are displayed in Table 12.6. The first column of the table presents the ln-earnings for Han by gender and year. The second column presents the ln-earnings for minorities by gender and year. In the third column, the difference between Han and minority ln-earnings is computed. This difference is positive in every year for both males and females except for 2002, when minority females register higher ln-earnings than Han

436

Ding Sai, Li Shi, and Samuel L. Myers, Jr.

females. These disparities require explanation. Therefore, ln-earnings are estimated separately for Han and minorities by gender to compute the predicted ln-earnings of minorities when they face the same treatment as Han. This computation is displayed in the column labelled (3) in the table. The difference between columns (2) and (3) is the unexplained residual difference in ln-earnings, or the portion of the gap in ln-earnings that cannot be explained by differences in endowments. The ratio of this difference to the actual disparity in ln-earnings (multiplied by 100) is the percentage of the disparity that is unexplained. Two key conclusions emerge from these calculations. First, the unexplained percentage declined from 1995 to 2007 for females, but it increased for males. The percentage of the minority versus Han earnings disparity that was unexplained in 1995 was 28 percent for males and 88 percent for females. In 2007 the measure of the unexplained gap rose to 47 percent for males, but it fell to 28 percent for females. A second conclusion is that between 2002 and 2007, a period of recovery, the unexplained gap declined slightly for males, from 50 percent to 47 percent, but it increased for females as females moved from a favored to a less-favored position. Still, the unexplained percentage for males was larger than that for females in 2007.

C. Determinants of Earnings Disparities Equations (5) through (7) provide a preliminary tool to explore the relative contributions of specific factors in determining minority-Han earnings disparities over time. The results of a unit increase in each factor show that for both males and females any increases in educational attainment are associated with increases in disparities. These positive values, however, are not always statistically significant and they are relatively small in magnitude. Moreover, as we have already seen, in recent years the actual returns to education have been remarkably similar both for minorities and for Han. Likewise, age effects are small and statistically insignificant. The larger impacts appear to be relative to changes in firm type. For males, an increase in employment in SOEs is associated with a large reduction in minority-Han earnings disparities between 1995 and 2007. The net effect for females combines two opposing impacts. Table 12.7 shows that from 1995 to 2002 inequality for females increased as a result of employment in SOEs (.386). From 2002 to 2007, however, inequality declined (−0.316). Thus, inequality during the period from 1995 to 2007 increased slightly

Intertemporal Changes in Ethnic Urban Earnings Disparities in China 437 Table 12.7. Determinants of changes in the disparities in ethnic earnings ∂I (t, t + 1) h m h = βm t − βt − βt+1 + βt ∂x i

(Eq. 7) 1995–2002

2002–2007

1995–2007

Male Age Education Household head Owner or manager of private enterprise Manager of institution Worker or other State-owned enterprise Private or self-employed Others

− 0.01 0.011 − 0.225 0.212a − 0.141 0.021 − 0.285 − 0.247 0.206b

0.006 0.011 0.418 − 1.078 − 0.142 − 0.193 0.035 0.46 − 0.362

− 0.004 0.021 0.194 − 0.866 − 0.283 − 0.171 − 0.251 0.213a − 0.518

Female Age Education Household head Owner or manager of private enterprise Manager of institution Worker or other State-owned enterprise Private or self-employed Others

− 0.036 0.006 0.111 − 0.206 − 0.145 0.044 0.386 − 0.583 0.695

0.02 0.015 − 0.409 0.887 0.555 0.286 − 0.316 0.214 0.039

− 0.016 0.021 − 0.298 0.681 0.409 0.33 0.071 − 0.369 0.733

Notes: Estimates include regional dummy variables (not shown) for western and central provinces. a The coefficients for these variables could not be estimated for the minorities in 1995. The number reported assumes that this coefficient is zero. b The coefficients for these variables could not be estimated for the minorities in 2002. The number reported assumes that this coefficient is zero.

(0.071). This contrasts with the larger reduction in inequality between Han and minority males due to employment in SOEs (−0.251). Table 12.7 also shows that for the 1995–2007 period a uniform increase in employment as a manager increased inequality for females (0.409) but reduced inequality for males (−0.283). An alternative way of thinking about the decomposition of the disparity measure is to consider intertemporal differences in treatment versus endowments and intratemporal differences in treatment and endowments. The results from computing the values detailed in Equations (8) to (11) are provided in Table 12.8.

438

Ding Sai, Li Shi, and Samuel L. Myers, Jr.

Table 12.8. Intratemporal and intertemporal decomposition of the disparity measure 1995–2002 Males I(t + 1) Intratemporal decomposition Treatment effect Endowment effect Intertemporal decomposition Treatment effect Endowment effect

2002–2007

1995–2007

Females

Males

Females

Males

Females

0.082

−0.781

−0.007

0.41

0.075

−0.371

82% 18%

101% −1%

143% −43%

93% 7%

76% 25%

109% −9%

−133% 233%

−3% 103%

43% 57%

18% 82%

−92% 192%

−6% 106%

Notes: The ln-earnings estimates include age, education, household head, occupation, type of firm, and a regional dummy variable for the western and central provinces.

The first row in the table reports I(t, t +1), or the change in the disparity measure for males and females for the 1995–2002, 2002–2007, and 1995– 2007 periods. Note that when this index is positive, the ratio of minorityto-Han earnings is declining, or the earnings disparity is rising. When the ratio is negative, the earnings disparity is declining. The first row indicates that from 1995 to 2002 the earnings disparity widened for males and narrowed for females, with the values of I(t, t + 1) equal to 0.082 and −0.781 respectively. From 2002 to 2007, the earnings disparity narrowed slightly for males and widened for females, with the values of I(t, t +1) equal to −0.007 and 0.410 respectively. The net effect from 1995 to 2007, equal to the sum of the effects from 1995 to 2002 and from 2002 to 2007, was a widening of the earnings disparity for males and a narrowing of the earnings disparity for females, with the values of I(t, t + 1) equal to 0.075 and −0.371, respectively. Next, we explain how this earnings disparity breaks down between differences in treatment and differences in endowments.

D. Intratemporal Decomposition The second set of rows in Table 12.7 reports the decomposition of the disparity measure into portions that can be explained by differences in the coefficients between the minority and Han ln-earnings regressions within each period and the differences in endowments within each period. The former is called the treatment effect. The latter is called the endowment effect. The computation asks how much of the observed change in earnings disparities

Intertemporal Changes in Ethnic Urban Earnings Disparities in China 439

can be attributed to differences in treatment between minorities and Han and how much of it can be attributed to differences in endowments. Between 1995 and 2002, almost all of the change in earnings disparities for females can be attributed to differences in treatment. In fact, one can argue that the differences in treatment disproportionately favored minority females over Han females in urban areas. During the same period, most of the change in earnings disparities among males – 82 percent – can also be attributed to differences in treatment. Similar findings emerge for the 2002–2007 period, leading to the conclusion that the dominant component of the intratemporal change in earnings disparities can be attributed to minority-Han differences in treatment. Because earnings disparities declined for females, however, the differences in treatment favored minority females, or acted as a form of what some analysts might call reverse discrimination. An alternative interpretation is that the preferred position of minority females is the result of preferences for urban minority females that produce higher lnwages than identically situated Han females. Among the males, the opposite impact is found. Because the earnings disparities increased, differences in the returns to endowments between Han and minority males produced an adverse impact on the relative wages of minority males.

E. Intertemporal Decomposition The second decomposition displayed in Table 12.8 considers the partitioning of the disparity measure into portions attributable to differences in endowments within a group between periods and differences in the returns to such endowments. The same group is being compared to itself during the two periods. As we noted in the data description, the composition of the groups changed as did the relative earnings of younger and older members of each group. Unsurprisingly, almost none of the intertemporal changes in earnings disparities can be attributed to differences in treatment of minorities in one period versus treatment of minorities in another period or to differences in treatment of Han in one period versus treatment of Han in another period. Instead, most of the changes can be attributed to changes in endowments. These results are tempered by the fact that we focus solely on urban wage earners. There are three forms of selection that this analysis does not take into account. The first is the selection of wage earners among all potential workers. Darity and Myers (2001), using data on blacks and whites in the United States, show that this type of selection will bias upward measures of minority-majority earnings. The second form of selection, alluded to in

440

Ding Sai, Li Shi, and Samuel L. Myers, Jr.

the introduction to this chapter, involves the migration of the most talented minorities from rural areas to urban areas. This sort of (self-)selection helps to explain how it is possible for minority-majority income disparities to be narrowing in urban areas while they are widening in rural areas. A third unexplored form of selection is a policy-induced selection. Preferences for minorities in college admissions or in hiring for government jobs or SOEs can produce a concentration of highly qualified minorities in locations like Beijing, the seat of the central government where there are large numbers of college graduates. In short, the underlying measures of urban earnings disparities examined in this chapter reflect multiple sources of selection that merit additional investigation in future research.

VII. Summary and Conclusion This chapter provides documentation on a pattern of first narrowing and then widening of minority-Han earnings disparities between 1995 and 2007 among urban workers. The patterns differ for males and females, with a widening occurring among minority versus Han males but an initial narrowing occurring among minority versus Han females followed by a more recent widening of the earnings gaps.

A. Females A key component in the change in minority-Han earnings gaps is the difference in treatment. For 1995, we estimate the portion of the gap between minority and Han females that is unexplained. By 2002, there continued to be minority-Han differences in treatment. Consistent with a national policy of preferences for minorities, the differences in treatment on some variables favored minority females, who experienced higher earnings in 2002 than did Han females. Accounting for differences in human capital, occupation, firm type, and province does not eliminate this apparent advantage experienced by minority females in 2002. Using a single regression equation and controlling for age, education, occupation, firm type, and province, in 2007 minority status has a small but statistically insignificant impact on ln-earnings. When a full residual difference model is estimated, a small unexplained gap is measured. On balance, any unexplained disparity in 1995 seems to have dissipated by 2007. Why did the ratio of minority-to-Han earnings rise for females between 1995 and 2002 and then fall in 2007? Only part of the answer can be found in the analysis of the returns to education or the returns to employment in

Intertemporal Changes in Ethnic Urban Earnings Disparities in China 441

SOEs. Differences in the rates of return to education and to employment in SOEs between Han and minorities can be interpreted as policy-induced differences in treatment. The policy instrument is understood to be preferences in college admission and hiring in SOEs. Table 12.4 shows that the returns to education and returns to employment in SOEs essentially converged for Han and minority females by 2007. The answer lies in part in the nature of the sample and the changing hukou policies. The urban sample does not include migrants with hukou in another jurisdiction. All persons in the urban sample are individuals with their hukou in the specified urban area. Thus, technically, the sample excludes migrants and persons with their hukou in other locations. However, an important policy change regarding hukou affects this interpretation. At the time of the data collection for the 2002 CHIP sample, rural persons admitted to universities in urban areas were permitted to change their hukou to the urban area. As such, some of the urban workers in the 2002 sample may well be migrants in the sense that their original hukou was in another jurisdiction, but their change in hukou was brought about because of their clearly selective admissions to universities. This permissive policy changed again by the time of the collection of the 2007 data. Therefore, even with selective admissions to universities, persons from other jurisdictions will not appear in the 2007 data set. Further evidence of a selective process involved in the determination of the earnings of minority females is found in Table 12.5. Minority educational levels jumped from 1995 to 2002. The share of minority females employed in professional jobs increased considerably. The percent employed as managers in government enterprises almost doubled, from 2.57 percent to 5.02 percent. In short, minority females earned more in 2002 than did Han females because the share of minority females among high earners increased.

B. Males The minority-Han ratio of wage and salary earnings among males, by way of contrast, declined steadily throughout the period examined. This widening gap in earnings cannot be attributed solely to ethnic differences in endowments. Our results suggest that there has been a rise in the unexplained portion of the overall gap in earnings between minority males and Han males in urban areas. Although the education attainment of urban minority males approached that of urban Han males and other measures of human capital also improved, the gap in earnings widened for males. The

442

Ding Sai, Li Shi, and Samuel L. Myers, Jr.

reason? Changes in the impacts that these human capital factors had on earnings. The improved endowments of urban minority males were overshadowed by their differential treatment relative to that of Han workers. An important insight is that minority male employment in managerial jobs and jobs in SOEs helps to reduce earnings disparities, partly because in recent years the estimated returns to managerial jobs and SOE employment have been higher for minorities than for Han. Still, other factors counteract these impacts. One of the greatest is the growing private-sector employment relative to SOE employment. Thus, although there are positive effects for urban minority males employed in SOEs, as the private sector expands, SOE employment represents a declining share of employment. Thus, to answer the core question raised at the outset of this chapter, we find that the two state policies of preferential treatment in education and employment had their intended impacts, at least among males. These policies worked to reduce gaps in earnings between Han and minority males. The returns to education were essentially equalized between Han and minority males by 2007 whereas the returns to employment in SOEs steadily rose for minority males to the point that by 2007, they were higher than the returns for Han males. However, these impacts were not large enough to prevent an otherwise widening disparity in earnings between Han and minorities. Among females, we find that the “unexplained” portion of the disparity in wage and salary incomes between Han and minorities narrowed from 1995 to 2007. We cannot directly attribute this narrowing of wage and salary incomes among females to state policies of preferential treatment in education and employment. But the evidence clearly shows that Hanminority wage and salary gaps were smaller in 2007 than they were in 1995. We have speculated that changes in hukou policies might account for the surprising jump in the ratio of minority-to-Han wage and salary incomes from 1995 to 2007. The sample is restricted to urban residents and technically excludes migrants. However, the policy that permitted university students from rural areas to change their hukou to the urban location of their universities could serve as a selection mechanism associated with the higher mean earnings of the beneficiaries of these policies. This, along with the policy of permitting rural women who marry urban men to change their hukou, the policy in force in 2002 of permitting minority students to change their hukou might explain the narrowing of the wage gap for females. In short, several forms of selection are likely to be underlying the following trends: widening earnings disparities between Han and minority males; and

Intertemporal Changes in Ethnic Urban Earnings Disparities in China 443

narrowing, and then widening, disparities among females. One obvious form relates to differential labor-force participation. Another relates to rural-urban disparities in income and the resulting migration from rural areas to urban areas (Sicular et al. 2007). Yet another form of selection that might explain the surprising results for females in 2002 is the policyinduced selection affecting the ability of minority females to change their hukou by marriage to urban males and/or via university enrollment. Other, less-well-understood forms of selection include the process of being selected for employment in SOEs via membership in the Communist Party. These complexities provide areas for future research. References Cai, F., D. Wang, and Y. Du (2002), “Regional Disparity and Economic Growth in China: The Impact of Labor Market Distortions,” China Economic Review, 13(2–3), 197– 212. Central People’s Government (2011), “Minwei fabu 2010 nian shaoshu minzu diqu nongcun pinkun jiance jieguo” (State Ethnic Affairs Commission Publishes the 2010 Poverty Monitor Results for Rural Ethnic Areas), July 29, http:www.gov.cn/gzdt/ 2011-07/29/content 1916420.htm. Accessed September 29, 2011. Chow, G.C. (1993), “Capital Formation and Economic Growth in China,” Quarterly Journal of Economics, 108(3), 809–842. Chow, G.C. and K.W. Li (2002), “China’s Economic Growth: 1952–2010,” Economic Development and Cultural Change, 51(1), 247–256. Darity, W.A., Jr. (1982), “The Human Capital Approach to Black-White Earnings Inequality: Some Unsettled Questions,” Journal of Human Resources, 17(1), 72–93. Darity, W. and S.L. Myers, Jr. (2001), “Why Did Black Relative Earnings Surge in the Early 1990s?” Journal of Economic Issues, 35(2), 533–542. Darity, W., S.L. Myers, Jr., and C. Chung (1998), “Racial Earnings Disparities and Family Structure,” Southern Economic Journal, 65(1), 20–41. Friedman, E. (2006), “Jiang Zemin’s Successors and China’s Growing Rich-Poor Gap,” in T.J. Cheng, J. deLisle, and D. Brown, eds., China Under Hu Jintao, 97–134, Singapore: World Scientific Publishing Co. Gustafsson, B. and S. Ding (2009), “Villages Where China’s Ethnic Minorities Live,” China Economic Review, 20(2), 193–207. Gustafsson, B. and S. Li (2003), “The Ethnic Minority-Majority Income Gap in Rural China During Transition,” Economic Development and Cultural Change, 51(4), 805– 822. Gustafsson, B.A., S. Li, and T. Sicular (2008), Inequality and Public Policy in China, New York: Cambridge University Press. Hannum, E. (2002), “Educational Stratification by Ethnicity in China: Enrollment and Attainment in the Early Reform Years,” Demography, 39(1), 95–117. Hannum, E. and X. Yu (1998), “Ethnic Stratification in Northwest China: Occupational Differences between Han Chinese and National Minorities in Xinjiang, 1982–1990,” Demography, 35(3), 323–333.

444

Ding Sai, Li Shi, and Samuel L. Myers, Jr.

Luo, C. and T. Sicular (2012), “Minority-Han Income Differentials in Rural China, 1988– 2002,” unpublished draft paper, University of Western Ontario, London, Ontario, Canada. Maurer-Fazio, M. (1999), “Earnings and Education in China’s Transition to a Market Economy: Survey Evidence from 1989 and 1992,” China Economic Review, 10(1), 17–40. Myers, S.L., Jr. (forthcoming), “Prosperity and Inequality: Lessons from the United States,” Chinese Academy of Social Sciences Proceedings. National Bureau of Statistics (NBS) (2009a), Zhongguo minzu tongji nianjian 2008 (China’s Ethnic Statistical Yearbook 2008), Beijing: Zhongguo tongji chubanshe. National Bureau of Statistics (NBS) (2009b), Zhongguo tongji nianjian 2009 (China Statistical Yearbook 2009), Beijing: Zhongguo tongji chubanshe. Organisation for Economic Co-operation and Development (OECD) (2010), Economic Surveys: China, Paris. Rong, X.L. and T. Shi (2001), “Inequality in Chinese Education,” Journal of Contemporary China, 10(26), 107–124. Sicular, T., X. Yue, B. Gustafsson, and S. Li (2007), “The Urban-Rural Income Gap and Inequality in China,” Review of Income and Wealth, 53(1), 93–126. Smith, J.P. and F.R. Welch (1975), “Black-White Earnings and Employment, 1960–1970,” (Report R-1666), Santa Monica, CA: The RAND Corporation. Smith, J.P. and F.R. Welch (1977), “Black-White Male Wage Ratios: 1960–70,” American Economic Review, 67(3), 323–338. Smith, J.P. and F.R. Welch (1989), “Black Economic Progress after Myrdal,” Journal of Economic Literature, 27(2), 519–564. Zang, X. and L. Li (2001), “Ethnicity and Earnings Determination in Urban China,” New Zealand Journal of Asian Studies, 3(1), 34–48. Zhang, X. and R. Kanbur (2005), “Spatial Inequality in Education and Health Care in China,” China Economic Review, 16(2), 189–204.

APPENDIX I

The 2007 Household Surveys Sampling Methods and Data Description Luo Chuliang, Li Shi, Terry Sicular, Deng Quheng, and Yue Ximing

To track the dynamics of income distribution in China, the Chinese Household Income Project (CHIP) has conducted four waves of household surveys, in 1988, 1995, 2002, and lastly 2007. These surveys were carried out as part of a collaborative research project on incomes and inequality in China organized by Chinese and international researchers, with assistance from the National Bureau of Statistics (NBS). The CHIP project participants and other researchers have analyzed the data from the first three waves and published a wide range of articles, reports, and books. Descriptions of the CHIP surveys and key findings can be found in Griffin and Zhao (1993); Riskin, Zhao, and Li (2001); and Gustafsson, Li, and Sicular (2008). This volume not only contains analyses based on the data from the fourth wave, 2007 but also uses data from the earlier waves to understand trends over time. Eichen and Zhang (1993) describe the 1988 survey, and Li et al. (2008) describe the 1995 and 2002 surveys. This Appendix provides basic information about the 2007 survey. The CHIP surveys are closely related to the NBS household survey. Li et al. (2008) discuss how the NBS household survey samples were selected. Additional details about the NBS household surveys can be found in recent NBS statistical reports and publications. All the CHIP waves contain surveys of urban and rural households. In view of the increased importance of rural-urban migration, and because the urban and rural household subsamples do not adequately cover migrants, the 2002 survey added a survey of rural-urban migrants. Thus, the 2002 CHIP survey includes three subsamples. The same procedure was adopted for the 2007 survey, which is also composed of three parts: the urban household survey, the rural household survey, and the rural-urban migrant

445

446

Luo Chuliang, Li Shi, Terry Sicular, Deng Quheng, and Yue Ximing

household survey.1 This structure reflects China’s urban-rural division and the increased number of rural individuals who have migrated into the urban areas, especially during the last two decades. There are both similarities and differences between the data collection in 2007 and that in the previous three waves, as we describe in this Appendix. In the CHIP surveys the sample sizes for the urban, rural, and ruralurban migrant surveys are not proportional to their shares in the national population. For this reason, for many analyses population-based weights are needed in order to obtain representative results. Moreover, the regional and provincial sample sizes are not proportional to their regional and provincial shares in the national population; thus, multilevel weights are needed. Appendix II provides a detailed discussion of weights for the 2002 and 2007 CHIP survey samples. In this Appendix, however, all information is reported without reweighting. The statistics reported in this Appendix are intended to describe the original survey data and may not be representative of China as a whole. The 2002 and 2007 migrant surveys include various kinds of rural-urban migrants, some of whom may also be covered in the rural and urban household surveys. The migrant samples include residents of cities with local agricultural residence registrations (hukou); this group is also found in the urban samples. The migrant samples also contain temporary and shortterm rural migrants with nonlocal hukou; this group is also included in the rural samples. Due to such overlap, analyses combining the migrant sample survey data with the urban or rural survey data may require adjustments to the samples to avoid double counting. Appendix II discusses some ways to address such double counting. In this Appendix we discuss the entire survey samples, including the types of households that may be double counted. Again, the statistics in this Appendix are intended to describe the original survey data, without modifications or adjustments.

I. Sampling and Sample Sizes Table AI.1 presents the sample sizes for the rural, urban, and migrant subsamples of the 2007 survey. The urban survey covered 10,000 households 1

The 2002 surveys were carried out by the NBS. The 2007 urban and rural surveys were conducted by the NBS, but the rural-urban migrant survey was conducted by a survey company. The 2007 survey is also a part of the larger RUMiCI (Rural-Urban Migrants in China and Indonesia) survey project. The sampling procedure and survey method for the 2007 migrant survey are described in detail in the Rural-Urban Migration in China and Indonesia Project Survey Documentation. See http://rse.anu.edu.au/rumici/ documentation.php. Accessed September 3, 2011.

The 2007 Household Surveys

447

Table AI.1. CHIP sample size for each subgroup, 2007

Individuals Households Provinces Counties/districts/cities Regions

Urban

Rural

Rural-urban migrants

29,262 10,000 16 302 4

51,847 13,000 16 287 4

8,404 4,978 9 15 4

Notes: For the urban and rural areas, the row marked “counties/districts/cities” provides the number of counties and districts covered by the survey. For migrants, this row provides the number of cities covered by the survey. Note that the sample sizes for the urban and migrant surveys reported in this table are slightly different from the sample sizes reported in Appendix II. This is due to different treatment of some duplicates found in the urban data set and because here we have dropped the small number of migrant observations for which the personal and household data sets could not be merged.

containing 29,262 individuals selected from 302 cities in sixteen provinces, whereas the rural survey covered 13,000 households containing 51,847 individuals selected from 287 counties in sixteen provinces. The migrant survey covered nearly 5,000 households containing 8,404 individuals selected from fifteen cities in nine provinces. To obtain a nationally representative sample, the provinces were selected from four distinct regions to reflect variations in economic development and geography. Beijing and Shanghai were selected to represent China’s large metropolitan cities; Liaoning, Jiangsu, Zhejiang, Fujian, and Guangdong to represent the eastern region; Shanxi, Anhui, Hebei, Henan, Hubei, and Hunan to represent the central region; and Chongqing, Sichuan, Yunnan, and Gansu to represent the western region. The provinces covered in the urban and rural surveys are almost identical, with the exception that Shanghai is included only in the urban survey, and Hebei is included only in the rural survey. The migrant household survey was conducted in fifteen cities in nine provinces that are also represented in the urban and rural surveys, including Shanghai (large metropolitan city region); Guangzhou, Shenzhen and Dongguan in Guangdong (eastern region); Nanjing and Wuxi in Jiangsu (eastern region); Hangzhou and Ningbo in Zhejiang (eastern region); Wuhan in Hubei (central region); Hefei and Bengbu in Anhui (central region); Zhengzhou and Luoyang in Henan (central region), Chongqing (western region); and Chengdu in Sichuan (western region). The majority of migrants in China are concentrated in the aforementioned cities.

448

Luo Chuliang, Li Shi, Terry Sicular, Deng Quheng, and Yue Ximing

The 2007 sample size is larger than the 2002 sample, in terms of numbers of individuals and numbers of households. Compared with the 2002 survey, the 2007 survey covers more provinces in the urban survey but fewer provinces in the rural survey. In the 2007 urban survey Shanghai, Zhejiang, Fujian, and Hunan were added and twelve provinces in the 2002 urban survey were retained – Beijing, Shanxi, Liaoning, Jiangsu, Anhui, Henan, Hubei, Guangdong, Chongqing, Sichuan, Yunnan, and Gansu. Hebei appears in the urban survey in 2007 but not in 2002. The provinces covered in the rural survey in both 2002 and 2007 are Beijing, Hebei, Shanxi, Liaoning, Jiangsu, Zhejiang, Anhui, Henan, Hubei, Hunan, Guangdong, Chongqing, Sichuan, Yunnan, and Gansu. Seven provinces, Jilin, Jiangxi, Shandong, Shaanxi, Guangxi, Guizhou, and Xinjiang are excluded from the 2007 rural survey, but Fujian is added. In the 2002 migrant household survey, the sample contains 2,000 rural-urban migrant households from provincial capitals and one or two medium-sized cities in the same provinces that are also included in the 2002 urban survey. The migrant household sample in 2007, as opposed to that in the 2002 survey, is drawn from cities where migrants are more concentrated nationally. In addition to the household and individual surveys outlined previously, in some areas surveys were also conducted at the village level to obtain relevant information about the communities where the rural households were located. Village-level variables are available for the 8,000 rural households about which we have information from the CHIP questionnaires (see Section II). Information about the village-level survey data is available on request.

II. Data from the CHIP Questionnaires versus Data Provided by the NBS The CHIP data set contains two types of data. One type was collected by the NBS as part of its annual urban and rural household surveys and then provided to the CHIP. The other type was collected through household interviews conducted using independent CHIP questionnaires. The CHIP questionnaires were designed to supplement the NBS survey data. They contain questions asking for some information that was also collected in the NBS surveys, as well as for some information on variables that are unavailable in the NBS household surveys. With respect to the migrant surveys, because the NBS does not conduct a survey of rural-urban migrants,

The 2007 Household Surveys

449

Table AI.2. Samples covered by the CHIP and NBS data (number of households)

Urban Rural Migrant

CHIP data only

CHIP and NBS data

NBS data only

(5,000)1 0 4,978

0 8,0002 0

10,000 5,000 0

Notes: 1. As discussed in the text, data are available for 5,000 urban households from the CHIP questionnaire, but there are no matching NBS data. Because the data for these 5,000 households are incomplete, we do not include them in the data descriptions and tables that follow. 2. Partial NBS data are available for these 8,000 households. See the text for further discussion.

information in the migrant data set is based entirely on interviews using the independent CHIP questionnaires. Unfortunately, not all types of information are available for all households and individuals. The types of data collected for each subsample are summarized in Table AI.2. For the 2007 urban survey, the NBS provides comprehensive data for 10,000 households. These 10,000 households, however, did not answer the CHIP questionnaires. We note that the CHIP questionnaires were used for an additional sample of 5,000 urban households that we do not include in our description of the 2007 survey sample here. The additional 5,000 households are in nine provinces (Shanghai, Jiangsu, Zhejiang, Anhui, Henan, Hubei, Guangdong, Chongqing, and Sichuan). They had been part of the NBS urban household survey sample in 2006, but due to sample rotation, they were not retained in the 2007 survey. For these additional 5,000 households NBS data are not available. As the CHIP urban questionnaire was designed to be matched with the data provided by the NBS, information on these 5,000 households is incomplete and of limited use for analysis of incomes, inequality, and poverty. Consequently, in this Appendix we limit our discussion to the 10,000 households for which NBS data are available. Most of the chapters in this volume that analyze the urban survey use only the 10,000 households for which we have NBS data. For the 2007 rural survey, data from the CHIP rural questionnaire are available for 8,000 households. For these 8,000 households, income and expenditure data are also available from the NBS household survey. The NBS also provided its household survey data for an additional 5,000 households, but these additional households were not interviewed using the CHIP

450

Luo Chuliang, Li Shi, Terry Sicular, Deng Quheng, and Yue Ximing

questionnaires. In other words, CHIP data and partial NBS data are available for 8,000 households, and NBS data but no CHIP data are available for an additional 5,000 households. Because comparable data on key household characteristics as well as on household incomes and expenditures are available for all 13,000 households, these two rural subsamples can be combined for analyses of incomes, inequality, and poverty. In this Appendix we discuss the full sample of 13,000 households. For the migrant sample only CHIP data and no NBS data are available. The CHIP migrant questionnaire was designed accordingly, with questions that collect information comparable to that in the urban and rural household surveys as well as additional information on topics of special relevance to migrants and migration. We note, however, that all the data in the CHIP migrant survey, including data on incomes and expenditures, are based on recall questions. For the urban and rural data sets, income and expenditure data are diary-based. In principle, diary-based income data are more accurate than are recall data, but recall data can still be useful. Analysts who use the migrant data should be aware of this difference between the data for the migrant sample and for the urban and rural samples.

III. Characteristics of the 2007 Urban Survey The distribution of households and individuals by province in the 2007 urban survey sample of 10,000 households for which NBS data are available is reported in Table A1.3. The sample was designed to draw more households from the more populous provinces. The sample size for each province or region, however, was not strictly proportional to its actual population, so that for some analyses it may be necessary to reweight the sample in order to obtain results that are representative (see Appendix II). The gender composition in the 2007 urban data, by province, is presented in Table AI.4. Slightly more female individuals than male individuals are contained in the overall urban sample. The gender compositions are almost identical in the 2002 and 2007 surveys, 100:102.6 (male:female) in the 2002 data (Li et al. 2008: 348), and 100:102.1 in the 2007 data. Within provinces, however, there are some differences between the two years. For example, in Beijing the percentage of females relative to males is 101 percent in the 2002 survey but 95.5 percent in the 2007 survey. Although we mention these differences, we do not attempt to explain them in detail. Table AI.5 shows the distribution of households by household size in the 2007 urban survey. More than 57 percent of the sampled households were composed of three members, reflecting the enforcement of the one-child

Table AI.3. Distribution of households in the 2007 urban sample, by province

Province Beijing Shanxi Liaoning Shanghai Jiangsu Zhejiang Anhui Fujian Henan Hunan Hubei Guangdong Chongqing Sichuan Yunnan Gansu

Number of counties/districts

Number of households

Number of individuals

18 24 52 12 12 16 11 32 19 23 8 23 13 9 17 13

800 600 800 500 600 600 550 800 650 800 400 700 400 600 600 600

2,289 1,771 2,244 1,519 1,669 1,653 1,572 2,443 1,893 1,160 2,327 2,268 1,186 1,740 1,794 1,734

Note: Here, and in the following tables, statistics on the urban sample are for the 10,000 urban households for which NBS data are available. The additional 5,000 households for which we only have data from the CHIP questionnaire are not included.

Table AI.4. Gender composition of individuals in the 2007 urban sample, by province

Province Total Beijing Shanxi Liaoning Shanghai Jiangsu Zhejiang Anhui Fujian Henan Hubei Hunan Guangdong Chongqing Sichuan Yunnan Gansu

Male

Female

Females as a percentage of males

14,478 1,171 902 1,089 751 828 815 782 1,195 925 579 1,160 1,110 589 849 876 857

14,784 1,118 869 1,155 768 841 838 790 1,248 968 581 1,167 1,158 597 891 918 877

102.1 95.5 96.3 106.1 102.3 101.6 102.8 101.0 104.4 104.6 100.3 100.6 104.3 101.4 104.9 104.8 102.3

452

Luo Chuliang, Li Shi, Terry Sicular, Deng Quheng, and Yue Ximing

Table AI.5. Distribution of households in the 2007 urban sample, by household size and province (%) Number of household members Province Total Beijing Shanxi Liaoning Shanghai Jiangsu Zhejiang Anhui Fujian Henan Hubei Hunan Guangdong Chongqing Sichuan Yunnan Gansu

1

2

3

4

5

6

≥7

All

1.7 0.0 1.7 2.1 0.4 3.2 3.5 0.6 1.1 1.7 2.3 2.5 0.3 1.0 1.2 5.0 1.5

25.1 22.5 25.7 34.1 16.4 35.5 28.3 25.6 21.3 28.2 23.5 26.1 10.1 25.8 29.8 24.7 24.0

57.7 70.6 53.7 50.5 69.8 47.2 60.5 63.6 57.8 53.4 61.0 55.4 65.0 54.8 52.3 47.0 61.5

10.3 5.1 14.5 8.0 6.4 9.2 5.0 7.8 11.8 11.4 8.8 10.6 15.9 13.0 11.7 15.3 10.2

4.6 1.8 3.8 4.9 6.4 4.2 2.2 2.4 7.5 4.8 4.3 4.8 7.6 5.3 4.5 6.3 2.7

0.5 0.0 0.7 0.4 0.6 0.8 0.5 0.0 0.4 0.6 0.3 0.6 0.9 0.3 0.5 1.2 0.2

0.1 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.3 0.0 0.0 0.0 0.3 0.0 0.0 0.5 0.0

100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0

policy in urban China. The proportion of three-member households decreased by 4 percentage points, from 61.6 percent in 2002 to 57.7 percent in 2007, whereas the proportion of small-size households (fewer than three members) increased slightly. More than 80 percent of the urban sample in 2007 consists of households with two or three members, suggesting that “nuclear households” dominate in the urban sample. The proportion of twoand three-member households is generally higher in the more developed provinces. The average household size in the NBS national urban household survey was 2.91 in 2007, very close to 2.93, the average (unweighted) household size calculated from our urban survey. Table AI.6 reports the distribution of individuals among different age groups in the 2007 urban survey. Figure AI.1 compares the age-gender profiles in the 2002 and 2007 urban surveys. In 2007, individuals between the ages of twenty and sixty account for 68.8 percent of all sampled individuals. The percentages vary among provinces from 65 percent to 76 percent. In Beijing and Shanghai, for example, the percentages of individuals between the ages of twenty to sixty are higher than that in other provinces. Compared with the 2002 urban data, in 2007 a higher percentage of individuals are in older cohorts and a lower percentage in younger cohorts. This reflects the aging of China’s urban population.

The 2007 Household Surveys

453

Table AI.6. Distribution of individuals in the 2007 urban sample, by age group and province (%) Age group Province

0–5

6–10

11–20

21–30

31–40

41–50

51–60

61–70

>70

Total

Total Beijing Shanxi Liaoning Shanghai Jiangsu Zhejiang Anhui Fujian Henan Hubei Hunan Guangdong Chongqing Sichuan Yunnan Gansu

2.7 1.2 2.1 1.7 3.1 2.5 2.6 2.0 3.2 3.0 2.8 3.4 4.7 1.9 2.2 4.2 2.8

4.2 2.1 5.9 3.2 2.3 3.2 4.4 3.4 4.2 5.2 2.1 5.8 6.5 4.1 4.4 3.7 5.8

11.9 10.6 16.0 10.3 9.6 9.4 11.4 14.1 13.1 12.4 10.3 10.7 13.1 10.5 10.9 12.4 14.0

10.7 14.2 8.4 10.7 15.0 10.3 11.1 10.0 8.2 9.6 13.9 9.6 11.2 11.2 9.2 11.4 8.9

17.8 11.3 23.6 14.7 12.1 14.9 17.9 15.0 20.0 21.5 12.9 19.3 23.6 18.0 17.2 18.7 21.2

21.8 24.4 20.4 21.2 20.4 18.6 23.4 28.9 22.3 18.3 22.9 20.1 21.6 19.1 22.0 20.8 23.5

18.5 25.7 13.5 22.5 25.9 21.0 17.4 15.7 16.0 16.4 22.2 17.9 12.8 22.4 19.3 16.0 14.9

8.1 7.7 6.8 11.1 7.6 11.2 7.7 6.4 8.5 9.1 7.8 8.3 4.4 7.9 9.6 9.1 5.9

4.3 2.9 3.2 4.7 4.2 8.9 4.1 4.6 4.5 4.5 5.2 5.0 2.2 4.8 5.2 3.6 2.9

100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0

4% 3%

female 2002

female 2007

male 2002

male 2007

2% 1%

–1% –2% –3% –4%

Figure AI.1. Age-Gender Profiles, Urban.

80 84 88 92 96 106

0 4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 64 68 72 76

0%

Age

454

Luo Chuliang, Li Shi, Terry Sicular, Deng Quheng, and Yue Ximing

Table AI.7. Educational attainment of individuals over the age of fifteen in the 2007 urban sample, by province (%)

Province Total Beijing Shanxi Liaoning Shanghai Jiangsu Zhejiang Anhui Fujian Henan Hubei Hunan Guangdong Chongqing Sichuan Yunnan Gansu

Less than Junior Senior primary Primary middle middle Professional Junior College college and above Total school school school school school* 1.8 0.5 0.8 1.5 1.0 2.9 1.9 1.3 2.2 1.4 1.3 1.8 1.0 1.6 2.7 4.5 2.6

6.0 2.2 4.6 5.3 2.6 6.7 9.5 4.6 7.7 3.5 3.3 8.1 5.9 7.2 6.9 13.6 4.6

24.5 19.0 31.8 31.8 24.9 25.7 31.7 21.8 25.8 19.6 19.3 24.0 16.3 25.4 24.8 26.8 23.8

25.7 23.4 25.1 23.1 29.1 26.3 22.7 32.0 26.5 26.7 26.6 25.3 30.4 26.4 24.2 16.0 30.1

11.0 11.6 11.8 10.2 11.4 9.5 6.4 10.0 14.6 11.6 12.0 10.8 10.7 8.7 9.4 14.4 11.3

18.9 22.5 15.2 19.1 16.9 14.9 17.2 20.1 14.8 24.1 21.7 18.3 23.0 20.2 20.9 15.7 17.1

12.1 20.8 10.8 9.1 14.1 14.0 10.6 10.2 8.4 13.1 15.7 11.7 12.8 10.6 11.0 9.0 10.4

100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0

* Here, and elsewhere, “professional school” refers to zhongzhuan, a technical/occupational stream of middle school.

Table AI.7 reports the distribution of individuals over the age of fifteen in the 2007 urban survey by educational attainment. The ratio of those with “junior college” and “college and above” educations increased from 16.7 percent and 8.7 percent in 2002 to 18.9 percent and 12.1 percent in 2007, respectively. This increase in postsecondary attainment reflects the expansion in the availability of college education since 1999. In Beijing, individuals over the age of fifteen with a postsecondary education accounted for more than 43 percent of all individuals over the age of fifteen; this is much higher than the percentages in the other provinces. In general, educational attainment is higher in the more developed provinces.

IV. Characteristics of the Rural Survey Table AI.8 presents the distribution of households and individuals in the 2007 rural sample by province. In order to capture the fact that the population in the rural areas is larger than that in the urban areas and that rural China is more heterogeneous, the rural survey sampled more households and individuals than the urban survey. The number of households

The 2007 Household Surveys

455

Table AI.8. Distribution of households in the 2007 rural sample, by province

Province Beijing Hebei Shanxi Liaoning Jiangsu Zhejiang Anhui Fujian Henan Hunan Hubei Guangdong Chongqing Sichuan Yunnan Gansu

Number of counties

Number of households

Number of individuals

13 5 35 24 10 10 9 26 10 37 10 12 5 11 40 30

500 500 700 800 1,000 1,000 900 800 1,000 800 1,000 1,000 500 1,100 700 700

1,717 1,826 2,777 2,694 3,714 3,426 3,683 3,435 4,089 3,168 4,026 5,082 1,782 4,163 3,015 3,250

selected in the survey design is based on the population in each province, with the more populous provinces assigned more households. Similar to the urban survey, however, the provincial sample sizes are not consistent with the actual provincial distributions of the population. Consequently, depending on the question being analyzed, reweighting may be required to obtain representative results (see Appendix II). Although fewer provinces were surveyed in 2007 than in 2002, in 2007 the sample sizes in terms of the number of households and individuals increased and more counties were drawn from within the provinces. Table AI.9 provides the gender composition of the 2007 rural survey sample, both overall and by province. In contrast to the urban survey (see Table AI.4), the rural survey contains fewer females than males, both overall and in each individual province. This difference reflects the strong preference for males in rural China. The gender composition changed only slightly between the 2002 and 2007 rural surveys. The ratio of females to males was 92.2 percent in 2002 and 93.2 percent in 2007. Table AI.10 shows the distribution of households by household size in the 2007 rural survey. Households with four members account for 30 percent of all the sampled households, and three- and four-member households together account for 56 percent of all sampled households. The larger household sizes in rural China reflect rural-urban differences in implementation

Table AI.9. Gender composition of individuals in the 2007 rural sample, by province

Province Total Beijing Hebei Shanxi Liaoning Jiangsu Zhejiang Anhui Fujian Henan Hubei Hunan Guangdong Chongqing Sichuan Yunnan Gansu

Male

Female

Females as a percentage of males

26,838 859 930 1,436 1,386 1,901 1,782 1,932 1,792 2,119 2,080 1,651 2,640 928 2,148 1,581 1,673

25,009 858 896 1,341 1,308 1,813 1,644 1,751 1,643 1,970 1,946 1,517 2,442 854 2,015 1,434 1,577

93.2 99.9 96.3 93.4 94.4 95.4 92.3 90.6 91.7 93.0 93.6 91.9 92.5 92.0 93.8 90.7 94.3

Table AI.10. Distribution of households in the 2007 rural sample, by household size and province (%) Number of household members Province Total Beijing Hebei Shanxi Liaoning Jiangsu Zhejiang Anhui Fujian Henan Hubei Hunan Guangdong Chongqing Sichuan Yunnan Gansu

1

2

3

4

5

6

≥7

All

0.3 0.6 0.4 0.3 0.6 0.2 0.5 0.7 0.0 0.4 0.2 0.0 0.3 0.2 0.5 0.1 0.0

12.0 12.2 19.6 12.6 21.8 15.3 18.4 9.7 5.6 10.6 9.9 12.3 2.4 20.6 14.7 6.9 4.1

25.8 48.4 27.2 19.3 38.4 34.1 40.6 21.6 20.9 16.9 25.1 23.5 8.0 33.2 32.9 18.1 12.6

30.2 24.2 29.6 37.9 23.3 23.5 24.9 33.4 35.9 39.0 32.8 34.8 26.8 24.6 23.1 38.0 31.4

19.1 11.8 15.6 20.1 11.9 19.2 10.9 22.0 21.6 20.2 20.2 18.3 28.3 14.8 17.9 21.3 27.1

9.1 2.4 5.8 7.9 3.8 6.3 3.9 9.4 10.0 11.0 9.0 8.9 19.1 4.6 9.1 10.1 18.3

3.5 0.4 1.8 2.0 0.4 1.4 0.8 3.2 6.0 1.9 2.8 2.4 15.1 2.0 1.8 5.4 6.4

100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0

The 2007 Household Surveys

457

Table AI.11. Distribution of individuals in the 2007 rural sample, by age group and province (%) Age group Province

0–5

6–10

11–20

21–30

31–40

41–50

51–60

61–70

>70

Total

Total Beijing Hebei Shanxi Liaoning Jiangsu Zhejiang Anhui Fujian Henan Hubei Hunan Guangdong Chongqing Sichuan Yunnan Gansu

4.6 1.9 4.8 3.1 3.0 4.2 3.7 5.7 4.4 6.0 5.0 4.6 5.5 4.8 4.5 5.3 4.4

4.7 2.6 4.6 4.6 3.4 4.3 4.6 4.2 4.3 5.9 2.8 4.4 6.2 3.5 5.1 6.4 5.6

18.6 17.3 16.5 22.5 14.9 17.2 13.1 20.6 18.9 21.7 17.9 18.2 22.8 12.1 14.8 19.4 23.5

15.3 13.3 13.2 15.2 13.1 12.1 13.4 15.5 18.5 14.4 19.2 16.8 19.2 13.3 12.2 17.0 13.1

15.3 15.8 14.7 14.5 14.3 18.6 14.1 15.4 16.7 16.3 12.4 13.3 11.5 15.8 18.8 16.1 18.0

17.5 24.3 18.1 19.0 21.0 18.4 21.3 16.1 17.4 15.6 19.5 16.5 14.5 16.0 14.7 16.3 17.0

15.3 16.2 19.7 14.5 19.9 15.5 19.7 14.0 12.5 12.9 16.9 17.1 12.8 21.0 18.7 10.2 9.8

5.9 6.2 6.0 4.9 6.6 6.7 7.0 5.7 4.7 5.2 4.2 6.3 4.4 9.2 8.0 5.8 6.1

2.9 2.6 2.6 1.7 3.8 3.1 3.2 2.7 2.7 2.2 2.2 3.0 3.2 4.4 3.4 3.5 2.6

100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0

of population planning policies. In most counties rural couples are allowed to give birth to a second child. The average (unweighted) household size in the 2007 rural survey was 4.0 members (as calculated from Table AI.1), which is close to the 4.03 members per household officially reported by the NBS in its annual rural household survey. The distribution of individuals among different age groups in the 2007 rural survey is presented in Table AI.11 and Figure AI.2. The rural sample is younger than the urban sample. The proportions of individuals between the ages of zero and five and between the ages of eleven and twenty are 4.6 percent and 18.6 percent, respectively, which is much higher than the proportions in the urban data. However, there are relatively fewer workingage individuals in the rural sample than in the urban sample. Individuals between the ages of twenty and sixty account for 63.4 percent of the rural sample, 5 percentage points less than in the urban sample, possibly reflecting that some rural laborers who had migrated to the urban areas were excluded from the rural survey. The treatment of individuals who were away from their homes at the time of the survey is explained in more detail in Appendix II. Similar to the urban sample, between 2002 and 2007 the age distribution

458

Luo Chuliang, Li Shi, Terry Sicular, Deng Quheng, and Yue Ximing

3%

2%

female 2002

female 2007

male 2002

male 2007

1%

0 4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 64 68 72 76 80 84 88 92 96 100

0%

Age

–1%

–2%

–3%

Figure AI.2. Age-Gender Profiles, Rural.

of the rural sample shifted upward, reflecting the aging of the population (Figure AI.2). Table AI.12 gives the distribution of individuals over the age of fifteen in the 2007 rural sample by level of educational attainment. As expected, educational attainment in the rural areas is generally lower than that in the urban areas. The majority of rural adults have attended primary or junior middle school; these two groups account for 37.5 percent and 43.0 percent of all the rural sampled adults, respectively. Compared with the 2002 rural survey, the share of adults who had attended primary school increased by nearly 8 percentage points, whereas the proportion who had attended junior middle school decreased by 2 percentage points.

V. Characteristics of the Rural-Urban Migrant Survey Since the mid-1990s, rural laborers have increasingly migrated to the urban areas to seek employment. Neither the rural nor the urban household survey regularly conducted by the NBS adequately captures the rural-urban migrants. The NBS rural household survey includes some migrants, as it counts as household members individuals who are present in the household for up to six months or who are away from the household but still maintain strong economic ties (see Appendix II). Historically, the NBS urban household survey only covered households with a local hukou, but in recent years, the NBS has deliberately expanded its urban sample to include households without an urban hukou; still, rural-urban migrants are generally underrepresented in the NBS data.

The 2007 Household Surveys

459

Table AI.12. Educational attainment of individuals over the age of fifteen in the 2007 rural sample, by province (%)

Province Total Beijing Hebei Shanxi Liaoning Jiangsu Zhejiang Anhui Fujian Henan Hubei Hunan Guangdong Chongqing Sichuan Yunnan Gansu

Less than primary school

Primary school

Junior middle school

Senior middle school

Professional school

Junior college and above

Total

3.8 3.3 0.0 3.7 3.9 0.0 0.0 0.0 9.7 0.0 0.0 4.8 0.0 0.0 0.0 18.4 18.6

37.5 7.7 40.4 20.2 21.4 44.3 49.3 39.8 23.9 43.8 47.9 29.5 42.1 57.2 55.3 35.0 25.5

43.0 44.7 48.5 54.7 55.7 41.0 35.7 51.2 40.8 44.0 41.4 45.5 46.0 33.9 36.1 35.7 35.3

11.3 20.3 9.0 16.1 10.5 11.1 11.0 6.9 16.5 9.8 8.5 15.5 10.1 7.6 6.9 7.7 15.8

2.4 11.0 1.1 2.5 2.9 2.2 2.3 1.1 4.5 2.1 1.3 2.8 1.4 0.9 0.9 2.1 2.7

2.1 13.1 1.0 2.8 5.6 1.3 1.7 1.1 4.5 0.4 0.8 1.9 0.4 0.3 0.9 1.1 2.1

100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0

Our 2002 survey includes a sample of rural-urban migrant households within the same provinces as the urban sample. In the absence of a sampling frame specifically designed to capture migrants, we relied on residential neighborhood committees, grassroots organizations in urban China, to identify and select the migrant households for our survey. Details about selection of the migrant sample for the 2002 survey are discussed in Li et al. (2008). The 2007 migrant survey covers nine of the sixteen provinces covered in the urban survey. Similar to the urban survey, the provincial sample sizes are not consistent with the actual distribution of the migrant population among provinces. Therefore, depending on the question being analyzed, reweighting may be required to obtain representative results (see Appendix II). In 2007 the migrant survey was conducted under the auspices of the Rural-Urban Migration in China (RUMiC) project, with assistance from Datasea, a company specializing in market research. The researchers made great efforts to construct the sampling frame, as explained at length in Kong (2010). Here, we only briefly describe the process. First, each sample city

460

Luo Chuliang, Li Shi, Terry Sicular, Deng Quheng, and Yue Ximing

selected was divided into equal-sized blocks averaging 0.25 square kilometers in size, based on up-to-date, equal-scale maps. Second, 10 percent of the blocks were randomly selected. The enumerators listed in a given order all the workplaces within each block and maintained a record of the number of migrants working in each workplace. These records were considered to be the sampling frame, and from this frame the researchers randomly drew the individuals for the migrant survey. The survey questionnaire then asked the migrant respondents for information about their living situations in the cities, including their household situation in the cities, that is, whether they lived in the city with other people from the household and shared income and expenditures. The migrant sample includes individuals with agricultural hukou who were not living in the location of their hukou registrations, including both temporary and long-term migrants.2 Because temporary migrants are also included in the rural sample as household members, there is a potential for double counting. Consequently, several of the analyses in this volume drop the temporary migrants and only use the subsample of long-term, stable migrants. The migrant sample also includes some individuals with agricultural hukou in the local urban area. Such individuals are included in the urban sample as well, also creating the potential for double counting. Appendix II provides a detailed discussion of how to handle potential double counting between the migrant and the rural and urban surveys. Tables AI.13 through AI.16 report information from the rural-urban migrant sample, similar to the information reported earlier for the urban and rural samples. The statistics in these tables are calculated from the full migrant sample, including both temporary and long-term migrants and including individuals with local agricultural hukou. Table AI.13 shows the distribution of migrant households and individuals among cities. In the table one can see that the average household size in the rural-urban migrant sample is much smaller than that in the urban and rural household samples. The majority of migrant households are composed of a single person. Looking at the gender composition in Table AI.14, we find that the migrant sample contains more males than the other two samples, because married women typically return to their places of origin. Not surprisingly, working-age individuals comprise a higher share of the migrant sample. 2

In 2007 the hukou reform, which removed the agricultural/nonagricultural distinction, had been implemented in only a few provinces. Most households still had either agricultural or nonagricultural hukou.

Table AI.13. Distribution of households and individuals in the 2007 rural-urban migrant sample, by city

City Total Guangzhou Dongguan Shenzhen Zhengzhou Luoyang Hefei Bengbu Chongqing Shanghai Nanjing Wuxi Hangzhou Ningbo Wuhan Chengdu

Number of individuals

Number of households

Average household size

8,404 617 427 365 658 366 705 428 682 852 611 331 639 331 692 700

4,978 400 272 302 350 200 350 200 400 503 400 200 400 200 400 401

1.7 1.5 1.6 1.2 1.9 1.8 2.0 2.1 1.7 1.7 1.5 1.7 1.6 1.7 1.7 1.7

Number of household members (%) 1 60.5 66.0 66.9 84.4 56.9 59.5 47.4 43.5 56.3 58.3 65.3 63.5 58.0 61.0 61.5 57.9

2

3

≥4

Total

18.9 18.5 17.3 12.3 14.9 12.5 20.3 18.5 25.3 20.5 22.8 12.0 28.5 20.5 12.8 18.7

13.7 12.3 9.9 1.7 14.9 17.5 19.4 22.0 12.0 15.7 7.0 20.0 10.5 11.5 18.8 16.7

6.9 3.3 5.9 1.7 13.4 10.5 12.9 16.0 6.5 5.6 5.0 4.5 3.0 7.0 7.0 6.7

100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0

Table AI.14. Gender composition of individuals in the 2007 rural-urban migrant sample, by city

Province

Male

Female

Females as a percentage of males

Total Guangzhou Dongguan Shenzhen Zhengzhou Luoyang Hefei Bengbu Chongqing Shanghai Nanjing Wuxi Hangzhou Ningbo Wuhan Chengdu

4,777 343 257 235 386 220 416 251 367 484 342 153 381 184 378 380

3,627 274 170 130 272 146 289 177 315 368 269 178 258 147 314 320

75.9 79.9 66.1 55.3 70.5 66.4 69.5 70.5 85.8 76.0 78.7 116.3 67.7 79.9 83.1 84.2

462

Luo Chuliang, Li Shi, Terry Sicular, Deng Quheng, and Yue Ximing

Table AI.15. Distribution of individuals in the 2007 rural-urban migrant sample, by age group and city (%) Age group City

0–5

6–10

11–20

21–30

31–40

41–50

51–60

61–70

>70

Total

Total Guangzhou Dongguan Shenzhen Zhengzhou Luoyang Hefei Bengbu Chongqing Shanghai Nanjing Wuxi Hangzhou Ningbo Wuhan Chengdu

4.5 3.7 4.9 1.9 8.5 7.7 5.0 6.3 2.8 3.6 2.6 4.5 3.1 5.1 5.2 4.0

3.3 2.6 2.8 0.3 4.3 4.4 4.3 6.1 3.2 2.9 1.6 3.0 2.7 2.7 4.1 4.0

17.2 14.3 11.9 20.6 19.6 24.0 21.7 20.8 14.8 14.3 19.8 16.3 13.0 15.1 17.5 17.4

32.5 36.1 42.6 51.0 29.5 25.4 25.3 24.3 30.1 32.8 36.2 38.1 36.8 27.8 34.5 25.3

24.9 25.9 24.6 15.6 20.5 22.1 26.4 29.0 27.9 27.4 21.1 16.9 26.8 26.9 21.7 32.4

12.8 14.3 8.7 7.4 9.4 12.0 11.8 8.9 16.7 16.7 11.5 17.8 13.3 15.4 14.0 10.9

3.8 2.6 3.3 3.3 5.5 2.7 5.3 3.3 3.4 2.1 5.7 3.3 3.8 5.7 2.5 4.1

0.8 0.5 0.9 0.0 2.4 0.8 0.4 1.2 0.7 0.2 1.3 0.0 0.6 1.2 0.6 1.1

0.2 0.0 0.2 0.0 0.3 0.8 0.0 0.2 0.4 0.0 0.2 0.0 0.0 0.0 0.0 0.7

100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0

Table AI.16. Educational attainment of individuals over the age of fifteen in the 2007 rural-urban migrant sample, by city (%)

city Total Guangzhou Dongguan Shenzhen Zhengzhou Luoyang Hefei Bengbu Chongqing Shanghai Nanjing Wuxi Hangzhou Ningbo Wuhan Chengdu

Less than Junior Senior primary Primary middle middle Professional Junior College school school school school school college and above Total 2.5 1.6 1.1 0.6 3.2 1.4 6.0 6.6 1.5 2.2 3.4 0.7 2.2 3.6 0.9 2.5

13.3 8.4 10.2 7.1 14.5 7.1 18.0 18.2 17.0 13.9 9.5 6.9 15.2 25.8 8.2 17.2

55.3 60.0 57.3 49.1 46.2 62.6 54.1 57.6 54.1 57.6 54.2 75.9 49.0 46.4 60.0 52.7

19.8 22.2 23.3 26.9 18.3 18.2 14.5 11.9 20.6 18.0 23.5 12.1 25.7 19.2 19.3 19.3

5.2 3.6 4.2 10.9 10.3 6.7 3.7 3.6 3.6 4.9 5.7 2.8 2.0 4.0 8.4 5.1

3.3 3.2 3.4 5.1 7.0 2.7 3.4 1.5 2.9 2.2 3.2 0.7 5.4 0.7 2.9 3.0

0.6 0.9 0.5 0.3 0.6 1.4 0.4 0.6 0.3 1.2 0.5 1.0 0.5 0.3 0.5 0.2

100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0

The 2007 Household Surveys

463

6%

4%

female 2002

female 2007

male 2002

male 2007

2% 0%

Age 1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 75 82

–2%

–4%

–6%

Figure AI.3. Age-Gender Profile, Migrants.

The age distribution in the 2002 migrant sample is dominated by individuals between the ages of twenty-five and forty (Figure AI.3). This probably reflects the sampling method, which in 2002 disproportionately captured longer-term migrants who had settled in urban neighborhoods. In 2007 the migrant sampling method captured more young migrants between the ages of seventeen and twenty-five. Adjustments of the migrant samples, as discussed in Appendix II through weighting and eliminating the short-term migrants, will reduce this discrepancy. Educational attainment in the 2007 sample of migrants over the age of fifteen is shown in Table AI.16. Most of the migrant adults have only a junior middle-school education, which is similar to the findings from the rural sample and reflects the lower educational attainment among migrants compared to their urban counterparts. References Eichen, M. and M. Zhang (1993), “Annex: The 1988 Household Sample Survey: Data Description and Availability,” in K. Griffin and R. Zhao, eds., The Distribution of Income in China, 331–346, New York: St. Martin’s Press. Griffin, K. and R. Zhao, eds. (1993), The Distribution of Income in China, Basingstoke: Macmillan. Gustafsson, B., S. Li, and T. Sicular, eds. (2008), Inequality and Public Policy in China, New York: Cambridge University Press. Kong, Sherry Tao (2010), “Rural-Urban Migration in China: Survey Design and Implementation,” in X. Meng, C. Manning, S. Li, and T. N. Effendi, eds., The Great Migration:

464

Luo Chuliang, Li Shi, Terry Sicular, Deng Quheng, and Yue Ximing

Rural-Urban Migration in China and Indonesia, 135–150, Northampton, MA: Edward Elgar. Li, S., Luo, C., Z. Wei, and X. Yue (2008), “Appendix: The 1995 and 2002 Household Surveys: Sampling Methods and Data Description,” in B. Gustafsson, S. Li, and T. Sicular, eds., Inequality and Public Policy in China, 337–353, New York: Cambridge University Press. Riskin, C., R. Zhao, and S. Li, eds. (2001), China’s Retreat from Equality: Income Distribution and Economic Transition, Armonk, NY: M.E. Sharpe.

APPENDIX II

The 2002 and 2007 CHIP Surveys Sampling, Weights, and Combining the Urban, Rural, and Migrant Samples Song Jin, Terry Sicular, and Yue Ximing I. General Remarks The China Household Income Project (CHIP) data sets consist of urban, rural, and, for 2002 and 2007, rural-urban migrant samples. The sizes of these samples are not proportional to their shares in the Chinese national population. Also, their regional distributions differ from those in the population. Consequently, weights are needed in order to make the samples nationally representative. In this Appendix we discuss the calculation of sample weights that can be used for analysis of the 2002 and 2007 CHIP data. We calculate these weights using data provided by the National Bureau of Statistics (NBS) from the 2000 census and the 2005 1 percent population sample survey, hereafter called the “2005 mini census.” The census and mini census are the most complete available accountings of China’s population. Our sample weights are designed to reflect population shares in the census and the mini census. We begin with a discussion of the CHIP sampling design and its implications for the calculation of weights (Section II). The calculation of weights requires data on population shares by geographic location and by urban, rural, and migrant classification, which we obtain using data from the 2000 census and the 2005 mini census. Section III discusses the census and minicensus data that we use for this purpose. In order to construct and apply the weights consistently, we must classify the location of residence for all The need for careful attention to weights was raised by Samuel L. Myers, Jr., Ding Sai, and Li Shi in “Sample Weights and the Analysis of Per Capita Income: The Case of CHIPs,” presented at the CHIP workshop in May 2009. This Appendix builds upon their work. We thank Li Shi for contributing key ideas and for making available the subsamples of the 2000 census and the 2005 1 percent population sample survey for use in calculating the weights. This work was supported in part by the Roy Wilkins Center for Human Relations and Social Justice, Hubert H. Humphrey Institute of Public Affairs, University of Minnesota.

465

466

Song Jin, Terry Sicular, and Yue Ximing

individuals and households and make sure that there is no double counting. The classification of location is discussed in Section IV. The last section of this Appendix raises some suggestions for implementation of the weights in the analysis of the data.

II. Calculation of Weights In the CHIP surveys some groups are oversampled and others are undersampled relative to their shares in the national population. Here we discuss the construction of weights that can be used to adjust the CHIP samples so that they reflect selection probabilities from the national population. In past analyses of the CHIP data, a weight adjustment was made only for the rural and urban dimensions. In 2002, for instance, according to the National Bureau of Statistics (NBS) population data, China’s rural population was 782.4 million and the urban population was 502.1 million, implying that rural and urban shares in the total population were 60.91 percent and 39.09 percent, respectively. The 2002 CHIP urban and rural sample shares, however, were 64.78 percent and 35.22 percent, respectively, so that the rural population was oversampled and the urban sample was undersampled. The use of rural-urban weights with the 2002 CHIP data was intended to adjust the shares of the urban and rural samples so that they were identical to the shares of the urban and rural populations in China’s national population. In light of questions raised by the project participants and following extensive discussions, we concluded that the sample weights should reflect not only the rural and urban population shares, but also the population shares of the major regions of China. This conclusion was based on the principle that weights should be determined in light of the approach used to construct the CHIP samples. The CHIP urban and rural sampling methods were designed to represent the conditions in four regions of China – coastal, central, western, and a separate category for large municipalities with provincial status.1 Table AII.1 provides a list of the provinces and their regional classifications for all rounds of the CHIP survey from 1988 1

See Li Shi, Luo Chuliang, Wei Zhong, and Yue Ximing, “Appendix: The 1995 and 2002 Household Surveys: Sampling Methods and Data Description,” in B.A. Gustafsson, S. Li, and T. Sicular, eds., Inequality and Public Policy in China (New York: Cambridge University Press, 2008), for an explanation of the sample selection. The geographic regions used to construct the CHIP sample frame are (1) large municipalities with provincial status (Beijing, Tianjin, and Shanghai are treated together as a separate geographic region; Chongqing is treated as part of western China for consistency with earlier rounds of the survey, when it was included in Sichuan), (2) coastal China (Hebei, Liaoning, Jiangsu,

Table AII.1. Provinces and their regional classifications in the CHIP samples, 1988 through 2007 2007 Province

467

Beijing Tianjin Shanghai Hebei Liaoning Jiangsu Zhejiang Fujian Shandong Guangdong Hainan Shanxi Jilin Helongjiang Anhui Jiangxi Henan Hubei Hunan Inner Mongolia Guangxi

1988

1995

2002

Province code

Region

Rural

Urban

Rural

Urban

Rural

Urban

Migrant

11 12 31 13 21 32 33 35 37 44 46 14 22 23 34 36 41 42 43 15

1 1 1 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 3 4

* * * * * * * * * * * * * * * * * * * *

*

*

*

*

*

*

* *

* * * *

* *

* * * *

45

4

*

*

* *

*

* *

Rural CHIP

Urban CHIP

Migrant CHIP

*

*

* *

* *

* *

* *

*

*

* *

*

*

*

*

*

*

*

*

*

*

*

*

*

* *

*

* *

*

*

*

* * * * *

*

* * * * *

*

*

*

*

*

* *

* *

* *

* *

* *

* *

Urban NBS

*

*

* *

Rural NBS

* (continued)

Table AII.1 (continued) 2007 Province

468

Chongqing Sichuan Guizhou Yunnan Tibet Shaanxi Gansu Qinghai Ningxia Xinjiang

1988

Province code

Region

50 51 52 53 54 61 62 63 64 65

4 4 4 4 4 4 4 4 4 4

Rural * * * * * * *

1995

2002

Urban

Rural

Urban

Rural

Urban

Migrant

Rural CHIP

Urban CHIP

Migrant CHIP

Rural NBS

Urban NBS

(*) *

* *

* *

* *

* *

*

* * * *

* *

*

(*) * * *

*

*

*

*

*

* *

*

* *

*

*

*

*

*

Notes: 1. * indicates that the province is in the sample. For 2007 the columns denoted by CHIP and NBS indicate whether the provinces are covered by the CHIP questionnaire and/or by the supplementary data set supplied by the NBS. 2. The geographic regions are (1) large municipalities with provincial status, (2) the coastal regions, (3) the central regions, and (4) the western regions. 3. In the original 1988 CHIP sampling frame, Hebei was classified as part of the central region, but its official NBS classification is coastal. Here we have adopted the NBS classification. 4. Chongqing became a separate province in 1997. It was included in the urban Sichuan sample starting in 1995. For consistency over time, and because Chongqing is less urbanized and does not resemble the other large municipalities, we classify Chongqing in the western region.

The 2002 and 2007 CHIP Surveys

469

through 2007. In each round, sample provinces were selected from each region so as to reflect the economic characteristics of that region. This was done separately for the urban and rural samples, yielding a total of eight strata. The CHIP migrant survey was designed to cover the same four regions as the CHIP urban and rural surveys.2 Including the migrant survey, the 2002 and 2007 CHIP survey data sets comprise twelve strata: rural coastal, rural central, rural western, rural provincial-level municipality; urban coastal, urban central, urban western, urban provincial-level municipality; and migrant coastal, migrant central, migrant western, and migrant provinciallevel municipality. We recommend the use of sample weights based on the population shares of these strata. For the 2002 and 2007 rounds, which are the main focus of this Appendix, weights can be applied for analysis of the CHIP rural sample, urban sample, and migrant subsample (keeping only long-term, stable migrants from the migrant sample so as to avoid double counting – as discussed below), whether they are used separately or in combination. For example, analysis of China’s formal urban population only (i.e., the urban population with a local urban household registration [hukou]) would apply weights from the four urban strata to the CHIP urban survey data. Analysis of China’s total urban population (including rural-urban migrants) would apply weights from the four urban strata to the CHIP urban survey data and weights from the four migrant strata to the CHIP migrant data (long-term, stable migrants only). Analysis of China’s national population would use weights for all twelve strata applied to the respective CHIP rural, urban, and long-term, stable migrant data. Researchers may wish to use weights that reflect not only regional populations, but also provincial populations. The provinces covered in the CHIP surveys have different population sizes, but the CHIP provincial samples are quite similar in size. In principle, whether or not the sample weights should reflect provincial populations depends on the way that the samples are constructed within the regions. If regional samples are selected deliberately to

2

Zhejiang, Fujian, Shandong, Guangdong, and Hainan); central China (Shanxi, Jilin, Heilongjiang, Anhui, Jiangxi, Henan, Hubei, and Hunan); and western China (Inner Mongolia, Guangxi, Chongqing, Sichuan, Guizhou, Yunnan, Tibet, Shaanxi, Gansu, Qinghai, Ningxia, and Xinjiang). The migrant survey for 2002 was carried out in the same twelve provinces as the urban survey, but it covered fewer cities within each province. The migrant survey for 2007 was carried out in nine provinces that were also covered in the 2007 urban survey, but in total the 2007 urban survey covered sixteen provinces.

470

Song Jin, Terry Sicular, and Yue Ximing

ensure that they are representative of the region, then the sample weights need not reflect the provincial population shares. Unfortunately, selection of the CHIP provinces was not done in an entirely transparent manner, and thus, it is unclear whether the sample weights should reflect provincial populations. Here we discuss both approaches and provide two sets of weights. Researchers can decide which approach they prefer.

A. Construction of Weights to Reflect Regional Populations Our sample consists of individuals, each of whom belongs to a stratum. Here “stratum” refers to any of the twelve subgroups discussed earlier, for example, urban-coastal, migrant-central, and so on. The weight wi k for individual i in stratum k is equal to w ki =

Nk , nk

(1)

where N k is the population of stratum k, and nk is the sample size from stratum k. Thus, for example, if the sample from stratum k contains 1 percent of the population of that stratum, then each sample observation represents 100 people, and the weight for each observation is 100. Weighting in this way guarantees that the combination of weighted samples from different strata reflects the combined size of those strata in the national population. For example, the size of the combined weighted samples for all urban strata will equal the size of the national urban population. Similarly, the size of the combined weighted samples for all strata in a region, for example, central China, will equal the size of that region’s population. These weights are a function of the sample and population shares. Let S k = N k /N be the share of stratum k in the national population N, and let sk = nk /n be the share of the sample from stratum k in the overall sample size n. Then the weight wi k for individual i in stratum k can be written as  w ki

=

Nk × n nk × N

 ×

N Sk N = k × . n s n

(2)

In other words, the weight is equal to the stratum’s population share divided by the stratum’s sample share, scaled up by the ratio of the national population to the total sample size.

The 2002 and 2007 CHIP Surveys

471

Equation (2) is appealing intuitively. Note that N/n is the same for all strata. As regression methods and inequality measures are typically scaleinvariant, for most analyses this scaling factor can be dropped and the weights can be calculated simply as the ratio of the population shares to the sample shares. Then equation (2) tells us that the weights depend on whether or not a stratum’s share of the population is bigger or smaller than its share of the sample. So, for example, if the share of rural-central China in China’s national population exceeds its share in the CHIP sample, then observations from rural-central China would receive a weight greater than one.

B. Construction of Weights to Reflect Regional and Provincial Populations Whether weights should also reflect provincial populations depends on how the sample provinces are selected. If the sample provinces and provincial samples within each region are deliberately selected so that their pooled samples are representative of the region, then weights need not reflect the provincial populations. Such would be the case, for example, if J sample provinces were selected out of the M provinces in stratum k, and these provinces were chosen because their combined populations are representative of the stratum. In this case, drawing a random sample of nk individuals from the pooled populations of the sample provinces would be identical to drawing nk individuals from the entire stratum. The probability of an individual being chosen would be pi k = nk /N k . The same result would apply if random samples were drawn separately for each province, with the sample size for each province nj k being proportional to its population size Nj k . Then the probability of an individual being chosen would be pi k = (nj k /Nj k ) × (Nj k /Nk ) = nk /Nk . In either case, the sample weights would be identical to those given in Equation (1). Therefore, the weights would only need to reflect the regional populations, not the provincial populations. Suppose instead that the provinces are chosen to be jointly representative of the region, but the size of each provincial sample is not proportional to its population. This is possibly the case for the CHIP samples. Then the weights should reflect that the probability of being selected differs among provinces. Let Nj k be the population and nj k be the sample size of province j in stratum k. Then the probability of an individual being drawn within a

472

Song Jin, Terry Sicular, and Yue Ximing

province is pi j,k = nj k /Nj k . The weight vi j,k for an individual i located in province j of stratum k can then be written as     Njk Nk j ,k ×  k (3) vi = nkj j Nj One can see that the second term is equal to the population of stratum k divided by the sum of the populations of sample provinces in stratum k. As is the case for the stratum weights wi k shown in Equation (1), the sum of the combined weighted samples for multiple substrata will equal the combined population of those substrata. For example, the size of the combined weighted provincial rural samples will equal the national rural population. One can restate Equation (3) in terms of population and sample shares as follows:       S jk Sk N j ,k × . (4) × vi = k k s n sj In Equation (4), Sj k is province j’s population divided by the sum of the populations of sample provinces in stratum k, that is, it is province j’s share of the population of all the sample provinces in the stratum. Similarly, sj k is province j’s sample share in the total sample of stratum k. The second two terms in Equation (4) are identical to Equation (2).

III. Population Shares: From the 2000 Census and the 2005 Mini Census Calculation of weights as outlined earlier requires information about the populations N k or Nj k of the different location strata. For 2002 we obtain this information from the Chinese 2000 census, and for 2007 we obtain it from the 2005 mini census, which is a 1 percent sample of the national population. Note that the mini census was not constructed entirely according to population shares across provinces. The NBS provides weights that can be used to adjust the mini-census data so that they reflect more accurately the provincial populations.3 We use these weights when we calculate population shares from the 2005 mini census. 3

The weight variable’s name is power_2. The data contain a value of this variable for each individual, taking 590 different values ranging from .082149 to 2.454594.

The 2002 and 2007 CHIP Surveys

473

The census and mini census counted individuals at a specific point in time (for the census, at midnight, October 31, 2000; for the mini census, the night of October 31, 2005). For each individual, the census and mini census contain a location flag as well as other information, such as gender, age, relationship to the household of residence, type of hukou, length of time away from the location of the hukou, and so forth. We do not have access to the full data sets for the 2000 census and the 2005 mini census; however, the NBS has provided us with randomly selected subsamples. For the 2000 census we have a 0.095 percent sample, and for the 2005 mini census we have a 20 percent subsample. The NBS selected these subsamples using systematic interval sampling, so they should be representative of the full census and the full mini census. We checked the composition of our subsamples of the census and mini census against the aggregated data from the full census and the full mini census published by the NBS. The subsamples’ population shares among provinces, by gender, and by city/town/village are similar to those for the full census and the full mini census. For calculation of weights, we make use of each individual’s location (city, town, or village) flag. The location flags in the 2000 census followed certain criteria that were designed to ensure that the census counted stable residents and that people who had moved were not double counted. An individual was flagged in his or her location at the time of the census if (a) he or she was living in and had a hukou in that location (including members of households in the location who were not present at the time of the census but had been away for less than six months) or (b) he or she had a hukou elsewhere but was living in that location at the time of the census and had been living there for more than six months.4 The 2005 mini census used a different approach. All individuals were flagged in their location at the time of the mini census. In addition, individuals who were members of households in a location and had a hukou in that location but were away (waichu renkou) at the time of the mini census

4

People present in a location at the time of the census were also flagged in that location if (a) they had lived in the location for less than six months but had left the place of their hukou for more than six months, or (b) they had no hukou but they were living or used to live in that location (e.g., newborns, or people studying abroad temporarily). The details of how the locations were flagged in the 2000 census can be found at http://www.stats.gov .cn/tjsj/ndsj/renkoupucha/2000pucha/html/append5.htm. Accessed October 11, 2010.

474

Song Jin, Terry Sicular, and Yue Ximing

were flagged as residents of that location. This approach might lead to some double counting of individuals who were away from their households at the time of the mini census.5

IV. Classification of Location In order to construct weights, we need to classify individuals according to their location of residence into the different strata. This classification must be done consistently for all data sets used to construct weights, that is, for the 2000 census, the 2005 mini census, and the 2002 and 2007 CHIP urban, rural, and migrant samples. As each location is either urban (including cities and towns) or rural (villages), the consistent classification of individuals by location ensures consistent classification of individuals as urban or rural. The classification is applied to all individuals, including migrants. Migrants who, according to the classification criteria, are classified as residents of a city or town will be counted as urban; those classified as residents of a village will be counted as rural. The criteria we adopt for classification of location are those used by the NBS in its annual rural and urban household surveys. The CHIP rural and urban household survey samples are subsets of the NBS rural and urban household surveys, therefore, using the same criteria is practical. The NBS criteria consider not just the location and length of residence, but also the strength of economic ties between the individuals and the households. The NBS (and CHIP) urban and rural survey samples consist of households and their members. An individual is counted as a resident in a location if he or she is a member of a household in that location and if he or she is usually living in the household or has lived in that household for six months or more during the survey year. An individual who is not usually living in the household or who is away from the household for more than six months is counted as a resident if most of his or her income is returned to the household, or if he or she maintains a close economic relationship with the household. Individuals who do not satisfy these criteria are not counted as residents of the location.

A. Reclassifications of the Census and Mini-Census Samples The criteria used to flag location in the 2000 census and 2005 mini census are different from those used in the NBS household surveys, so we must 5

The details of how the locations were flagged in the 2005 mini census can be found in 2005 nian quanguo 1% renkou chouyang diaocha ziliao (Tabulation of the 2005 National Sample Survey of 1 Percent of the Population) (Beijing: Zhongguo tongji chubanshe, 2008), 833.

The 2002 and 2007 CHIP Surveys

475

reclassify the individuals in the census and mini census before constructing the population shares and sample weights. The most important difference is in the treatment of individuals who are away from their households for more than six months but maintain an economic relationship with the household. The census and mini census count these individuals in their place of residence; we must reclassify them in the location of their households of origin. The census and mini census do not contain information about the strength of an individual’s economic relationship with his or her household of origin, but they contain information about marital status and about whether the individuals are living with their spouses. We use this information as a proxy for the strength of their relationship with the household of origin. If an individual with a nonlocal hukou is married and not living with his or her spouse, we consider that person as having a significant economic relationship with his or her household in the location of the hukou. We consider such individuals to be unstable migrants. If an individual with a nonlocal hukou is not married (single, divorced, or widowed) or is married and is living together with his or her spouse, then we consider that person as not having a strong economic relationship with his or her household in the location of the hukou. We consider such individuals to be stable migrants. We must also carry out some additional reclassifications of individuals in the 2005 mini census because the approach used to flag location in the 2005 mini census is different from that used in the 2000 census. For consistency with the census and the NBS household surveys, we reclassify individuals who have lived in the location at the time of the mini census for less than six months in the place of their hukou.6 In order to carry out these location reclassifications, we examine all individuals in our subsamples of the census and mini census. We accept the flagged location and do not reclassify individuals who satisfy either of the following two conditions: 6

The mini census contains information about place of hukou (province), type of hukou (agricultural or nonagricultural), and length of time away from the place of the hukou. We use this information to carry out this reclassification. Information about the length of time is given by the answer to the question (R8), “How much time since he/she left the place of his/her hukou registration” (likai hukou dengji di shijian). This is slightly different from asking how long the individual has lived in the current location. For example, it is possible that a migrant may have left the original place of the hukou a long time ago, first going elsewhere and only recently moving to the current place of residence. We have no information about where the individuals have resided since leaving the place of their hukou registration. We assume that individuals who have been away from the place of their hukou for six months or more have been living for the last six months in their current place of residence.

476

Song Jin, Terry Sicular, and Yue Ximing

1. They hold a local hukou (regardless of whether the local hukou is agricultural or nonagricultural), and (a) they are currently living in the location, or (b) they are absent but they are members of local households and have been away for less than six months. 2. They do not hold a local hukou but have been living in the flagged location for more than six months and are either (a) single, divorced, or widowed, or (b) married and living with spouse. All other individuals are reclassified as a resident in the province of their hukou. In other words, all individuals who do not hold a local hukou and have been living in the flagged location for less than six months are reclassified, as are all individuals who do not have a local hukou and have been living in the flagged location for more than six months, are married, and are not living with their spouses. Individuals who are reclassified back to the province of their hukou will be designated as rural or urban, based on whether they have an agricultural or nonagricultural hukou. If they have an agricultural hukou, they are reclassified as a rural resident of the province of their hukou; if they have a nonagricultural hukou, they are reclassified as an urban resident of that province. This reclassification scheme effectively treats temporary migrants and long-term, unstable migrants as residents of the place of their hukou. Migrants who are long term and stable are not reclassified. Note that reclassification can occur for any type of migrant, including urban-urban, ruralrural, urban-rural, or rural-urban. Rural-urban migrants, however, are of particular interest and are the most numerous. Table AII.2 gives a summary of the 2000 census and 2005 mini-census samples before and after reclassification. For the 2000 census, reclassifications were mainly confined to individuals who lived in the location for more than six months and were married but not living with a spouse. There were more reclassifications for the 2005 mini census because in the mini census individuals who have lived in the location for less than six months were also reclassified.

B. Reclassifications of the CHIP Survey Samples Because we have adopted the location criteria used in the NBS urban and rural household surveys, and because the CHIP urban and rural samples are drawn from the NBS household surveys, we do not need to reclassify individuals in the CHIP urban and rural survey samples. We treat all

Table AII.2. Summary of the 2000 census and 2005 mini-census samples before and after reclassification

Original Number 477

2000 Census urban rural 2005 Mini Census urban rural

% of population

432,315 747,795

36.6% 63.4%

1,147,410 1,417,005

43.7% 55.3%

Reclassified out To rural

Reclassified in

After reclassification

To urban

From rural

From urban

Missing data (dropped)

Number

% local

% migrant

% of population

0 191

191 0

0 8,293

2,380 223

421,833 755,674

93.1% 100%

6.9% 0%

35.8% 64.2%

25,598 5,914 7,769 926

926 7,769

5,914 25,598

4,642 1,137

1,118,167 1,440,825

92.9% 56.3%

7.1% 100%

43.7% 56.3%

8,293 0

Notes: The numbers for the 2005 mini census are weighted by power_2. In principle, the original number plus the numbers “reclassified in” minus the numbers “reclassified out” and missing should equal the post reclassification number. This is true for the 2000 census numbers, but small discrepancies exist for the 2005 mini-census numbers due to weighting. Without weighting, the equality holds.

478

Song Jin, Terry Sicular, and Yue Ximing

individuals in the CHIP rural sample as residents in their given rural locations, and all individuals in the CHIP urban sample as residents in their given urban locations.7 The NBS rural surveys treat individuals who live in their rural households of origin most of the time, or who live away from the household for more than six months but maintain a close economic relationship with the household as members of the rural households. Short-term, unstable, ruralurban migrants are therefore counted in the rural survey. The problem of underrepresentation of migrants, then, occurs mainly for longer-term ruralurban migrants who do not maintain a close economic relationship with their rural households. This group of migrants is included in the CHIP migrant surveys. The CHIP migrant surveys also include other types of individuals. These surveys are samples of individuals with agricultural hukou who live in urban areas, including not only long-term, stable rural-urban migrants, but also individuals with local agricultural hukou, short-term rural-urban migrants, and long-term rural-urban migrants who maintain a close economic relationship with their rural households of origin. For the purpose of calculating weights, we need to drop these latter types of individuals, as they are already included in the NBS and CHIP rural surveys. On this basis, we only keep individuals in the CHIP migrant surveys who have nonlocal hukou and satisfy one of the following criteria:8 1. They have been living in the urban location for more than six months and they are single, divorced, or widowed. 2. They have been living in the urban location for more than six months and they are married and living with their spouses. We call these individuals long-term stable migrants. Individuals who have been living in the urban location for less than six months, or for more than six months but are married and not living with their spouses, are dropped. We call these individuals short-term or long-term unstable migrants. Individuals who have local agricultural hukou are also dropped. 7

8

We checked the CHIP urban surveys and in fact found some individuals who have nonlocal rural hukou, but these proportions are very small – less than 1 percent of the total observations. The 2007 CHIP migrant survey contains the question, “How many months have you stayed outside your hometown for work or business?” (Zuijin 12ge yue nei, zai waichu wugong jingshang di yigong shenghuole jige yue?). The 2002 CHIP migrant survey contains a similar question, “How many months did you stay in an urban area in 2002” (Zai 2002 nian nin zonggong zai chengzhen juzhu shijian duoshao [yue]?). We use the answers to these two questions to determine the individual’s migration duration.

The 2002 and 2007 CHIP Surveys

479

Table AII.3. Composition of the CHIP migrant samples, 2002 and 2007 Category (1) Local agricultural hukou (2) Long-term stable (3) Short-term and long-term unstable (4) Missing Total

2002

2007

1,938 (36.4%) 2,976 (55.9%) 278 (5.2%) 135 (2.5%) 5,327 (100%)

1,806 (21.4%) 5,303 (62.8%) 1,289 (15.3%) 98 (1.2%) 8,446 (100%)

Note: This table gives the number of individuals; percentages of the migrant sample for that year are shown in parentheses.

Table AII.3 shows the number (and percentage) of individuals in the 2002 and 2007 CHIP migrant surveys who satisfy the preceding criteria for long-term, stable migrants. It also shows the number of individuals in the migrant surveys who belong to other categories. Note that the different compositions of the 2002 and 2007 migrant samples reflect in part the differences in the sampling methods used to construct the migrant samples in the two years. The use of residential neighborhood committees as the sampling frame in 2002 led to a higher proportion of long-term stable migrants and individuals with local agricultural hukou. We have created a variable catg for the 2002 and 2007 migrant data sets that identifies individuals as long-term, stable migrants according to these criteria. The Stata data files mcatg02.dta and mcatg07.dta contain this variable and ID variables to facilitate merging with the CHIP migrant survey data set (available on request from the authors). The variable catg can be used to keep or drop observations. When calculating weights and using the migrant data in combination with the CHIP urban and rural data sets, observations with catg = 2 satisfy the criteria for long-term, stable migrants and should be kept; all other observations should be dropped.9 9

The variable catg takes a value of “1” if the individual has a local agricultural hukou, “2” if he or she is a long-term stable migrant, and “3” if he or she is a short-term or unstable long-term migrant. A missing value indicates that the individual cannot be identified as a member of any of these three groups. We drop individuals if they have a missing value. Researchers can follow our approach and drop individuals with missing values of catg, or they can use other information in the data sets to classify them and include them in their calculations.

480

Song Jin, Terry Sicular, and Yue Ximing

V. Implementation of Weights Tables AII.4 and AII.5 contain the numbers of individuals in each of the twelve strata and their component provinces in our subsamples of the 2000 census and the 2005 mini census, after reclassification. Researchers can use these numbers as values for Nk or Nj k in the calculation of weights. Sample sizes for each of the strata nk and its component provinces nj k will vary depending on the sample used in the analysis, so researchers will calculate these based on the set of observations used in their analyses. The numbers in Tables AII.4 and AII.5 are appropriate for calculation of weights in analyses at the individual or per capita level. Analyses at the household level should use weights calculated using counts of households because the number of individuals per household differs among the strata. Tables AII.6 and AII.7 give the counts of households in our subsamples of the 2000 census and 2005 mini census for each stratum and its component provinces. Researchers can use these numbers as the population frequencies Nk or Nj k for calculation of the household-level weights. Sample counts of households will depend on the observations actually covered in the analysis and thus should be calculated by the researcher accordingly. This Appendix discusses the calculation of weights based on the geographic distribution of the population among regions and provinces, as well as among urban, rural, and migrant groups. Some researchers may be interested in different subdivisions of the population, for example, between Han and minority groups, or among education groups or age cohorts. Researchers who are analyzing such subgroups will wish to construct weights to ensure that the results are representative of those subgroups. In these cases, one can combine weights based on the regional strata discussed here with weights based on the populations and sample sizes of the subgroups of interest. For example, Chapter 5 in this volume by Knight, Sicular, and Yue about intergenerational educational mobility uses weights based on age cohorts. Similarly, an analysis of the differences between Han and minority groups might use weights that reflect the sizes of the Han and minority populations in each stratum. To avoid double counting, in our classification of individuals (and households) by location, we have chosen to drop individuals in the CHIP migrant survey who have local agricultural hukou because such individuals are also included in the CHIP urban sample. Some researchers, however, may wish to add these urban residents to the CHIP urban sample rather than to drop them. If so, the weights will need to be adjusted accordingly. Similarly, we have dropped short-term and unstable migrants from the CHIP migrant

The 2002 and 2007 CHIP Surveys

481

Table AII.4. Population frequency by stratum, 2000 (individuals in the 0.095 percent subsample of the 2000 census)

Province code 11 12 31 13 21 32 33 35 37 44 46 14 22 23 34 36 41 42 43 15 45 50 51 52 53 54 61 62 63 64 65 Total

Province name

Region code

Urban locals

Stable long-term migrants

Rural locals

Beijing Tianjin Shanghai Subtotal Hebei Liaoning Jiangsu Zhejiang Fujian Shandong Guangdong Hainan Subtotal Shanxi Jilin Heilongjiang Anhui Jiangxi Henan Hubei Hunan Subtotal Inner Mongolia Guangxi Chongqing Sichuan Guizhou Yunnan Tibet Shaanxi Gansu Qinghai Ningxia Xinjiang Subtotal

1 1 1 1 2 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 3 3 4 4 4 4 4 4 4 4 4 4 4 4 4

8,597 6,369 11,796 26,762 16,202 21,479 26,866 18,647 11,311 31,845 32,291 2,617 161,258 10,382 11,935 16,473 14,395 9,461 19,552 20,500 14,974 117,672 8,769 10,642 8,801 19,112 7,316 7,953 325 10,022 5,425 1337 1,606 5,757 87,065 392,757

984 289 1,404 2,677 670 824 1,659 1,966 1,230 1,128 8,171 196 15,844 451 359 760 514 310 785 951 650 4,780 765 695 313 841 597 981 77 371 286 120 122 607 5,775 29,076

3,232 2,800 1,871 7,903 48,964 18,727 41,626 23,717 18,971 56,274 36,769 4,287 249,335 20,946 12,530 16,079 41,806 25,650 69,598 30,636 42,325 259,570 12,903 30,713 17,844 55,774 26,328 31,672 1,967 23,532 18,585 3,188 3,718 12,642 238,866 755,674

482

Song Jin, Terry Sicular, and Yue Ximing

Table AII.5. Population frequency by stratum, 2005 (individuals in the 20 percent subsample of the 2005 mini census)

Province code 11 12 31 13 21 32 33 35 37 44 46 14 22 23 34 36 41 42 43 15 45 51 50 52 53 54 61 62 63 64 65 Total

Province name

Region code

Beijing Tianjin Shanghai Subtotal Hebei Liaoning Jiangsu Zhejiang Fujian Shandong Guangdong Hainan Subtotal Shanxi Jilin Heilongjiang Anhui Jiangxi Henan Hubei Hunan Subtotal Inner Mongolia Guangxi Sichuan Chongqing Guizhou Yunnan Tibet Shaanxi Gansu Qinghai Ningxia Xinjiang Subtotal

1 1 1 1 2 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 3 3 4 4 4 4 4 4 4 4 4 4 4 4 4

Urban locals 20,476 13,408 24,010 57,894 46,357 46,773 72,030 43,135 28,168 80,865 86,997 8,211 412,535 27,336 26,608 40,744 51,554 31,078 59,245 47,604 43,862 328,032 24,114 28,168 24,945 49,475 18,709 25,777 1,559 30,038 15,383 3,742 4,573 13,512 239,996 1,038,458

Stable long-term migrants

Rural locals

3,085 1,355 4,687 9,127 1,347 2,112 6,767 7,548 4,982 3,029 21,147 598 47,530 1,020 845 1,674 1,379 590 1,180 2,080 1,754 10,522 2,312 1,259 662 1,857 1,100 2,136 88 1,092 446 215 255 1,108 12,530 29,076

4,754 5,186 3,602 13,542 90,331 32,848 65,983 42,212 35,009 95949 72,754 8,068 443,153 39,042 25,644 32,127 85,144 52,585 129,229 64,398 81,831 510,001 21,754 58,636 31,407 116,263 55,322 62,381 4,754 45,526 38,466 6,483 7,203 25,932 474,128 1,440,825

The 2002 and 2007 CHIP Surveys

483

Table AII.6. Population frequency by stratum, 2000 (households in the 0.095 percent subsample of the 2000 census)

Province code 11 12 31 13 21 32 33 35 37 44 46 14 22 23 34 36 41 42 43 15 45 50 51 52 53 54 61 62 63 64 65 Total

Province name

Region code

Urban locals

Stable long-term migrants

Rural locals

Beijing Tianjin Shanghai Subtotal Hebei Liaoning Jiangsu Zhejiang Fujian Shandong Guangdong Hainan Subtotal Shanxi Jilin Heilongjiang Anhui Jiangxi Henan Hubei Hunan Subtotal Inner Mongolia Guangxi Chongqing Sichuan Guizhou Yunnan Tibet Shaanxi Gansu Qinghai Ningxia Xinjiang Subtotal

1 1 1 1 2 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 3 3 4 4 4 4 4 4 4 4 4 4 4 4 4

2,914 2,131 4,103 9,148 4,723 7,041 8,443 6,104 3,274 9,712 8,758 699 48,754 3,095 3,789 5,389 4,399 2,776 5,683 6,150 4,789 36,070 2,877 3,136 2,944 6,278 2,126 2,533 115 3,042 1,675 432 505 1,773 27,436 121,408

330 82 514 926 206 246 538 679 381 287 2,160 60 4,557 116 102 228 128 69 176 273 189 1,281 206 196 89 228 159 313 29 116 87 31 34 191 1,679 8,443

880 773 614 2,267 13,165 5,499 12,111 7,495 4,931 16,944 8,460 982 69,587 5,497 3,540 4,585 11,697 6,931 18,267 8,632 12,264 71,413 3,723 7,887 5,627 16,504 6,695 7,913 404 6,203 4,321 706 855 3,073 63,911 207,178

Note: Includes collective households for migrants, but not for urban and rural locals.

484

Song Jin, Terry Sicular, and Yue Ximing

Table AII.7. Population frequency by stratum, 2005 (households in the 20 percent subsample of the 2005 mini census)

Province code 11 12 31 13 21 32 33 35 37 44 46 14 22 23 34 36 41 42 43 15 45 51 50 52 53 54 61 62 63 64 65 Total

Province name

Region code

Urban locals

Stable long-term migrants

Rural locals

Beijing Tianjin Shanghai Subtotal Hebei Liaoning Jiangsu Zhejiang Fujian Shandong Guangdong Hainan Subtotal Shanxi Jilin Heilongjiang Anhui Jiangxi Henan Hubei Hunan Subtotal Inner Mongolia Guangxi Sichuan Chongqing Guizhou Yunnan Tibet Shaanxi Gansu Qinghai Ningxia Xinjiang Subtotal

1 1 1 1 2 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 3 3 4 4 4 4 4 4 4 4 4 4 4 4 4

13,413 24,032 24,199 61,644 11,190 16,966 17,522 12,994 8,531 23,875 61,217 5,830 158,125 18,449 15,042 16,014 12,116 9,288 11,142 15,712 13,159 110,922 10,833 8,585 11,467 12,524 6,411 14,221 1,318 17,008 10,714 4,658 3,956 6,132 107,827 438,518

2,025 1,798 5,287 9,110 335 741 1,853 2,606 1,851 833 13,158 420 21,797 677 423 550 308 145 195 577 512 3,387 853 354 253 412 334 1,218 103 589 293 280 181 503 5,373 3,9667

2,939 5,941 3,864 12,744 19,923 10,623 15,906 12,923 10,228 27,360 44,665 4,892 146,520 23,593 12,505 10,498 18,841 15,114 22,658 19,977 23,123 146,309 8,947 16,378 13,815 27,718 1,050 36,763 2,915 25,270 20,541 6,044 4,491 8,963 188,895 494,468

Note: Includes collective households for migrants, but not for urban and rural locals.

The 2002 and 2007 CHIP Surveys

485

survey because they are also included in the CHIP rural sample. Some researchers may wish instead to add this group to the CHIP rural sample, in which case once again the weights will need to be adjusted accordingly. Finally, as discussed in Appendix I, we note that for 2007 not all variables are available for the full rural and urban CHIP samples. The sample size will therefore depend on which variables are being used and the number of individuals or households for which they are available. Researchers will therefore need to pay close attention to their sample sizes and recalculate the weights accordingly.

Index

age. See also children; elderly; middle-aged and older workers; students; young adults age-wage profile for female workers, 396 homemakers and, 313 housing wealth and, 127 impact of on owning housing-reform housing, 115 in CHIP 2007 rural survey, 457 in CHIP 2007 rural-urban survey, 462 in CHIP 2007 urban survey, 453 income growth curves by, 275 migration and, 249 nonwork and nonworkers in urban China and, 304–316 of retirement by year, 302 of young men and young women beginning work, 292 poverty and, 276 unemployment rates by, 307 urban income and, 279–281 workforce average, 339 agricultural taxes and fees elimination of, 14, 44, 197, 218–221 agriculture, 150 agricultural income, 59, 77 agricultural subsidies for rural households, 15, 45 as major source of rural income, 198, 201, 202, 226 farmland preservation, 95 industrialization of agriculture policy, 17, 146

slowdown in agricultural production, 198 Ant Tribe (Lian Si), 295 asset income. See also property income housing wealth and, 121, 124, 126, 127 income inequality and, 34, 37, 77 migrants and, 61, 63 poverty and, 31 rural household income and, 202, 206, 207, 225 Basic Medical Insurance for Urban Residents Program (chengzhen jumin jiben yiliao baoxian zhidu), 7 Basic Pension Insurance Program for Enterprise Employees (qiye zhigong jiben yanglao baoxian zhidu), 6 benefits for laid-off workers (xiagang butie), 297 business income, 256, 261 defined, 259 income inequality and, 271 business tax, 364 central region, rise of (zhongbu jueqi) scheme, 46 children. See also educational inequality income growth curves and, 275 migration and, 246 poverty and, 276 urban income inequality and, 261, 273, 278–281 China Household Income Project (CHIP). See CHIP surveys (general); CHIP urban surveys; CHIP rural surveys; CHIP rural-urban migrant surveys

487

488

Index

CHIP income, 28, 200 defined, 53 measurement and definition of, 19–21 CHIP rural survey (2002), 23, 50, 199 parental education information in, 144–145 provinces covered in, 199, 234 CHIP rural survey (2007), 23, 51, 199 distribution of households in, by province, 455 distribution of households in, by size and province, 456 distribution of individuals in, by age and province, 457 education level of parents in, 142 educational attainment of individuals over the age of 15 in, by province, 459 gender composition of individuals in, by province, 456 intergenerational transmission of education and, 157–159 provinces covered in, 199, 235 sample size for, 447 CHIP rural-urban migrant survey (2002), 23–26, 50, 51, 199, 230, 234 bias in, 235 composition of, 479 hukou registration and, 446 migrant wages and, 236–244 CHIP rural-urban migrant survey (2007), 23–26, 51, 199 bias in, 235 composition of, 479 distribution of households and individuals in, by city, 461 distribution of individuals in, by age and city, 462 educational attainment of individuals over the age of 15 in, by city, 462 gender composition of individuals in, by city, 461 hukou registration and, 446 migrant wages and, 236–244 provinces covered in, 235 RUMiCI (Rural-Urban Migrants in China and Indonesia), 22 sample size for, 447 CHIP surveys (general), 4, 51 alternative responses to employment question on, 305

as basis for the analyses of income inequality, 37 census and mini-census population information and, 472–474 census and mini-census sample reclassification, 474–476 classification of location and, 474 comparison of CHIP and NBS per capita household incomes, 28 construction of weights to reflect provincial populations, 471–472 construction of weights to reflect regional populations, 470–471 data sources of, 26 design and survey question changes, 27 exclusion of self-built/inherited older housing in urban, 112 four regional groups covered in, 33 four waves of, in 1988, 1995, 2002, and 2007, 445 geographic regions used to construct sample frame of, 52 implementation of weights in, 480–485 lack of information on intra-household allocation in, 256 number of households covered in, 449 provinces covered in, 22, 467 questions on early retirement in, 304 reclassification in, 476–479 samples from earlier years, 22 sampling and sample size of, 445–448 summary of key indicators of inequality and poverty in, 30 three subsamples in, 22 vs. NBS surveys, 28, 448–450 weight calculations in, 27, 466–472 CHIP urban survey (1988), 261 nonworkers in, 300 CHIP urban survey (1995), 261 gender-wage gap in, 389 nonworkers in, 300 CHIP urban survey (2002), 23, 50 enterprise ownership in, 337–341 gender-wage gap in, 389 income inequality in, 261 inconsistent data on urban rental values in, 133–136 nonworkers in, 300 provinces covered in, 261

Index CHIP urban survey (2007), 23, 51 distribution of households in, by age group and province, 453 distribution of households in, by province, 451 distribution of households in, by size and province, 452 education level of parents in, 142 educational attainment of individuals over age 15 in, by province, 454 enterprise ownership in, 337–341 gender composition of individuals in, by province, 451 gender-wage gap in, 389 income inequality in, 261, 369 intergenerational transmission of education in, 157–159 nonworkers in, 300 personal income tax in, 363, 369 provinces covered in, 235, 261 sample size for, 447 Communist Party, 96, 119, 121, 443 private enterprise ownership and, 258, 332 Company Law, 336 Compulsory Education Law, 151 consumer prices, 53 consumption tax, 364 corporate income tax, 365 cost of living. See also purchasing power parity (PPP) differences among regions, 53, 59, 65 inequality and, 203 Cultural Revolution educational trends during, 150, 187, 295 cumulative density functions, 255 danwei. See work unit dibao. See minimum living standard guarantee (dibao) program discrimination. See also gender-wage gap against female workers, 38, 386, 402 law and policy regarding ethnic, 420–422 disposable income, 19, 20, 52, 65, 114, 270, 280, 321, 326 measurement of, 234 early retired and retirement, 11, 292, 293, 321 as nonworkers, 290, 291 defined, 303

489

phrasing in CHIP surveys in 2007 vs. previous surveys, 304 rate of, by age, 302 underrepresentation of, in lowest decile, 327 earnings. See also earnings gap across ownership sectors; gender-wage gap; Han-minority earnings gap; income; urban-rural income gap; wages hourly, 341–345 earnings gap across ownership sectors, 332–335 descriptive statistics on, 342 hourly earning determinants, 345–348 hourly earnings by ownership category and year, 343 Juhn-Murphy-Pierce decomposition of earnings differentials, 353–358 Oaxaca-Blinder decomposition of earnings differentials, 348–353 summary statistics on earnings by ownership, 341 working time and, 343 economic growth, 44, 414 differential among regions, 46 income inequality and, 1 migration and, 230 proportion of children and elderly and, 257 rise in percentage of nonworkers and, 289 surplus labor and, 231 urbanization and, 46 education. See also educational inequality attainment among minorities, 416, 417, 418 class background (chengfen) and, 144, 146 credit constraints and, 188 discrimination against female workers and, 384 earnings by employment ownership sector and, 346 economic growth and, 2 gender and, 296 gender-wage gap and, 391 NBS estimates of, 385–386 Han-minority earnings gap and, 432–435 income growth curves and, 275 inequality in, 154, 178 middle school progression rates, 144, 146 nine-year free, 45, 188, 197

490

Index

education (cont.) nonwork and nonworkers and, 289, 292, 293–296, 309 of individuals over the age of 15 in CHIP 2007 rural survey, 459 of individuals over the age of 15 in CHIP 2007 rural-urban migrant survey, 462 of individuals over the age of 15 in CHIP 2007 urban survey, 454 parental, 144–145, 165–167 population shares, mean income, and income inequality by level of, 274, 277 preferential policies for minorities, 425 primary net enrollment rates, 144, 146 survey of policies and goals, 145–154 urban-rural income gap and, 33 work force and, 341 educational inequality, intergenerational transmission of education and, 142, 177–181, 185–188 aggregate educational mobility and, 159, 185 aggregate educational mobility defined, 142 average years of the sons‘/daughters’ education by the educational levels of fathers/mothers, 165–167 by birth cohort, 171–174 causal relationships as factor in policy, 187 changes across policy periods, 186 data issues, 189–190 differences between rural and urban areas, 186 gender-specific transmission, 187 general findings, 182–185 in 2007 CHIP survey, 142, 157–159 international literature on educational level of parents, 143–144 levels used in analysis of, 191 low and high educational level household comparison, 174–177 men and women, 168–171 microeconomic level of, 142 overall, and by urban-rural and birth year, 160–165 policies and goals affecting, 145, 153–154 rural areas and, 144–145 spatial considerations and, 144 theory and methodologies, 154–156, 181–182

urban and rural cohorts, 168–171 urban areas and, 146, 154 elderly, 261, 273, 276, 278, 279–281 imputed rents from owner-occupied housing and, 260 income growth curves and, 275 migration and, 246 pension income of as factor in income inequality, 260 pensions of in rural areas, 6 employment. See also self-employment; unemployment labor force and, in China (1995–2007), 233 labor-market policies and minimum wage regulations, 12 trends in, 390 ethnic minorities. See also Han-minority wage gap gender-wage gap and, 392 law and policy regarding, 420–422 one-child policy and, 421 rules on classifying, 420 farm income. See agriculture farmland, 34, 71 farmland preservation, 95 rural land for housing use (zhaijidi) and, 95 female workers. See also gender; gender-wage gap; nonwork and nonworkers in urban China discrimination against, 38, 290, 299, 384, 386, 389, 395, 398, 401 unemployment and, 391 unskilled labor and, 384 fiscal decentralization, 151 fiscal recentralization, 152 five-guarantee program (wubao), 6, 11 Five-Year Plan, First (1953–57), 145, 187 Foster, Greer and Thorbecke (FGT) indices of poverty, 264 GDP annual growth in, 44, 414 gender, 298–299, 326, 327 composition of individuals in CHIP 2007 rural survey, 456 composition of individuals in CHIP 2007 rural-urban migrant survey, 461

Index composition of individuals in CHIP 2007 urban survey, 451 earnings and, 317 education and, 296 increase in age of beginning work and, 292 middle-age workers and, 309 unemployment rates and, 307, 391 urban employment in SOEs and, 338 gender-wage gap, 384–386 age-wage profile for female workers, 396 decomposition of changes in, 399–400, 402 educational attainment and, 391 international context, 387 low-wage groups and, 402 methodology of decomposition of, 394–395 NBS estimates for, 29 ownership sectors and, 398 postreform widening of, 384, 402–404 previous findings on, 387–389 privatization of state-owned enterprises and, 385 quantile regression results, 400 regression analysis of, 397 wage structure by age, minority group, and marital status, 392 women’s economic dependency by year, 318 Gini coefficient, 29, 53 calculation of, 55 cost of living (PPP) and, 59 decomposition of by income source (2002 and 2007), 58, 63, 206, 268 educational inequality and, 178, 182 for city employment rates, 300 for hourly earnings, 343 for household per capita incomes, 30, 138 for housing wealth, 107 for income sources, 270 for personal income among people of work-active ages, 323 imputed rents from owner-occupied housing and, 202 income inequality and, 267–271 international comparisons of, 55, 76 NBS estimates for, 29 pre- and post-tax income, 378 regional, 71 rise of 2002 to 2007, 55

491

rural, 29, 203 urban, 29, 415 Government Employee Health Insurance Program (gongfei yiliao), 7 government subsidies for agricultural production, 197 Great Leap Forward (1958), education during, 148 grey income, 21 Han-minority earnings gap, 392, 416, 418, 426–427, 430–432 descriptive statistics, 427–430, 434 determinants of, 436–438 determinants of change in, 424–425 females and, 440–441 human capital and occupational model, 423–424 in 1995, 2002, and 2007, 428 income ratio of, 429 intertemporal decompositions of, 426, 439–440 intratemporal decompositions of, 425–426, 438–439 males and, 441–443 ordinary least squares estimates of the effects of minority status on ln-earnings, 431 ratio of mean and median family household total incomes in urban China (1995, 2002, and 2007), 418 ratio of wage and salary incomes (1995, 2002, and 2007), 419 residual difference analysis, 435–436 return to education and SOE employment and, 425, 432–435 harmonious society (hexie shehui), 3, 414 health, 2, 303, 411 migration and, 246 high-income groups, underrepresentation of in household surveys, 21, 77 homemakers, 292, 309, 320, 321, 387 age as factor in, 313 as nonworkers, 290, 291 reappearance of traditional, 327 homeownership, 4, 86, 114, 129 international rates of, 129 mean income growth and, 55 migrant households and, 105 mortgage debt among, 101 NBS surveys yield rates of, 23, 26, 91, 99, 132–133, 448

492

Index

homeownership (cont.) rural households and, 104 urban households and, 104 household income. See also rural household income; urban household income by employment sector, 273 components and growth of per capita (2002 and 2007), 268 Gini coefficients of income sources (2002 and 2007), 270 increase in, between 2002 and 2007, 55 per capita, 52, 269 personal income and, 321–323 regional differences and, 67–72 household registration. See hukou housing. See also homeownership; imputed rental income from owner-occupied housing comparisons of market value and equity value per capita (2002), 101 floor area of urban (1990–2007), 91 housing provident fund (zhufang gongjijin), 90 housing-related questions in 2002 CHIP survey, 50 income inequality and, 108–110 inequality in access to, 94 marriage and, 124 prices in urban China, 93, 260 rent-price ratios and, 135 self-built/inherited older housing exclusion in CHIP urban surveys, 112 housing reform, 85, 87 chronology of rural, 88 chronology of urban, 88 commercialization of (zhufang shangpinhua), 87 privatization of in urban China and, 34, 86, 259 rent reform (zujin gaige), 87 rural, 94–97 rural-urban migrants and, 131 urban, 48, 87–94 housing tenure, 92 determinants of in urban areas, 110–118 for rural, urban, and migrant households (2002 and 2007), 92 multinomial logit analysis of choice in urban areas (2002 and 2007), 116 multiple tenure choices and, 115

housing wealth, 4, 85 average annual increases in per capita (2002 to 2007), 106 calculating, 97 determinants of in rural areas, 110–129 determinants of in urban areas, 118–124 distribution of across income quintiles, 108 estimates for other countries, 98 gap in between urban and rural areas, 106 Gini coefficient for, 107 income inequality and, 92, 129–131 inequality of, 106–108, 115 market value of housing and, 105 mean of, per capita (2002 and 2007), 105 methodology and data of, 97–104 mortgage data and, 132–133 negative housing equity and, 133 ratios of per capita, between urban, rural, and migrant households (2002 and 2007), 106 hukou, 475. See also rural-urban migrants and migration exclusion from housing refoms and, 94 in 2002 and 2007 CHIP migrant surveys, 446 loosening of regulations, 213 qualifications for status change, 18 reform policies, 17–19, 252 human capital, 1, 416, 418, 420, 422, 423–424 Hu-Wen New Policies, 3, 5–19, 37, 44, 255 housing and, 260 increase in enterprise and property income, 259 nonwork and, 290, 296–298 positive distributional effects, 30 imputed rental income from owner-occupied housing, 4, 34, 50, 57, 81, 86, 98, 200, 259 alternative estimates of, 55, 102, 136–138 alternative estimates of, based on market value vs. equity value of, 102 as part of household income, 261 as slightly equalizing for rural inequality, 203 elderly and, 260 estimates of per capita (2002 and 2007), 109 income inequality and, 110, 271

Index inconsistent data in 2002 NBS and CHIP surveys and, 133–136 market rent estimates, 31, 81, 97, 98, 102, 103, 134, 136, 137, 318 mean income growth and, 55 methodology and data of, 97–104 rate of return estimates, 31, 81, 97, 98, 102, 103, 134, 136, 137 rent-price ratios and, 135 urban inequality and, 256 income. See also asset income; business income; earnings; household income; housing wealth; imputed rental income from owner-occupied housing; rural household income; property income; urban household income; urban-rural income gap; wages Gini coefficients of, 206 high-income groups, 21 labor, 231, 374, 375 self-employment, 19 income growth curves, 255 education levels and, 275 for 1988–1995, 1995–2002, and 2002–2007, 262 in households connected with state sector, the private sector, and those with no workers (2002 and 2007), 274 income inequality, 271–272, 414–415 agricultural taxes and fees and, 219 alternative estimates of imputed rental income from owner-occupied housing and, 81 among children, adults, and elderly (2002 and 2007), 276 asset income and, 34, 37, 77 business income and, 256 China’s ranking worldwide, 30 CHIP surveys as basis for analyses of, 37, 255 cost of living (PPP), with and without adjustments for, 30, 59 decomposition of by income source (2002 and 2007), 58, 63, 268 downsizing of state and collective sector and, 257 earnings and, 343 economic growth and, 1 education and, 277 equalizing processes in, 48, 77

493

estimates with and without cost of living (PPP) adjustments (2002 and 2007), 60 first decade of the 21st century as important juncture in, 36 housing and, 48, 92, 107, 108–110, 260 housing wealth and, 85, 106–108, 129–131 imputed rental income from owner-occupied housing and, 110, 256 income measurement considerations in estimates of, 19–21 increase in, between 2002 and 2007, 4, 256 indices (1988, 1995, 2002, and 2007), 263 international comparisons, 76 key indicators of, 31 mean incomes with alternative weights (2002 and 2007), 79 measurement of, 55 migrants and migration and, 35–36, 37, 59–64 minimum living standard guarantee (dibao) program and, 9, 226 national mean income and (2002 and 2007), 54 past trends in, by year, 48 pensions and, 260 personal income tax and, 366, 381 post-2007 trends, 38 property income and, 34, 77 regional income differences and, 33, 67–72 rural, 77, 198, 202–207, 226 social security programs and, 21 taxation and, 226, 362 unskilled labor and, 252 urban, 48, 255–257, 260, 267–271, 281–283 urban-rural income gap as contributing factor to, 32, 48, 64–67, 77, 197 work status and ownership sector, 273 income measurement and definition. See also CHIP income; NBS income alternative calculation of, 52 measurement preference, 52 rural as net household income per capita, 28 urban as disposable household income per capita, 28 income tax. See personal income tax income, business. See business income

494 inequality indices. See Gini coefficient; Lorenz curve; Theil indices intertemporal decompositions, 425, 426, 438, 439–440 intratemporal decompositions, 425–426, 438–439 job tenure, 257, 290 Juhn-Murphy-Pierce decompositions, 353–358 kernel density estimations of hourly earnings, 343 Labor Contract Law, 12 Labor Health Insurance (laobao yiliao) Program, 7 labor income. See also earnings; wages inequality in, 231 Personal Income Tax Law (PITL) and, 374, 375 labor-market trends, 232–234, 252, 289, 296–298 competition in urban due to influx of rural migrant workers, 290, 385 discussion about issues related to, 291 employment and unemployment by gender and age, 390 labor force and employment in China (1995–2007), 233 labor force defined, 307 labor policies and minimum wage regulations, 12 labor surplus, 231 laws and regulations for female workers, 386 privatization of state-owned enterprises and, 384, 385 scarcity of unskilled labor, 252 labor mobility. See rural-urban migrants and migration laid-off (xiagang) workers, 291, 297, 385, 387 Land Administration Law, 95 laws Company Law, 336 Compulsory Education Law, 151 Labor Contract Law, 12 Land Administration Law, 95 Law of Population and Family Planning, 421

Index Law of Regional National Autonomy, 422 Law of the People’s Republic of China on Employment Contracts, 298 Law on the Protection of Women’s Rights and Interests, 386 National Program for Women’s Development (2001–2010), 386 National Unemployment Insurance Rules, 297 Personal Income Tax Law (PITL), 13, 373–376 Real Rights Law, 96 Regulations for Enterprise Minimum, 12 Regulations for Unemployment Insurance, 8 Regulations on the Administration of Land for Housing in Villages and Rural Townships, 95 Regulations on Urban Nationality Work, 422 Rules of Classifying the Nationality of Chinese Citizens, 420 Tentative Stipulations on Private Enterprises, 335 Lewis model, 231, 251 Lorenz curve, 54, 56, 62 defined, 54 of household per capita income (2002 and 2007), 56 of migrant per capita income (2002 and 2007), 62 mean logarithmic deviation (MLD) inequality index, 54, 60, 79, 80, 263 medical insurance policies, 7–8 middle-aged and older workers, 297, 307, 310, 326 education levels and gender of, 292 gender differences among, 309 migrants. See rural-urban migrants and migration migration. See rural-urban migrants and migration Mincerian earnings equation, 345 minimum living standard guarantee (dibao) program, 5, 9–12, 45, 221–224, 226, 266 official data on participation in, 224 participation rates by region, 223 poverty and, 225

Index statistics on individuals in dibao vs. non-dibao households, 222 minimum wage (zuidi gongzi) regulations, 12, 239, 242 migrant wages and, 12 minorities. See ethnic minorities; Han-minority wage gap mortgage data, 132–133, 137 mortgage debt, 101 Musgrave Thin (MT) index, 377–381 decomposition of effects of horizontal equity and vertical equity, 379 definition of and formula for, 366 personal income tax, 378 National Program for Women’s Development (2001–2010), 386 National Unemployment Insurance Rules, 297 NBS household surveys, 21, 23, 26, 49, 445 comparison of CHIP and NBS per capita household incomes, 28 inconsistent data on urban rental values in, 133–136 number of households covered in, 449 rates of growth in household income and, 44 rural-urban migrants underrepresentation in, 27 total gross income of urban households and, 370 vs. CHIP surveys, 28, 448–450 NBS income, 28, 200 disposable household per capita income calculations, 52 measurement and definition of, 19–21 poverty and, 73 NBS labor-market surveys, 297 negative housing equity, 133 net transfer income, 271 rural income and, 202 New Rural Pension System (xinxing nongcun shehui yanglao baoxian), 7, 8 New Socialist Countryside, 197 nonwork and nonworkers in urban China, 273, 278, 290, 326–327 age and, 292, 304–316 by category, age, and gender (1988, 1995, 2002, and 2007), 305, 306 CHIP urban survey data and, 300 data and description of, 300–307

495

determinants of among persons ages 30–55/60 (1995, 2002, and 2007), 314 determinants of among young adults, 311 early retired as, 290, 303 education and students and, 292, 293–296, 309 gender and, 292, 298–299, 317 income of other household members and, 291 increases in per year post-economic growth, 289 local labor-market conditions, 292 personal income and economic well-being of, 321–323 residual category, 304 spatial differences, 300 unemployment and labor trends pre- and post-economic restructuring, 296–298 Northeast strategy, revival of (zhenxing dongbei), 46 Oaxaca-Blinder decompositions earnings gap across ownership sectors and, 335, 348–353 gender-wage gap and, 394–395 of log hourly wages, 350 occupational class, 422 one-child policy, 421 ownership sectors. See also earnings gap across ownership sectors; labor-market trends; nonwork and nonworkers in urban China; state-owned enterprises (SOEs) categories of, defined, 339 coastal premiums in SOEs and UCEs, 348 data overview of, 337–341 descriptive statistics on individual characteristics by, 340 earning differential statistics and, 341–345 economic reforms and, 85, 335–337 gender-wage gap and, 348, 398 hourly earning determinants, 345–348 hourly wage functions by (2002), 346 hourly wage functions by (2007), 347 income distribution and, 271–272 Oaxaca-Blinder decomposition of log hourly wages, 350 work experience and, 347 P index, 366

496

Index

pensions, 6–7, 271, 278 in urban areas, 6 income inequality and, 260 reform of urban system, 6 personal income tax, 362–363, 365–366, 377–381 average and alternative rates, 370 average rate by decile, 377 international comparisons of, 379 mean business operating income and the proportion of individuals (non-) reporting, 372 mean income and proportion of individuals (non-) reporting, 371 methodology and data issues in study of, 366–376 Musgrave Thin (MT) index decomposition of effects of horizontal equity and vertical equity, 379 nonreporting of, 370–372 reforms of, 13 taxable income categories, deductions and rates of, 382 understatement of taxation in dataset, 373 Personal Income Tax Law (PITL), 13, 373–376 population control measures, 232 population weights, 51, 52, 55, 65, 104, 162, 465, 466–470 provincial, 471–472 regional, 470–471 poverty, 16, 72–76, 217 absolute poverty lines, 47, 73, 263, 281 absolute vs. relative, 36 among individuals living in households primarily connected to the state sector, the private sector, and those with no workers (2002 and 2007), 273 calculation of poverty lines for urban and rural, 74 decline in absolute, 36 education and, 144, 277 Foster, Greer and Thorbecke (FGT) indices of, 264 in autonomous ethnic areas, 417 in low- vs. high-income cities, 256, 281, 282 international PPP absolute line threshold, 73 key indicators of, 31

low-income (dishouru) line, 11 measurement of absolute, 49 minimum living standard guarantee (dibao) program and, 225 national incidence and composition of (2002 and 2007), 74 of children, adults, and elderly, 274, 276 of migrants, 75 poverty function, 284 PPP and, 74 predicted probabilities of urban relative poverty and affluence (2002 and 2007), 280 rates of, in migrant and nonmigrant households (2002 and 2007), 217 reduction record in China, 48, 72, 78 regional dimensions of, 49, 75, 76 relative poverty lines, 73 rural. See rural poverty urban, 36, 73, 256, 261–267 poverty, rural. See rural poverty prices. See also cost of living; purchasing power parity (PPP) consumer, 53 private enterprises, 258–259, 332 lifting of prohibitions on, 85 private property, 34 income inequality and, 57 private property rights, 85 property income. See also asset income; housing wealth categories of in PITL, 374 imputed rent and, 34 in rural China, 225 income inequality and, 34, 77 NBS urban income measure and, 19 proportion of to total household income, 259, 268 provinces, 46, 50, 51, 111, 445–448, 451, 454 CHIP survey compatibility and, 235 dibao participation rates in, 223 in the CHIP surveys (1988 through 2007), 234, 467 in the CHIP urban survey (2002 and 2007), 261 migrant wage contributions, by province, 216 population weights and, 471–472 purchasing power parity (PPP). See also cost of living

Index cost of living adjustments and, 30, 53, 59, 200 inequality with regional adjustments in, 67–72 poverty threshold and, 74 rural inequality and, 203 Rain and Dew Program (yulu jihua), 17 real estate. See homeownership; housing Real Rights Law, 96 reforms. See also housing reforms; Hu-Wen New Policies; laws; rural policies discrimination and, 420–422 economic, 1, 44, 332 enterprise ownership evolution and, 85, 258, 290, 332–337 gender-wage gap and, 387–389 Han-minority earnings gap and, 415 key indicators of, 10 of education, 14, 45, 145–154 of gender policies, 386 of housing in urban China, 34, 85, 87, 94 of hukou policies, 17–19 of minimum wage (zuidi gongzi) regulations, 12 of personal income tax, 13 of real estate, 85 of rents (zujin gaige), 87 of taxation, 13–14, 363 of unemployment insurance, 8, 298 of urban medical coverage, 7 of urban public pensions, 6 overview of policies, 44–46 regions, 46, 47, 52, 471–472 CHIP sampling strategy and, 22 cost of living differences among, 53, 59, 65 dibao participation rates and, 223 Gini coefficients in, 71 inequality between and within, 33, 48, 67–72 population weights in, 470–471 poverty differences in, 49, 75, 76 uneven economic growth among, 46 urban-rural income gap in, 72 Regulations for Enterprise Minimum Wages, 12 Regulations on Urban Nationality Work, 422 rent, imputed. See imputed rental income from owner-occupied housing retirement. See early retired and retirement

497

rise of central region (zhongbu jueqi) scheme, 46 RUMiCI (Rural-Urban Migrants in China and Indonesia), 22, 459 rural areas, 14–16 determinants of housing wealth in, 110–129 education and, 45, 144–145, 146–154, 188, 197 homeownership in, 104 housing reform in, 94–97 inequality in, 198, 202–207 labor force increase in, 232 one-child policy in, 421 pensions in, 6 self-employment in, 113, 128 surplus labor in, 251 tax and fee reform (nongcun shuifei gaige) in, 14, 226 rural household income. See also imputed rental income from owner-occupied housing agriculture and, 59, 198, 201, 226 average annual growth (2002 to 2007), 205 composition of, 201 data and methods, 200 distribution of, across decile groups, 56 estimates of Gini coefficient of (2002 and 2007), 203 growth in nonagricultural business, transfers, asset, and property, 202, 206, 207, 225 in poor and nonpoor households, 211–213 levels and growth by deciles, 57 mean income per capita by region (2002 and 2007), 81, 200 migrants and migration and, 35–36, 213–218, 225 net transfer income and, 202 richest and poorest groups in the income distribution, 204 rural inequality and, 77, 202–207, 226 rural taxation and, 218–221, 226 urban-rural income gap and, 198 rural policies. See also agricultural taxes and fees giving more, taking less (slogan), 13–14 government subsidies for agricultural production, 197

498

Index

rural policies (cont.) high priority of, 197 housing and land policies, 94–97 minimum living guarantee standard guarantee (dibao) program, 226 minimum living standard guarantee (dibao) program, 45, 221–224 pro, 14–16, 44 pro-agriculture policies, 202 public investments in rural infrastructure, 197 tax and fee reforms, 218 rural poverty. See also minimum living standard guarantee (dibao) program calculation of poverty lines for urban and rural, 74 changes in, 198, 207–213 decline in absolute, 36 in autonomous ethnic areas, 417 majority of all poverty lines as, 75 policies to alleviate, 16–17 poverty line for, 73 rates of, in migrant and nonmigrant households (2002 and 2007), 217 spatial dimensions of, 49, 76 taxation and, 220 rural taxation, 14, 44, 197, 218–221, 226 rural-urban migrants and migration, 46, 61, 244–250 by province, 215 CHIP surveys and, 50, 199 decomposition of migrant income inequality by income source (2002 and 2007), 63 definition of, 47, 60 effect on rural and urban incomes, 46 first inclusion in CHIP survey (2002), 230 gender-wage gap and, 400, 404 growth in the number of, 213 health and, 246 housing and, 94, 104 housing welfare policy and, 131 income inequality and, 37, 59–64 inequality among, 63 international financial crisis and, 25 labor-market trends and, 232–234, 385 poverty rates and, 217 probability of migrating, 245 rural income and, 213–218 unskilled labor and, 290 wages, 12, 236–244, 251

self-employment, 13, 17, 19, 127, 391 in rural areas, 113, 128 in urban areas, 117, 122, 236, 258 of migrants, 236–244 smashing the iron rice bowl (za tiefanwan), 290, 296, 333 social welfare and social security system. See also minimum living standard guarantee (dibao) program; reforms establishment of in 2000, 6 types of assistance, 11 urban-rural income gap and, 21 State Administration of Taxation (SAT), 369 state-owned enterprises (SOEs). See also ownership sectors corporatization of, 332 decrease of employees in, 257, 385 defined, 339 students, 292, 293–296, 304, 309, 321 as nonworkers, 291 living conditions among recent university graduates, 295 taxation. See also agricultural taxes and fees 1994 fiscal reform (tax-sharing reform) and, 363 business tax, 364 consumption tax, 364 corporate income tax, 365 horizontal and vertical equity, 367, 369 income inequality and, 260, 362 indirect taxes and, 363–365 Musgrave Thin (MT) index and, 366 P index as measure of progressivity, 366 phases of reform, 13–14 progressive, 362, 366, 368 rates, 370 rural tax and fee reform (nongcun shuifei gaige), 14, 226 share of major taxes in total tax revenue in selected years after the 1994 fiscal reform, 364 State Administration of Taxation (SAT), 369 value-added tax (VAT), 363 Theil indices, 53, 54, 55, 67, 204, 263 Two Exemptions, One Subsidy policy, 153 unemployment. See also nonwork and nonworkers in urban China insurance program, 8

Index international financial crisis and, 25 labor force participation and, 388 laid-off (xiagang) workers as, 387 missed categories in unemployment rate, 291, 297 National Unemployment Insurance Rules, 297 open, 292 rate by year, 307 rate defined, 304 rate for urban workers, 390 rates for men vs. women, 307, 385 recent graduates and, 290 trends in post-economic restructuring, 296–298 xiagang as, 387 urban areas determinants of housing tenure in, 110–118 determinants of housing wealth in, 118–124 dibao program in, 45 education in, 146, 154 geographical boundaries of, 11 housing in, 91, 93, 260 housing reform policies in, 87–94 income inequality in, 255–257, 260, 281–283 multinomial logit analysis of housing tenure choice in (2002 and 2007), 116 rise of private enterprises in, 258–259 self-employment in, 117, 122, 258 urbanization, 46 urban household income, 57, 267–271, 281–283 defined, 261 distribution of, across decile groups, 56 imputed rental income from owner-occupied housing and, 259 levels and growth by deciles, 57 mean income per capita by region (2002 and 2007), 81 overall trends in, 261–267 pensions and, 260 wage earnings’ share of total income, 259 wages and, 232

499

urban net transfers, 59, 77 urban-rural income gap, 46, 47 by region, 72 calculation of contribution to inequality, 107 contribution to inequality, 32, 48, 64–67, 77, 197 cost of living adjustments and, 65, 66 education and, 33 educational attainment among minorities and, 416, 417, 418 international comparisons, 65 ratio of in 2007 CHIP survey, 233 regional dimension of, 32 rural income growth and, 198 social security programs and, 21 widening of between 2002 and 2007, 198 widening of since late 1980s, 32 value-added tax (VAT), 363 wages. See also earnings; gender-wage gap; Han-minority earnings gap; incme; urban-rural income gap migrant, 12, 236–244, 251 minimum wage (zuidi gongzi) regulations, 12, 239, 242 wealth. See also housing wealth cost of living and, 59 data on, 21 variations among households, 1 western development strategy (xibu dakaifa zhanl¨ue), 46 women. See female workers; gender; gender-wage gap; homemakers work unit, 8, 90–92, 96, 257, 297, 387, 395 as social security in pre-reform era, 5 xiagang, to keep ties with, 8, 297, 387 working time, 343 young adults, 289, 307–313 three states of nonwork among, 309 unemployment rates for, 307