African Successes, Volume I: Government and Institutions 9780226316369

Citation preview

African Successes, Volume I

National Bureau of Economic Research Conference Report

African Successes, Volume I Government and Institutions

Edited by

Sebastian Edwards, Simon Johnson, and David N. Weil

The University of Chicago Press Chicago and London

Sebastian Edwards is the Henry Ford II Professor of International Economics in the Anderson Graduate School of Management at the University of California, Los Angeles. Simon Johnson is the Ronald A. Kurtz (1954) Professor of Entrepreneurship and professor of global economics and management at the MIT Sloan School of Management. David N. Weil is the James and Merryl Tisch Professor of Economics at Brown University. All three editors are research associates of the NBER.

The University of Chicago Press, Chicago 60637 The University of Chicago Press, Ltd., London © 2016 by the National Bureau of Economic Research Printed in the United States of America 25 24 23 22 21 20 19 18 17 16

1 2 3 4 5

ISBN-13: 978-0-226-31622-2 (cloth) ISBN-13: 978-0-226-31636-9 (e-book) DOI: 10.7208/chicago/9780226316369.001.0001 Library of Congress Cataloging-in-Publication Data Names: Edward, Sebastian, 1953– editor. | Johnson, Simon, 1963– editor. | Weil, David N., editor. Title: African successes: government and institutions / edited by Sebastian Edwards, Simon Johnson, and David N. Weil. Other titles: National Bureau of Economic Research conference report. Description: Chicago; London : Chicago and London, 2016– | Series: National Bureau of Economic Research conference report. Identifiers: LCCN 2015050080 | ISBN 9780226316222 (cloth : alk. paper) | ISBN 9780226316369 (e-book) | ISBN 9780226316055 (cloth : alk. paper) | ISBN 9780226316192 (e-book) | ISBN 9780226315720 (cloth : alk. paper) | ISBN 9780226315867 (e-book) | ISBN 9780226315553 (cloth : alk. paper) | ISBN 9780226315690 (e-book) Subjects: LCSH: Economic development—Africa. | Africa—Economic conditions—21st century. Classification: LCC HC800 .A56873 2016 | DDC 330.96—dc23 LC record available at http://lccn.loc.gov/2015050080 ♾ This paper meets the requirements of ANSI/NISO Z39.48–1992 (Permanence of Paper).

National Bureau of Economic Research Officers Martin B. Zimmerman, chairman Karen N. Horn, vice chairman James M. Poterba, president and chief executive officer Robert Mednick, treasurer

Kelly Horak, controller and assistant corporate secretary Alterra Milone, corporate secretary Denis Healy, assistant corporate secretary

Directors at Large Peter C. Aldrich Elizabeth E. Bailey John H. Biggs John S. Clarkeson Don R. Conlan Kathleen B. Cooper Charles H. Dallara George C. Eads Jessica P. Einhorn

Mohamed El-Erian Linda Ewing Jacob A. Frenkel Judith M. Gueron Robert S. Hamada Peter Blair Henry Karen N. Horn John Lipsky Laurence H. Meyer

Michael H. Moskow Alicia H. Munnell Robert T. Parry James M. Poterba John S. Reed Marina v. N. Whitman Martin B. Zimmerman

Directors by University Appointment Timothy Bresnahan, Stanford Pierre-André Chiappori, Columbia Alan V. Deardorff, Michigan Ray C. Fair, Yale Edward Foster, Minnesota John P. Gould, Chicago Mark Grinblatt, California, Los Angeles Bruce Hansen, Wisconsin–Madison

Benjamin Hermalin, California, Berkeley Marjorie B. McElroy, Duke Joel Mokyr, Northwestern Andrew Postlewaite, Pennsylvania Cecilia Rouse, Princeton Richard L. Schmalensee, Massachusetts Institute of Technology David B. Yoffie, Harvard

Directors by Appointment of Other Organizations Jean-Paul Chavas, Agricultural and Applied Economics Association Martin J. Gruber, American Finance Association Arthur Kennickell, American Statistical Association Jack Kleinhenz, National Association for Business Economics William W. Lewis, Committee for Economic Development Robert Mednick, American Institute of Certified Public Accountants

Alan L. Olmstead, Economic History Association Peter L. Rousseau, American Economic Association Gregor W. Smith, Canadian Economics Association William Spriggs, American Federation of Labor and Congress of Industrial Organizations Bart van Ark, The Conference Board

Directors Emeriti George Akerlof Jagdish Bhagwati Carl F. Christ Franklin Fisher

George Hatsopoulos Saul H. Hymans Rudolph A. Oswald Peter G. Peterson

John J. Siegfried Craig Swan

Relation of the Directors to the Work and Publications of the National Bureau of Economic Research 1. The object of the NBER is to ascertain and present to the economics profession, and to the public more generally, important economic facts and their interpretation in a scientific manner without policy recommendations. The Board of Directors is charged with the responsibility of ensuring that the work of the NBER is carried on in strict conformity with this object. 2. The President shall establish an internal review process to ensure that book manuscripts proposed for publication DO NOT contain policy recommendations. This shall apply both to the proceedings of conferences and to manuscripts by a single author or by one or more co-authors but shall not apply to authors of comments at NBER conferences who are not NBER affiliates. 3. No book manuscript reporting research shall be published by the NBER until the President has sent to each member of the Board a notice that a manuscript is recommended for publication and that in the President’s opinion it is suitable for publication in accordance with the above principles of the NBER. Such notification will include a table of contents and an abstract or summary of the manuscript’s content, a list of contributors if applicable, and a response form for use by Directors who desire a copy of the manuscript for review. Each manuscript shall contain a summary drawing attention to the nature and treatment of the problem studied and the main conclusions reached. 4. No volume shall be published until forty-five days have elapsed from the above notification of intention to publish it. During this period a copy shall be sent to any Director requesting it, and if any Director objects to publication on the grounds that the manuscript contains policy recommendations, the objection will be presented to the author(s) or editor(s). In case of dispute, all members of the Board shall be notified, and the President shall appoint an ad hoc committee of the Board to decide the matter; thirty days additional shall be granted for this purpose. 5. The President shall present annually to the Board a report describing the internal manuscript review process, any objections made by Directors before publication or by anyone after publication, any disputes about such matters, and how they were handled. 6. Publications of the NBER issued for informational purposes concerning the work of the Bureau, or issued to inform the public of the activities at the Bureau, including but not limited to the NBER Digest and Reporter, shall be consistent with the object stated in paragraph 1. They shall contain a specific disclaimer noting that they have not passed through the review procedures required in this resolution. The Executive Committee of the Board is charged with the review of all such publications from time to time. 7. NBER working papers and manuscripts distributed on the Bureau’s web site are not deemed to be publications for the purpose of this resolution, but they shall be consistent with the object stated in paragraph 1. Working papers shall contain a specific disclaimer noting that they have not passed through the review procedures required in this resolution. The NBER’s web site shall contain a similar disclaimer. The President shall establish an internal review process to ensure that the working papers and the web site do not contain policy recommendations, and shall report annually to the Board on this process and any concerns raised in connection with it. 8. Unless otherwise determined by the Board or exempted by the terms of paragraphs 6 and 7, a copy of this resolution shall be printed in each NBER publication as described in paragraph 2 above.

Contents

Series Introduction Sebastian Edwards, Simon Johnson, and David N. Weil

1

Volume Introduction Sebastian Edwards, Simon Johnson, and David N. Weil

5

I. Political Economy and Conflict 1. Healing the Wounds: Learning from Sierra Leone’s Postwar Institutional Reforms Katherine Casey, Rachel Glennerster, and Edward Miguel 2. The Political Economy of Government Revenues in Postconflict Resource-Rich Africa: Liberia and Sierra Leone Victor A. B. Davies and Sylvain Dessy 3. Does Decentralization Facilitate Access to Poverty-Related Services? Evidence from Benin Emilie Caldeira, Martial Foucault, and Grégoire Rota-Graziosi

15

33

57

vii

viii

Contents

4. Demographic Pressure and Institutional Change: Village-Level Response to Rural Population Growth in Burkina Faso 103 Margaret S. McMillan, William A. Masters, and Harounan Kazianga 5. New Tools for the Analysis of Political Power in Africa Ilia Rainer and Francesco Trebbi

145

II. Government Regulation and Basic Institutions 6. Deals versus Rules: Policy Implementation Uncertainty and Why Firms Hate It Mary Hallward-Driemeier, Gita Khun-Jush, and Lant Pritchett 7. The Unofficial Economy in Africa Rafael La Porta and Andrei Shleifer 8. State versus Consumer Regulation: An Evaluation of Two Road Safety Interventions in Kenya James Habyarimana and William Jack

215

261

307

III. Long-Run Assessments 9. Fifteen Years On: Household Incomes in South Africa Murray Leibbrandt and James Levinsohn 10. Is Tanzania a Success Story? A Long-Term Analysis Sebastian Edwards Contributors Author Index Subject Index

333

357 433 435 441

Series Introduction Sebastian Edwards, Simon Johnson, and David N. Weil

In the 1950s and early 1960s, there was a great deal of optimism about the prospects for economic development in sub-Saharan Africa. By early in the twenty-first century, the prevailing consensus had become much more negative—and for good reason. Decades of civil war, repeated state failure, corruption, and disappointing private sector performance characterized much of Africa’s postindependence experience. A wave of prominent papers in the economic literature tried to dig more deeply into the causes of these problems, with some scholars putting them in a broader comparative perspective and others focusing on the specifics of the African experience. There was no shortage of deep causes suggested as explanations for repeated African disappointments, including weak rule of law, a lack of democracy, colonial inheritance, the impact of the slave trade, the burden of tropical disease, some form of “resource curse,” and ethno-linguistic divisions among the population. The NBER Africa project, conceived in the middle of the first decade of the twenty-first century, took a different approach. Rather than trying to understand the causes of underperformance in Africa, we decided to focus on finding and understanding more positive aspects of what was happening Sebastian Edwards is the Henry Ford II Professor of International Economics at the University of California, Los Angeles, and a research associate of the National Bureau of Economic Research. Simon Johnson is the Ronald A. Kurtz (1954) Professor of Entrepreneurship and Professor of Global Economics and Management at the MIT Sloan School of Management and a research associate of the National Bureau of Economic Research. David N. Weil is the James and Merryl Tisch Professor of Economics at Brown University and a research associate of the National Bureau of Economic Research. For acknowledgments, sources of research support, and disclosure of the authors’ material financial relationships, if any, please see http://www.nber.org/chapters/c13651.ack.

1

2

Sebastian Edwards, Simon Johnson, and David N. Weil

on that continent south of the Sahara, along several different dimensions studied by economists. Our timing proved good for three reasons. First, scholars were turning their attention increasingly to the details of what was working well or at least better with regard to African development. This new wave of research involved working with or, in many cases creating, data sets suited to careful analysis—including sophisticated use of econometrics. We are fortunate to have involved and supported many leading empirical researchers as they broke new ground in multiple directions. Most likely some of this work would have happened in any case, but at the very least we can claim to have provided an appropriate catalyst for many projects. Second, Africa is changing—and the prospects for nearly a billion people are looking up. A decade or more of relatively good performance does not a miracle make, but across a wide range of countries there has been better economic growth, considerable progress with improving public health and other social indicators, and a range of attempts to improve the performance of the state. Many of the chapters in this series of four volumes address some aspect of this profound and important transformation. Third, the study of Africa is becoming much more integrated with the rest of economics. Just as happened earlier with research on Japan and China, a line of work that previously might have been seen as of purely regional interest can now appear in top journals. The broader trend, of course, has been the resurrection of development economics as a vibrant field. But also encouraging—and a central part of the NBER Africa project—has been the increase in interest among nonspecialists regarding what is actually happening on this dynamic and fascinating continent. Volume I in this series focuses on the most basic building blocks of economic development, including the rule of law, civil conflict, and institutions more broadly. Volume II contains chapters on human capital development in Africa, including some important work on public health improvements, but also research into education on the continent. Volume III examines whether Africa can glean any advantage from being relatively late to economic development, including being able to bypass some large investments (for example, in fixed-line telephones or in traditional branch banking) or being able to learn from others (for example, in the development of an export sector). Volume IV looks at whether recent growth can be sustained, in terms of agricultural development and more broadly. The research presented in these volumes covers a wide territory, in geographic and intellectual terms. However, our project was never intended to be comprehensive. Rather we attempted to act as a catalyst for rigorous and innovative thinking applied to recent African development. The work published here far exceeds our earliest expectations—a reflection, primarily, of how many serious scholars are now deeply engaged with these issues.

Series Introduction

3

It has been a great honor for us, as program directors and now as editors of these volumes, to participate in and help to facilitate this surge in serious interest. In terms of making this possible, we must thank, first and foremost, the Bill & Melinda Gates Foundation for their willingness to encourage economists in this endeavor. We have worked with a range of officials at the Foundation over the past decade; their support has been unstinting and their perspective on development is always refreshing. We would specifically like to thank Sara Sievers for many of our early interactions and Oliver Babson for his subsequent engagement. We also greatly appreciate the ideas and energy of Geoffrey Lamb, Kim Hamilton, Negar Akhavi, Adam Gerstenmier, and Mumukshu Patel. We must also recognize the founding insight and energy of Martin Feldstein, who emphasized, among many other relevant and helpful points, that there is an important link between raising the quality of economic research on an issue and improving the odds that policy discussions will become more informed. Since becoming president of the NBER in 2008, James Poterba has also provided us with great leadership and encouragement—including sage advice at every stage of the process. Our four published volumes present results from thirty-nine research teams, with the findings organized along thematic lines. Of the 100 authors whose names appear on one or more of the chapters in our collection, nineteen are from Africa and thirty-two are affiliated with the NBER as either Faculty Research Fellows or research associates. They range from experienced specialists on African development to scholars who had never previously worked on the continent; in seniority they ran the gamut from newly appointed assistant professors to distinguished, established thought leaders. Every project included a significant on-the-ground component. Some research teams combined money we provided with other funds, for example, to support the collection of very large field data sets. In other cases, funding from the NBER made it possible for researchers to interview policymakers, business people (both large and small), farmers, health workers, or others engaged in the process of economic development. Results from the research project were presented at a series of conferences in the United States and in Africa. Given the large size of our project and our explicit goal of including scholars who had not previously been working on Africa, we began with a background conference in February 2008, at the NBER in Cambridge, MA, bringing together researchers and policymakers from the United States, Africa, and Europe. The conference featured presentations on current research and an overview of available data, aimed especially at scholars who were new to the area. Our next meeting, also in Cambridge, MA, in February 2009 was a preconference at which preliminary findings from ten papers were presented. This was followed by a conference in December of the same year, again in

4

Sebastian Edwards, Simon Johnson, and David N. Weil

Cambridge, MA, that took a hybrid form: final versions of five projects were presented, as were preconference versions of another twelve projects. This was followed in July of 2010 by a conference in Accra, Ghana, held in cooperation with the Institute of Statistical, Social, and Economic Research (ISSER). Our final research conference was held in Zanzibar, Tanzania, in August 2011, in cooperation with the Bank of Tanzania. One goal of our project from inception, with full support from the Gates Foundation, was to help connect economic research with the African policy community. We arranged some interactions along these lines throughout our project, but we were fortunate to be able to make a special effort at the end, with a meeting devoted specifically to the Next Macroeconomic Challenges in Africa, held again in Zanzibar, in December 2012, also in cooperation with the Bank of Tanzania. There are many people to thank for the successful completion of this project. Benno Ndulu, governor of the Bank of Tanzania, provided wise guidance throughout the process and particularly helped us to focus on narrowing the gap between research and policy, including our two fruitful meetings in Zanzibar. His team at the Bank of Tanzania was most helpful in many ways, and we have special thanks for Patricia Mlozi, Mechtilda Mugo, Msafiri Nampesya, and Pamella Lowassa-Solomon. Our conference in Accra benefited greatly from the engagement and support provided by Ernest Aryeetey, Kwesi Botchwey, and Robert Osei. And we had great cooperation in many ways across the entire project with African Economic Research Consortium (AERC); thanks to Olu Ajakaiye, William Lyakurwa, and Lemma Senbet for making this possible. We would like to thank everyone who attended our various conferences and who worked hard on all dimensions of these research projects. We are also most grateful for all the inputs received from members of the Advisory Committee, formed especially for this project: Robert Bates, Paul Collier, Martin Feldstein, Benno Ndulu, Franklyn Prendergast, Antoinette Sayeh, Nicholas Stern, and John Taylor. The entire project ran smoothly due to the hard work, precision, and attention to detail of Elisa Pepe at the NBER. Elisa was with us from beginning to end, and words cannot sufficiently express our gratitude for all she has done. Others at the NBER provided outstanding help on a wide variety of tasks. Our meetings in the United States and in Africa were organized with exemplary competence by Carl Beck of the NBER’s conference department. For all their help in the production, management, and dissemination of research, we would also like to thank Alex Aminoff, Laura Bethard, Daniel Feenberg, Helena Fitz-Patrick, Wayne Gray, Steve Harriman, and Alterra Milone. And for their assistance with the data portal, we are grateful to Binh Thanh Le, Dimitry Legagneur, and C. Michelle Tejada.

Volume Introduction Sebastian Edwards, Simon Johnson, and David N. Weil

As colonial empires began to crumble in the decade after World War II, there were high expectations for growth in Africa south of the Sahara. The region had significant natural resources, a growing population, and strong ties with Europe that promised trade, aid, and helpful capital inflows. Independence was widely regarded as creating great opportunities. Over the next three decades, Africa’s economic outcomes were, for the most part, profoundly disappointing.1 Natural resources proved, at best, to be a mixed blessing—even sometimes fueling civil war. Increasing population often outstripped the resources governments could devote to health and education, and the end of colonialism in this part of the world was definitely not associated with sustained growth in incomes per capita. Trade remained sluggish, aid proved disappointing, and capital inflows did not have a more broadly transformative effect. This volume, the first of four books resulting from our project, provides expert assessments regarding some of the basics of recent economic development in Africa: the rule of law, the extent of political stability, the role of foreign aid, and what happens after civil war. The chapters here also take up key details of how the state operates, including actions and policies that Sebastian Edwards is the Henry Ford II Professor of International Economics at the University of California, Los Angeles, and a research associate of the National Bureau of Economic Research. Simon Johnson is the Ronald A. Kurtz (1954) Professor of Entrepreneurship and Professor of Global Economics and Management at the MIT Sloan School of Management and a research associate of the National Bureau of Economic Research. David N. Weil is the James and Merryl Tisch Professor of Economics at Brown University and a research associate of the National Bureau of Economic Research. For acknowledgments, sources of research support, and disclosure of the authors’ material financial relationships, if any, please see http://www.nber.org/chapters/c13386.ack. 1. Africa’s disappointing growth between 1960 and 2000 is the subject of a two-volume work (see Ndulu et al. 2008).

5

6

Sebastian Edwards, Simon Johnson, and David N. Weil

are more or less conducive to economic development. Our authors also provide some longer-run assessments of the growth process. Based on the evidence and analyses presented in this volume, we are convinced that over the past two decades Africa has made important progress and that many of the problems of the postcolonial period have now receded. Currently, sub-Saharan Africa has strong prospects for further development, even including potential convergence toward the income levels seen in today’s middle-income countries. But there is a long way to go and some of the obstacles remain severe. The findings of our project stress the ways in which these vary across countries and sometimes within countries. Political Economy and Conflict One of the insights that has informed the literature on economic growth and development over the last two decades has been an appreciation for the paramount importance of politics and social peace for any kind of reasonable economic outcome. For this reason we supported—and volume I contains—work on exactly these issues. Chapter 1, “Healing the Wounds: Learning from Sierra Leone’s Postwar Institutional Reforms,” by Katherine Casey, Rachel Glennerster, and Edward Miguel, deals with what must be considered one of the most fundamental issues for development: Can countries recover from civil wars, in the sense of building social cooperation, solid institutions, and an agreement not to engage again in destabilizing violence? The authors are cautiously optimistic. The brutal civil war in Sierra Leone did not completely destroy the capacity for local collective action, and while politics remain heavily influenced by ethnicity, voters are willing to cross ethnic lines when they have sufficient information. Ethnic segregation also does not impede the provision of local public goods, and decentralization of government functions can work, at least sometimes. In chapter 2, Victor A. B. Davies and Sylvain Dessy stress several lessons that can be drawn from the African experience in “The Political Economy of Government Revenues in Postconflict Resource-Rich Africa: Liberia and Sierra Leone.” Mineral and energy resources offer a great opportunity for accelerated development, including situations where peace has only recently been established, but rushing to bring in international companies can have unintended and unfortunate consequences. At the same time, there is a great deal that countries can learn from each other about how to structure and enforce contracts in ways that ensure benefits are shared more broadly. There are also important responsibilities for countries where mining companies and oil companies are headquartered, and from which they draw managerial talent. The boom in Africa since the 1990s is partly about natural resources, but it is also about a more robust approach to poverty reduction. This is largely

Volume Introduction

7

about creating a state—including a central government, but also its various regional and local components—that is focused on delivering services that people actually need. In this context, it is most encouraging to see governments experimenting with alternative ways of structuring both revenues and spending. An important role for economists is to study carefully what is tried and to distill lessons in terms of what works and along which dimensions. This is precisely the line taken by Emilie Caldeira, Martial Foucault, and Grégoire Rota-Graziosi in chapter 3, “Does Decentralization Facilitate Access to Poverty-Related Services? Evidence from Benin.” Decentralization to the local level can have important advantages, including making government more responsive to what people really need to improve their lives. But no kind of initiative in this sphere is without complications, and the authors study carefully both positive and more nuanced aspects of recent experience in Benin. The goal of our project was not to create a comprehensive picture of current circumstances in Africa, nor to survey the full range of government and private sector initiatives. Rather, we wanted to encourage research into leading questions—and this study of Benin is a nice example. Other researchers will hopefully find ways to examine this issue in other contexts and, over time, the field will establish a broader systematic assessment. In chapter 4, “Demographic Pressure and Institutional Change: VillageLevel Response to Rural Population Growth in Burkina Faso,” Margaret S. McMillan, William A. Masters, and Harounan Kazianga take up the issue of whether population increase can lead to helpful changes in institutions, such as the creation of property rights. In some situations, more crowding onto a finite amount of land can lead to greater conflict—and exactly the kind of downward spiral experienced by some parts of West Africa after 1960. But, as the authors show, other outcomes are also possible. It was very important for us to understand more precisely the conditions in which legitimate pro-growth rules and norms develop. At least in the case of Burkina Faso, seen through the lens of their study, there is some room for optimism. The same is true for the issues raised in chapter 5, “New Tools for the Analysis of Political Power in Africa,” by Ilia Rainer and Francesco Trebbi. The authors have done a great service by collecting and organizing very detailed data on the ethnic background of leading political figures across fifteen African countries. There is much more to do in this area in terms of linking these characteristics to both decision making and to outcomes. The work reported here should help other scholars push forward the conceptual frontier, for example understanding when ethnic identity gets in the way of reasonable economic policies. And, of course, the deeper question and concern always lurks: What prevents any part of Africa from sinking back into the kind of destructive conflict that characterized too much of the postindependence period? Still, there is a very real possibility that ethnic-based conflicts have receded, at least in some parts of the continent.

8

Sebastian Edwards, Simon Johnson, and David N. Weil

Government Regulation and Basic Institutions Lest anyone think that looking for success in Africa has led us collectively to become starry eyed, chapter 6 provides a substantial bucket of cold water. In “Deals versus Rules: Policy Implementation Uncertainty and Why Firms Hate It,” Mary Hallward-Driemeier, Gita Khun-Jush, and Lant Pritchett find that, across a large number of African countries, many firms face some form of “closed” environment, in which deals on reasonable terms are limited to firms with favored characteristics. Alternatively, there is a disordered environment in which it is difficult to reliably predict policy action outcomes, even when firms undertake influence activities. Countrylevel laws and regulations on the books seem to count for little; it is all about “implementation”—meaning how individual officials approach their jobs. This is often done at the local level and is influenced by ethnic and other local characteristics. The World Bank’s Doing Business indicators have been hugely influential and, in some places, prove extremely informative. But in the African countries examined carefully here, the de facto outcomes are quite different from what one would expect from looking at the de jure rules. Unless and until the relationship between government and the private sector improves— including along the precise dimensions identified by these authors—it will be difficult to accelerate growth in Africa. In chapter 7, “The Unofficial Economy in Africa,” Rafael La Porta and Andrei Shleifer investigate a related but distinct phenomenon. To some extent, entrepreneurs react to an unfavorable environment and to the absence of clear rules and regulations by not registering their firms and otherwise trying to stay beneath the radar of the official sector. La Porta and Shleifer find that these unofficial firms have low productivity compared with registered firms, both across a wide range of twenty-four countries and in their detailed investigation of Madagascar, Mauritius, and Kenya. Crossing this chasm will not be easy, and the obstacles may be more serious even than weak rule of law and pervasive corruption. In chapter 8, “State versus Consumer Regulation: An Evaluation of Two Road Safety Interventions in Kenya,” James Habyarimana and William Jack look more deeply at a specific and very important aspect of the privatepublic interaction. Road safety is now a serious issue in much of Africa, as the combination of modern vehicles and less-than-fully-modern roads too often proves fatal. The role of private bus companies running minibuses over long distances is particularly important, both for transportation and— tragically—as a cause of death. The authors study the effects of alternative ways to make this mode of transportation safer. One approach is to require engine modifications that limit speed. This makes sense from an engineering perspective, but incentives also matter. If the driver makes more money or

Volume Introduction

9

otherwise derives utility from going faster, then outcomes are not necessarily as intended. An alternative is to encourage passengers to put pressure on drivers to be more careful. The authors find themselves at a fascinating intersection between psychology, culture, and economics. Who will speak up, under what conditions, and to what effect? There are presumably applications and lessons across a wide range of social situations, both for the private sector and for the public sector. Long-Run Assessments We close volume I with two chapters that look at how institutional change played out over long periods of time, with a particular emphasis on how one measures and assesses both causality and economic progress. In volume IV we return to several more studies that follow developments in individual countries. Chapter 9 belongs to the great tradition of NBER papers that provide deep insight into the data and help to shape a field of study. In “Fifteen Years On: Household Incomes in South Africa,” Murray Leibbrandt and James Levinsohn use national household surveys from 1993 and 2008 to assess who has experienced what kind of income gains since the end of apartheid. Real gross domestic product (GDP) per capita since the democratic elections in 1994 has risen by an average of close to 1.5 percent per year—an apparently encouraging performance. However, measuring incomes at the household level (adjusting for household size) paints a potentially different and in some ways less positive picture. Including all household members, even those not participating in the labor market, confirms that average per capita incomes increased and that growth has been shared across the distribution of income. However, the increases are modest and inequality has increased—even beyond the high degree of inequality that was a legacy of apartheid. Access to basic services such as water, housing, and electricity has improved, but unemployment has risen to very high levels. Strong spending by the state on education and services led to measurable improvements in levels of education and access to essential services, but these improved endowments did not translate into generalized increases in real incomes. What dampened the translation between improved endowments and improved real incomes? The authors are not yet in a position to provide a definitive answer, but other evidence suggests there has been a twist in the returns to education in South Africa that has lowered the returns to education for all but the highest levels of schooling. Foreign aid has played an important role in discussions on African development. Recent controversies have pitched scholars that, mostly on the

10

Sebastian Edwards, Simon Johnson, and David N. Weil

bases of cross-country data, believe that aid has been ineffective (or even detrimental), and those that argue that—when properly dispensed—international assistance can help reduce poverty and further development. In chapter 10, “Is Tanzania a Success Story? A Long-Term Analysis,” Sebastian Edwards analyzes the role played by foreign aid in Tanzania between 1961 and 1996. As Edwards shows, foreign assistance—most of which took the form of bilateral aid—did not support growth in Tanzania during the early years (especially after 1967 when the policies of the Arusha Declaration were implemented), and it contributed to economic decline in the 1970s and 1980s. However, foreign aid was helpful for the subsequent post-1996 recovery of the country, when the aid community and the Tanzanian authorities found ways of coordinating their actions. Edwards’s work has subsequently been expanded into a book, Toxic Aid, published by Oxford University Press in mid-2014. Conclusion Taken together, the chapters in this volume paint a complex picture of how government and institutions shape development, and also the factors that determine how well institutions function. The vision of government-led development that accompanied the birth of many new countries in Africa in the 1950s and 1960s did not survive prolonged contact with reality. It was replaced by a vision of government as part of the problem: the people charged with enacting policies often followed political or ethnic self-interest, if they were not simply corrupt; natural resource revenues and foreign aid, both of which had seemed like ideal sources of the revenues needed to speed development instead turned out to have toxic effects on the political system; and waiting in the wings when politics failed were ethnic conflict and state failure. But experience in Africa, including closer investigation by the economists represented in this volume, suggests that such a dismal assessment is not fully warranted either. Countries with weak governance and a great deal of corruption can still contain pockets of functionality, and in countries as diverse as Sierra Leone, South Africa, and Tanzania, policies and political systems have improved. There is a lot that remains to be done, if growth across the continent is really to improve the lives of current and future generations. What exactly is involved in sustaining growth and sharing its benefits is the focus of the remaining volumes in this series. Volume II digs deeply into issues of human capital. Volume III looks at the specifics of modernization across some key sectors, including finance, mobile phones (including the dimension of mobile banking), and exports. Volume IV examines the sustainability of growth in more detail, from the perspective of agriculture and food production, as well as improvements

Volume Introduction

11

in market efficiency. Volume IV also includes studies of recent experience in countries that have—so far—managed to sustain some broad form of economic success.

References Ndulu, Benno J., Stephen A. O’Connell, Jean-Paul Azam, Robert H. Bates, Augustin K. Fosu, Jan Willem Gunning, and Dominique Njinkeu, eds. 2008. The Political Economy of Economic Growth in Africa, 1960–2000. New York: Cambridge University Press.

1

Healing the Wounds Learning from Sierra Leone’s Postwar Institutional Reforms Katherine Casey, Rachel Glennerster, and Edward Miguel

1.1

Introduction

Academics and development practitioners agree that strengthening the transparency, accountability, and inclusiveness of institutions could be important determinants of economic performance (Engerman and Sokoloff 1997; Acemoglu, Johnson, and Robinson 2001; Banerjee and Iyer 2005). Yet they also acknowledge that it remains unclear what types of interventions could successfully make progress toward these objectives. Such concerns take on heightened urgency in a country like Sierra Leone, which has suffered decades of extreme poverty and recently emerged from a devastating civil war. In this context, the objectives of inclusive and participatory governance are twofold. First, enhancing accountability aims to provide a more effective vehicle to channel government and donor resources toward the reconstruction of public infrastructure and restoration of basic services. Second, creating avenues for public participation allows citizens to voice and seek redress for grievances regarding government incompetence and Katherine Casey is assistant professor of political economy at the Stanford Graduate School of Business. Rachel Glennerster is executive director of the Abdul Latif Jameel Poverty Action Lab (J-PAL). Edward Miguel is the Oxfam Professor of Environmental and Resource Economics and faculty director of the Center for Effective Global Action at the University of California, Berkeley, and a research associate of the National Bureau of Economic Research. This chapter draws together findings from a number of different research projects. The authors gratefully acknowledge generous financial support from the National Bureau of Economic Research African Successes Project, the GoBifo Project, the Harry F. Guggenheim Foundation, the Institutional Reform and Capacity Building Project (IRCBP), the International Growth Centre (IGC), the International Initiative for Impact Evaluation (3ie), the MIT Horace W. Goldsmith Foundation, and the World Bank Development Impact Evaluation (DIME) Initiative. For acknowledgments, sources of research support, and disclosure of the authors’ material financial relationships, if any, please see http://www.nber.org/chapters/c13387.ack.

15

16

Katherine Casey, Rachel Glennerster, and Edward Miguel

corruption, as well as confront and amend long-standing social tensions and inequities that many believed helped fuel the recent violence. This chapter examines how different factors—including the legacy of war, ethnic diversity, decentralization, and community-driven development (CDD)—affect local institutions and collective action, as well as national political culture and outcomes in Sierra Leone. The story that emerges is nuanced and does not confirm reflexive biases: war does not necessarily destroy the capacity for local collective action; ethnicity affects residential choice, but does not impede local public goods provision; while politics remain heavily ethnic, voters are more willing to cross ethnic boundaries in local elections where they have better information about candidates; decentralization can work even where local capacity is highly constrained, although the results are mixed; and for all of its promise and some positive impacts on local public goods provision, CDD does not transform local institutions nor social norms of behavior. All of these results are somewhat “unexpected,” but they are quite positive in signaling that even one of the world’s poorest, most violent and ethnically diverse societies can overcome major challenges and progress towards meaningful economic and political development.1 The rest of the chapter is structured as follows: section 1.2 provides background on Sierra Leone’s protracted decline into poverty and unrest, exploring prominent social divisions that may have encouraged young men to take up arms; section 1.3 discusses the historical evolution of ethnic diversity and the complex role it plays in contemporary public life; section 1.4 details key postwar institutional reforms including the restoration of multiparty democracy, decentralization, and CDD; section 1.5 assesses the progress these reforms have made toward encouraging economic development, democratizing institutions, and changing social norms; and section 1.6 concludes. 1.2

Legacies of Poverty, Corruption, and Conflict

After achieving independence from Britain in 1961, Sierra Leone enjoyed only a brief period of free and competitive democracy. Increasing political instability, worsening governance, and deepening poverty marred the subsequent few decades, which terminated in institutional collapse and civil war. In the 1970s and 1980s the country was ruled by authoritarian leaders who enriched themselves through illicit deals involving diamonds, while doing little to provide needed services such as health care and education (Reno 1995). Eliminating threats to its absolute control, the government of 1. This chapter summarizes and brings together findings from four others: Bellows and Miguel (2009), Glennerster, Miguel, and Rothenberg (2013), Casey, Glennerster, and Miguel (2012), and Casey (2015). For complete details on theoretical models, identification strategies, empirical specifications and data sets, please see these original papers.

Healing the Wounds: Learning from Sierra Leone’s Postwar Reforms

17

President Siaka Stevens dismantled competitive democracy by abolishing district-level local government in 1972 and declaring the country a one-party state in 1978. By the early 1990s, Sierra Leone had the second lowest living standards of any country in the world (United Nations 1993). Figure 1.1 situates the economic experience of Sierra Leone in relation to the rest of sub-Saharan Africa. The comparative poverty of Sierra Leone is evidenced by real gross domestic product (GDP) per capita that falls significantly below the regional average for the entire period (1960–2010). Note further the economy’s stagnant performance from 1970–1990, followed by precipitous decline during the war. This chapter focuses on the institutional factors surrounding the positive postwar recovery apparent in the upward trend of the last decade. While the particular confluence of events and reforms is unique to Sierra Leone, note that its strong performance in the last several years mimics broader trends of growth for the region as a whole (Miguel 2009). The weak economic performance and poor governance of the 1970s and 1980s steered the country toward civil unrest. Partially as a result of the widespread discontent toward the corruption and ineffectiveness of the

Fig. 1.1 Trends in GDP per capita 1960–2010 (in constant 2000 US dollars) for sub-Saharan Africa and Sierra Leone Source: Source of data is World Development Indicators, World Bank. Note: The sub-Saharan African mean is the “all income levels” series.

18

Katherine Casey, Rachel Glennerster, and Edward Miguel

government, a small group of rebels, who had entered the country from Liberia in 1991, were successful in recruiting disenfranchised youth to rise up violently against the status quo. As their numbers swelled by early 1992, these rebels, known as the Revolutionary United Front (RUF), spread the armed conflict to all parts of the country. The brutal civil war that ensued saw an estimated 50,000 Sierra Leoneans killed, over half of the population displaced from their homes, and thousands of civilians victimized by amputation, rape, and assault (Human Rights Watch 1999). A small cadre of British troops, along with a large international peacekeeping mission, brought the war to a decisive end with peace officially declared in January 2002. Scholars point to a number of long-standing social divisions that created frustration and may have helped incite violence. First, some have claimed that the initial motivations of the RUF were idealistic and that the early rebels were guided by a strong sense of political grievances related to the failings of the corrupt regime (Richards 1996). Such frustrations were particularly acute for young men, who were largely excluded from decision making and at times subject to coerced labor and capricious fines by traditional authorities. Second, colonial rule enhanced the historical legacy of inequality between local chiefs and their subjects. During the colonial period, the British implemented direct rule over residents of the Western peninsula (the Colony), yet exerted indirect rule over residents of the interior (the Protectorate). This latter system promoted chiefs loyal to the British, and institutionalized—and in many cases augmented—their autocratic power over their subjects, thereby exacerbating inequality and reinforcing social divisions. Third, although not likely a direct cause of violence, women have historically held less power in local governance and possessed weaker socioeconomic status as compared to men. After the war ended, major institutional reforms aimed to address these root causes of dissension, to both promote greater equity and preclude a return to violence. Although devastating, the war did not leave the country so weakened as to be incapable of recovery. While violence inflicted during the war created incalculable human suffering and destroyed much of the physical infrastructure of the country, the impact on institutions is more nuanced. Bellows and Miguel (2009) unexpectedly uncover a positive association between exposure to violence and subsequent increases in political and social activism. Their research suggests that individuals whose households directly experienced war violence are more active political and civic participants than nonvictims: they are more likely to vote (by 2.6 percentage points), attend community meetings (by 6.5), belong to a social (by 6.6) or political (by 5.7) group, and serve on school management committees (by 3.8). In addition, these victims were no worse off in terms of standard consumption measures a few years after the war ended. While these findings underscore the extraordinary resilience of Sierra Leoneans, it is important to note that they are based on variation across individuals within the same village, and

Healing the Wounds: Learning from Sierra Leone’s Postwar Reforms

19

thereby do not estimate the net effect of civil war on the country as a whole. The authors conclude that while the “humanitarian costs of civil wars are horrific . . . it appears their legacies need not be catastrophic.” 1.3

Ethnic Diversity

Ethnicity plays a nuanced role in the social and political life of Sierra Leone, defying commonly held conceptions about the adverse effects of diversity. Many scholars have argued that ethnic diversity is an important impediment to economic and political development. Economic growth rates are slower in ethnically diverse societies, and local public goods provision often suffers (Easterly and Levine 1997; Alesina, Baqir, and Easterly 1999; Alesina et al. 2003; Fearon 2003). The inability to overcome the public good free-rider problem in diverse communities, due to monitoring and enforcement limitations, is the leading explanation proposed for less developed countries (Miguel and Gugerty 2005; Habyarimana et al. 2007, 2009). These issues are particularly salient in sub-Saharan Africa, the world’s most ethnolinguistically diverse region. Yet our research shows that while ethnicity is important for residential choice and political allegiances in Sierra Leone, it does not appear to hamper local collective action and was not an organizing factor in the civil war. 1.3.1

Historical View of Ethnicity

As background, Sierra Leone is very diverse, ranking fifteenth on the Taylor and Hudson (1972) list of countries with the highest levels of ethnolinguistic fractionalization.2 Specifically, of eighteen major ethnic groups, the Mende and Temne are numerically dominant, occupying shares of 32.2 percent and 31.8 percent, respectively, while the Limba, Kono, and Kuranko are the next largest groups at 8.3 percent, 4.4 percent, and 4.1 percent, respectively (National Population and Housing Census 2004). Other groups occupy a substantially smaller share, including the Krio (Creoles)—former slaves who returned to Africa to settle Freetown—whose population share fell to only 1.4 percent by 2004. These groups are characterized by distinct customs, rituals, and history, and, most importantly, language. With the exception of Krio, an English dialect, the other languages are members of the Niger-Congo language family. Within this family, the most salient distinction is between the Mande languages—including Mende, Kono, Kuranko, Susu, Loko, Madingo, Yalunka, and Vai—and the AtlanticCongo languages, including Temne, Limba, Sherbro, Fullah, Kissi, and Krim. These groups are mutually unintelligible to each other, and much further apart linguistically, for example, than English and German. Over the past two centuries national politics have been heavily influenced 2. As cited in Easterly and Levine (1997).

20

Katherine Casey, Rachel Glennerster, and Edward Miguel

by two distinct divisions along ethnic lines, where the early Colony versus Protectorate tension under the British was later surpassed by regional allegiances after independence. At the time of the founding of the Sierra Leone colony in the late eighteenth century and through much of the nineteenth century, the Krio enjoyed a relatively privileged political and economic position due to their facility with English and special links with the British even though they were numerically small. Before independence, the key political division in Sierra Leone was Krio versus non-Krio, but because of growing tensions between the Krio and “up country” ethnic groups, the British progressively limited their political power. Thus as Sierra Leone made its transition to independence, the primary source of political conflict shifted. As stated by Kandeh (1992, 90), “the salience of the Creole [Krio]-protectorate cleavage was eclipsed after independence by the rivalry between the Mendes of the south and Temnes of the north.” It is this largely regional divide that continues to galvanize national politics today. Two facts about the Krio may have prevented ethnic political divisions from escalating into violent conflict. One key difference between Sierra Leone and many other African countries is that the “favored” ethnic group during early colonialism was not truly indigenous and no longer holds a position of power. They historically served as a common antagonist for the Mendes and Temnes together, and have since lost their political influence. Second, the Krio people gave Sierra Leone their language, also called Krio, which is a dialect of English that has been influenced by Portuguese, Arabic, Yoruba, and many African languages as a legacy of the slave trade. Serving as a national lingua franca for decades, Krio is currently spoken (usually as a second language) by most Sierra Leoneans, and is increasingly taught in schools. In many other African countries the lingua franca is the former colonial language, usually English or French. While Krio has a base in English, it is unique to Sierra Leone and widely spoken even by those with no schooling. While the existence of a common national language is clearly insufficient to guarantee social stability—as the African cases of Rwanda and Somalia poignantly illustrate—Krio’s ubiquity in Sierra Leone may (through historical accident) help promote the consolidation of a common national identity that transcends tribe (Ngugi 2009), as with Swahili in postindependence Tanzania (Miguel 2004). Perhaps due to these unifying forces and in contrast to most popular media coverage on African civil wars, neither ethnic nor religious divisions played a central role in the Sierra Leone conflict. The Revolutionary United Front (RUF) rebels targeted people from all ethnic groups, and statistical analysis of documented human rights violations shows that no ethnic group was disproportionately victimized. There is also no evidence that civilian abuse was worse when armed factions and communities belonged to different ethnic groups (Humphries and Weinstein 2006). Ethnic grievances were not rallying cries during the war and all major fighting sides were ex-

Healing the Wounds: Learning from Sierra Leone’s Postwar Reforms

21

plicitly multiethnic (Keen 2005). The fact that the war was not fought along ethnic lines, and the central role that external actors played in bringing it to a conclusive end, may partially explain why there has not been a resurgence of violence. 1.3.2

How Ethnicity Matters Today

The fact that ethnic identity was not an organizing factor in the conflict does not, however, mean that it is unimportant. Glennerster, Miguel, and Rothenberg (2013) show that a preference for one’s own ethnic group is a key determinant of residential choice for rural Sierra Leoneans. Using nationally representative household data that collected respondents’ chiefdom of residence in 1990 and 2007, they estimate a discrete choice model to understand why different individuals moved across chiefdoms after the war. Since contemporary ethnic composition is in part endogenously determined by postwar migration choices, their empirical specifications use historical 1963 ethnic shares. They find that individuals are on average willing to travel an additional 10.1 kilometers to live in a chiefdom with a 10 percentage point greater share of her/his own ethnic group. While still strong, the coethnic preference is attenuated for people with some education (note that adult literacy is just 34.8 percent). In particular, educated individuals are only willing to travel an additional 8.6 kilometers to live in a chiefdom with a 10 percentage point greater share of her/his own ethnic group, which suggests that education dampens coethnic residential preferences. Yet the preference for living with one’s own ethnic group does not translate into a weaker ability to work together with those from other groups. In fact, conditional on other factors (including remoteness from cities as well as population size and density), individuals exhibit a positive preference for diversity, though this is smaller than the preference for a higher coethnic share. Standing in sharp contrast to the bulk of the ethnic diversity literature, Glennerster, Miguel, and Rothenberg extend their analysis and find no negative effects of ethnic diversity on the provision of local public goods. The authors use a mean effects approach to summarize the average impacts of the project across a family of related indicators (following Kling, Liebman, and Katz 2007). Specifically, they find no effect of ethnic—or religious— diversity on local collective action (as measured by road maintenance, group membership, self-expressed trust or disputes); no effect of ethnic diversity on the quality of primary schools (as measured by instructional supplies, facilities, or teaching); and if anything, positive impacts of ethnic diversity on health clinic quality, supplies, and staff presence and quality. Arguably the most salient arena of ethnic loyalties today lies in politics. Casey (2015) notes the tight correlation between voting choices and ethnic identity. The two major political parties—the Sierra Leone People’s Party (SLPP) and the All People’s Congress (APC)—have strong, long-standing ties to the Mende and other ethnic groups in the south and the Temne and

22

Katherine Casey, Rachel Glennerster, and Edward Miguel

other groups in the north, respectively. As an example of the strength of these loyalties, in the 2007 Parliamentary elections the APC won thirtysix of thirty-nine seats in the Northern Province, while the SLPP and its splinter party, the People’s Movement for Democratic Change (PMDC), swept twenty-four of twenty-five seats in the Southern. At the individual level, voters are readily forthcoming about ethnic-party allegiances: laughter is not an uncommon response to questions about why someone voted for a particular political party, often followed by an explanation of how their father and grandfather voted for the same one. When asked which party they supported in the first round of the 2007 presidential elections, 88.4 percent of Mende voters reported choosing the SLPP or PMDC, while 94.2 percent of Temne voters reported backing the APC.3 Yet Casey argues that these traditional ethnic alliances are not immutable, particularly in local elections where voters have access to better information about candidates. Detailed exit poll survey data collected in 2008 shows that while party is equally as important as a bundle of individual candidate characteristics (like reputation in his/her previous job, education, and kinship ties) in national elections, voters say that party is only half as important as these traits in local races. 1.4

Institutional Reforms

Sierra Leone emerged from the war in a very weak position economically, socially, and politically. The country remained at the bottom of the UN Human Development Index ranking, families were mourning the loss of loved ones and grappling with the emotional and physical traumas of war, the network of infrastructure and public services was largely destroyed, and the government faced an institutional vacuum. At the same time, the people were ready for change and committed to preventing a return to violence. The government of Sierra Leone and its donor partners responded with an ambitious program of institutional reform and economic development. The reforms aimed to foster economic growth while strengthening the institutional environment to promote better governance and lasting peace. The most high profile political reforms were the restoration of multiparty democracy and the reconstitution of local government after over thirty years of dormancy. The National Electoral Commission (NEC) oversaw competitive elections for Parliament and the presidency in 2007 and for the nineteen local councils in 2004 and 2008. These reforms brought democratic competition back to national, district, and ward-level institutions. The peaceful transfer of power from the ruling SLPP regime to the opposition APC 3. Source: IRCBP, National Public Services Survey 2008. This is a nationally representative household survey of over 6,000 households. Survey responses are limited to those respondents who stated their voting choice and could verify the fact that they had voted by producing their voter identification card with a hole punch made by polling center staff indicating participation in the 2007 first round voting.

Healing the Wounds: Learning from Sierra Leone’s Postwar Reforms

23

challengers in 2007 stands out as a particularly impressive achievement. When compared to its counterpart for the Kenyan elections of the same year, the NEC distinguished itself by taking decisive action to address charges of electoral irregularities from both sides—the challengers and the ruling party. A report by Freedom House (2009), the non-governmental organization (NGO) watchdog, notes how the NEC “functioned with remarkable independence and helped to ensure the success of the balloting, despite postponements and other difficulties.” Understanding the factors that enable electoral commissions to succeed—and what role external donors might play in helping to protect their independence (the UN among others provided substantial support in Sierra Leone)—is important. In addition to historical antecedent, the main theoretical motivations for decentralization seemed a good match for the postwar context, both in terms of providing local public goods and empowering the citizenry. On the public goods front, theorists point to economic welfare gains from decentralizing service provision to local governments that have better information about, and greater ability to tailor outputs to, differences in local preferences and costs (Oates 1999). By reducing the (geographic and bureaucratic) distance between frontline service providers and managers, decentralization can further reduce the cost of supervision and increase the speed and efficiency with which managers respond to needs on the ground. The informational and supervisory advantages of local government are particularly important in light of weak transportation and communication networks that exacerbate central government oversight challenges. Yet it is not clear that decentralized provision necessarily dominates centralized, as there are risks that the accountability gains from better information may be compromised by a greater risk of elite capture (particularly when only a small minority is literate and politically aware) or that services management will suffer in the hands of less competent local politicians (Bardhan and Mookherjee 2006). Aware of the potential benefits and risks, the government of Sierra Leone began to gradually transfer responsibilities and tied grants for public services in health, agriculture, education, and other sectors over to the local councils in accordance with the Local Government Act of 2004. The extent of decentralization to date varies widely across sectors, where health takes the lead and education lags conspicuously behind. Yet even in health, decentralization remains at best partial since the power to hire, fire, and remunerate staff is retained by the central government. To complement the conditional grants by sector, the councils were given several million dollars’ worth of discretionary funds to use toward development projects in their districts under the Local Government Development Grants (LGDG) program. Turning to civic engagement, scholars of decentralization emphasize the political value of creating greater opportunities for citizens to participate in government, which can be particularly important in developing countries that inherited highly centralized regimes from colonial powers (Oates 1999).

24

Katherine Casey, Rachel Glennerster, and Edward Miguel

They suggest that inclusion and participation carry both intrinsic value in empowering citizens as well as accountability gains in enabling the public to better monitor and constrain the behavior of elected officials. In Sierra Leone, encouraging the participation of women and youth takes on added importance in light of their long-standing exclusion and the role that resulting frustrations may have played in the war. The government and donors sought to complement these national reforms with “bottom-up” initiatives aimed at strengthening community-level institutions. Community-driven development (CDD) is one such approach that has become very popular throughout the developing world: the World Bank alone has spent US $50 billion on CDD initiatives over the past decade (Mansuri and Rao 2012). By emphasizing local participation in and control over project implementation, CDD aims to provide public goods through a process that empowers the poor. Advocates of participatory local governance promise a long and varied list of benefits ranging from more cost-effective construction of infrastructure, to a closer match between project choice and village needs, to the weakening of authoritarian village institutions.4 Critics hold concomitant concerns that participation requirements serve as a regressive tax, widening political participation clogs up rather than expedites decision making (Olson 1982), and external resources attract new leaders, crowd out the most disadvantaged (Gugerty and Kremer 2008), or are captured by elites if the program is unable to change the nature of de facto political power (Bardhan 2002). Any real world program risks manipulation during implementation, and skeptical observers fear that donors simply use the jargon of participatory development for political or public relations purposes while continuing to operate in a “top-down” manner. Few studies provide rigorous empirical evidence regarding these claims (Mansuri and Rao 2004). Using a randomized experiment, Casey, Glennerster, and Miguel (2012) evaluate the success of one particular CDD project in Sierra Leone, the “GoBifo” project (which means “move forward” in Krio). As is typical of CDD projects, GoBifo established village-level structures and provided tools to plan and manage development resources; provided communities with financing and technical assistance in implementing small-scale projects (totaling $4,667 per community or roughly $100 per household); and created links between these processes and local government institutions. This “hardware” support was coupled with intensive community facilitation, or “software,” that promotes democratic decision making, the participation of socially marginalized groups, and transparent budgeting practices. The CDD approach attempts to bolster local coordination—for example, by setting up village development committees—and to enhance participation 4. For instance, Dongier et al. (2003, 303) write that: “Experience demonstrates that by directly relying on poor people to drive development activities, CDD has the potential to make poverty reduction efforts more responsive to demands, more inclusive, more sustainable, and more cost effective than traditional centrally led programs . . . achieving immediate and lasting results at the grassroots level.”

Healing the Wounds: Learning from Sierra Leone’s Postwar Reforms

25

by requiring women and youth (adults under age thirty-five) to hold leadership positions, sign off on project finances, and attend meetings. Project emphasis on inclusive and democratic decision-making aims to empower women and youth in other realms of local governance, and the learningby-doing experience with successfully implementing communal projects aims to catalyze collective action beyond the immediate project sphere. The idea is that once communities have the institutions in place—a village development committee, plan, bank account, and experience in budgeting and management—they should be better able to take advantage of new opportunities that arise after the program itself has ended. This latter emphasis on “help yourself ” activism echoes current President Ernest Bai Koroma’s “Attitudinal Change” public messaging campaign that urges citizens to take responsibility for their own development. 1.5

Effectiveness of Institutional Reforms

The following sections examine how successful these initiatives—multiparty democracy, decentralization, and community-driven development— have been in instigating economic development and democratizing institutions, with an eye on changes in social and political norms of behavior. After a brief look at the postwar macroeconomic situation, section 1.5.1 explores three areas of public goods provision: the impact of decentralization on access to services and supervision of field staff, how electoral pressures and ethnic-party allegiances influence the allocation of public spending across constituencies, and how effective CDD has been in delivering small-scale public goods to communities. Section 1.5.2 then turns to participation, discussing the impact of the postwar reforms on information flows between citizens and their elected officials, as well as on the direct participation of individuals in community-level decisions. 1.5.1

Progress toward Economic Development

Postwar recovery on the macroeconomic level has been relatively rapid and robust. The ministry of finance estimates that the economy has been growing at roughly 5 percent per year since the end of the war. Efforts to rebuild public infrastructure have largely restored the prewar (although still very low) stock of key public goods and flow of basic services. While this progress is encouraging, the country remains extremely poor: it again fell to the bottom of the latest Human Development Index, ranking 180th out of the 182 countries included (United Nations 2009). While we have no rigorous test of whether local or centralized service provision performs better, evidence suggests that decentralization has been consistent with improvements in public services on the ground. Foster and Glennerster (2009) note that access to public services has improved over the period 2005 to 2007: the percentage of households reporting access to a primary school within half an hour increased by 5.5 percentage points,

26

Katherine Casey, Rachel Glennerster, and Edward Miguel

to a health clinic by 4.4 ppts., to a motorable road by 7.3 ppts., to a market by 13.0 ppts. (within one hour), and to drinking water by 12.4 ppts. (within fifteen minutes). The only significant reduction was in access to an agricultural extension officer, which fell by 5.5 percentage points. For the sector that devolved the most—health—a panel survey of clinics reveals that nearly all indicators of health care quality (including clinic staffing, supplies, and equipment) improved significantly between 2005 and 2008. They further provide suggestive evidence that the gains from decentralization were largest where the reductions in the distance to power (i.e., from the national capitol to the relevant district capitol) were greatest. This implies that service improvements were most pronounced in areas located close to a new district government headquarters, but far from Freetown. As a robustness check, they find no such differential change for education outcomes, which experienced no decentralization during this time period. Despite the theoretical gains in supervision, evidence suggests that the newly elected local councilors were not particularly active in overseeing services in the field. Focusing on health care, while councilors managed to visit roughly half of all clinics during their first year of office, these visits fell by 50 percent over the subsequent two years. Furthermore, interviews with clinic staff suggest that the real gains in oversight and technical assistance occurred within the ministry itself, with authority devolving from central government bureaucrats to district-level members of the health management teams. Taken together, these findings suggest that the substantial improvements in health care may be due more to deconcentration within the ministry than the oversight of local politicians driven by electoral concerns. A key sector for economic development, agriculture, remains one of the areas where the decentralization process is yet to establish a clear division of roles between local councils and the central government (Casey 2009). Market construction represents one of the major items of spending of local councils’ discretionary funding. Yet, the lack of a central agency devoted to market development5 and the absence of specific transfers to local councils in this area are among the factors limiting access to markets for agricultural producers in the country (Srivastava and Larizza 2011). One of the authors is currently leading a research team exploring potential innovative solutions to improve farmers’ access to product market, credit market, and agricultural extension. Results from the 2010 Agricultural Household Tracking Survey, the first postwar nationally representative agricultural household survey, informed some of these pilots. In one of the projects the team is partnering with the Ministry of Agriculture Forestry and Food Security (MAFFS) to test the impact of an inventory credit scheme. By providing loans to palm oil producers, who can use stored product as collateral, the pilot aims to satisfy immediate liquidity shortages that farmers may face at 5. The Sierra Leone Produce Marketing Board was dismantled in 1989.

Healing the Wounds: Learning from Sierra Leone’s Postwar Reforms

27

harvest time. Such cash needs are reported to be one of the main reasons most sales occur in the peak harvest period, a time when farm-gate prices are typically low. Evaluation of the pilot will assess the extent to which improved access to credit may change storage decisions and the timing of product sales. Another project combines agricultural extension with temporary subsidies to promote the adoption of improved varieties among rice producers. One area where electoral politics, service provision, and ethnicity come together is in the allocation of public resources across constituencies. While we have seen that ethnic diversity does not affect public goods provision within communities, it does appear to influence how much political patronage external agents bestow upon their constituents. Since ethnic diversity signals greater political competitiveness, Casey (2015) argues that it attracts greater investment from candidates on the campaign trail and from politicians once in office. Given the long-standing ties between particular ethnic groups and political parties, she estimates the competitiveness of a given race for a seat in Parliament or one of the local councils based on the ethnic composition of its constituency. To illustrate: the SLPP and APC have an equal chance of winning a maximally competitive constituency that has 50 percent Mende and 50 percent Temne residents, while the SLPP has little chance of winning a constituency that is 90 percent Temne. This logic provides a novel empirical test of the classic swing voter proposition that both parties favor constituencies with the weakest underlying party preference (Lindbeck and Weibull 1987). A test based on ethnic composition—which is largely stable over time and not subject to short-term shifts in response to political patronage—does not suffer the endogenity bias that plagues previous tests based on reported voting choices.6 Under this framework, Casey finds evidence that moving from a jurisdiction that is perfectly homogenous to one that is maximally competitive results in a 0.89 standard deviation unit increase in the bundle of campaign goods distributed by national candidates and $19,575 more public spending by elected local politicians. In a two-group setting, these findings are equivalent to saying that political competition and thus patronage are increasing in ethnic diversity, which could provide an alternative explanation for the counterintuitive findings regarding clinic quality discussed in section 1.3.2.7 Moving down to the community level, decentralizing service provision to villages through CDD proved effective in delivering smaller-scale public 6. Voting choices in part reflect transfers from political parties, which arise endogenously from the strategic game played by parties seeking to win elections. See Larcinese, Snyder, and Testa (2013) for estimates of the resulting bias. 7. Note that of all the measures considered by Glennerster, Miguel, and Rothenberg (2013), clinic quality is the least influenced by local collective action, where major policy decisions— including clinic location choice and procurement and hiring/firing of medical staff—are all controlled more centrally. Health clinic quality may thus provide another example of how winning politicians reward more diverse and politically competitive jurisdictions with better public goods once in office.

28

Katherine Casey, Rachel Glennerster, and Edward Miguel

goods. Casey, Glennerster, and Miguel (2012) find that strong implementation performance by the GoBifo project improved the stock and quality of local public goods as well as enhanced general economic welfare in treatment communities. In particular, the project successfully established village development committees, plans and community bank accounts, and facilitated greater interaction between villages and elected local politicians and chiefdom officials. Community verification of project financial receipts further attests to minimal leakage of project resources, which is no small feat in an environment of endemic corruption. It also appears that communities used these grants productively: again using a mean effects approach, GoBifo led to a 0.204 standard deviation unit increase in the average stock and quality of local public goods (including the presence of a functional traditional midwife post, latrine, and community center, as well as better quality construction of primary schools and grain-drying floors, among other outcomes); and GoBifo enhanced general economic welfare by 0.376 standard deviation units (including the presence of petty traders, number of goods on sale, household assets and amenities, among others). These results suggest that CDD is a reasonable approach to delivering small-scale local public goods in a way that is equitable, accountable, and low cost. 1.5.2

Progress toward Democratizing Institutions

Our research suggests that the postwar institutional reforms have increased information flows between citizens and local politicians, with positive impacts on voting behavior and oversight of local public goods. Yet we find little evidence to suggest that community-level interventions have led to fundamental changes in social norms, particularly with respect to the voice and participation of women and youth. Decentralization has lived up to its purported informational advantages in terms of increasing how much voters know about their politicians, which in turn reduces the salience of ethnic identity in voting choices and campaign spending. As decentralization brings government closer to the people, Casey notes that voters have more information about local politicians (for example, while 37 percent of respondents could correctly name their local councilor, only 17 percent could name their parliamentarian) and more opportunities to interact with them (while 50 percent of communities reported being visited by their elected councilor in the past year, only 25 percent report a visit from their MP). Adding information provision into a standard swing voter investment model, Casey demonstrates that increasing the amount of information voters have about candidates (beyond party affiliation), makes voters more likely to cross traditional ethnic-party lines for particularly attractive rival party candidates. Exploiting the information differences across levels of government and using individual fixed effects, she shows that the same voters are 11.3 percentage points more likely to cross traditional ethnic grouppolitical party allegiances when voting in local as opposed to national elections. Knowing this, parties in turn respond by de-emphasizing ethnic-party

Healing the Wounds: Learning from Sierra Leone’s Postwar Reforms

29

allegiances in allocating campaign spending across constituencies. While parties continue to favor more diverse constituencies for both local and national races, the spending in local races is roughly half as responsive to ethnic composition as that for national elections. An optimistic interpretation of these results suggests that providing voters with better information about candidates could help reduce the salience of ethnicity in politics. While CDD strengthened village-level public good provision and created meaningful links between villagers and the lowest tiers of elected government, it did not fundamentally influence social norms nor communal capacity for collective action. Yet, taking a step back, let us first question the assumption that the war left the communities highly compromised in terms of social cohesion and their ability to work together. Note that baseline levels of cohesion were high: over 81 percent of respondents had trusted a neighbor with goods to sell on their behalf in the local market, the average person was a member of more than two of five common social groups, and only 22 percent reported having a financial conflict in the preceding year. Glennester, Miguel, and Rothenberg (2013) further show that there is no evidence to negatively link ethnic diversity with such cohesion and the resulting ability to act collectively. In addition, Bellows and Miguel emphasize that if anything, survivors of violence emerged from the war even more politically and socially active than they were before. Such initially high levels of cohesion throw into doubt the necessity of the social facilitation aspect of CDD in this context. That said, there were clear social divisions—between women and men, and youth and their elders—that were creating frustration, and chronic poverty suggests that greater collective action toward local development would likely be welfare enhancing. Yet while CDD explicitly targeted these areas for remedy, we find no evidence that it had any lingering impact on the voice and participation of women and youth, nor on the likelihood that communities were able to take up development opportunities arising after the project ended. While these null results held true across a wide variety of outcomes, the following two examples from our “structured community activities” (SCAs) are illustrative. The SCAs were designed to unobtrusively observe members of treatment and control communities engaged in concrete, realworld collective activities and decisions. Regarding voice, the research team found no difference in the number of women who spoke publicly in a community meeting to decide between two small gifts offered by the research team as a thank you for their participation in the study. Regarding collective action, exactly sixty-two treatment and sixty-four control villages (roughly 50 percent across the board) took up a voucher opportunity to purchase building materials at a subsidized price for use in a community project. These findings (along with the other indicators studied) suggest quite conclusively that CDD had no impact on underlying attitudes toward women and youth, and did not serve as a catalyst for collective action beyond the life of the project.

30

1.6

Katherine Casey, Rachel Glennerster, and Edward Miguel

Concluding Remarks

Many would have said that Sierra Leone was condemned to be a basket case because of its war history, ethnic diversity, and poverty, but the economic and political progress of the last decade belies this. In fact, war legacies and ethnic diversity do not appear to necessarily hinder local collective action. And while national politics remain heavily ethnic, power is transferred peacefully between parties, and voters are willing to cross traditional ethnic allegiances when they have better information. Turning to public goods, decentralization has been compatible with steady improvements in service delivery, greater interaction between citizens and their elected representatives, and enhanced supervision of front line workers by district-level managers. Ten years ago, few observers would have thought this last decade of peace and prosperity was possible for Sierra Leone. In terms of policy lessons, it is useful to consider the role that external actors played in these achievements. Taken together, our research suggests that external assistance and interventions to support large-scale institutional reforms have met with greater success than those targeting community-level norms and dynamics. In particular, international donors and foreign governments have contributed to ending the civil war, restoring multiparty democracy, and decentralizing public services. Implemented in partnership with the government of Sierra Leone, these reforms have translated into substantial, and quite tangible, benefits on the ground. The experience with CDD, however, suggests that fundamentally altering community-level hierarchies and social norms is incredibly difficult, and not something that we, as outsiders, yet know how to do effectively. Moving forward, much uncertainty remains regarding the potential for sustained economic growth and institutional development in Sierra Leone. The discovery of large deposits of natural resources (including iron ore and oil) provides hope for improved incomes, but may also place additional pressure on weak institutions. As Sierra Leone celebrated fifty years of independence in 2011, the country faced an open question of whether these newly found resources would contribute more to general development than diamonds have to date. As with the successes we have seen in the last several years, the country’s further progress ultimately rests on the resilience and determination of Sierra Leoneans striving for a better future.

References Acemoglu, Daron, Simon Johnson, and James A. Robinson. 2001. “The Colonial Origins of Comparative Development: An Empirical Investigation.” American Economic Review 91 (5): 1369–1401.

Healing the Wounds: Learning from Sierra Leone’s Postwar Reforms

31

Alesina, Alberto, Reza Baqir, and William Easterly. 1999. “Public Goods and Ethnic Divisions.” Quarterly Journal of Economics 114 (4): 1243–84. Alesina, Alberto, Arnaud Devleeschauwer, William Easterly, Sergio Kurlat, and Romain Wacziarg. 2003. “Fractionalization.” Journal of Economic Growth 8 (2): 155–94. Banerjee, Abhijit, and Lakshmi Iyer. 2005. “History, Institutions, and Economic Performance: The Legacy of Colonial Land Tenure Systems in India.” American Economic Review 95 (4): 1190–213. Bardhan, Pranab. 2002. “Decentralization of Governance and Development.” Journal of Economic Perspectives 16 (4): 185–205. Bardhan, Pranab, and Dilip Mookherjee. 2006. “Decentralisation and Accountability in Infrastructure Delivery in Developing Countries.” Economic Journal 116 (508): 101–27. Bellows, John, and Edward Miguel. 2009. “War and Local Collective Action in Sierra Leone.” Journal of Public Economics 93 (11–12): 1144–57. Casey, Katherine. 2009. “Decentralization in Practice.” In Decentralization, Democracy, and Development: Recent Experience from Sierra Leone, edited by Yongmei Zhou. Washington, DC: World Bank. ———. 2015. “Crossing Party Lines: The Effects of Information on Redistributive Politics.” American Economic Review 105 (8): 2410–48. Casey, Katherine, Rachel Glennerster, and Edward Miguel. 2012. “Reshaping Institutions: Evidence on Aid Impacts Using a Pre-Analysis Plan.” Quarterly Journal of Economics 127:1755–812. Dongier, Philippe, Julie Van Domelen, Elinor Ostrom, Andrea Rizvi, Wendy Wakeman, Anthony Bebbington, Sabina Alkire, Talib Esmail, and Margaret Polski. 2003. “Chapter 9: Community-Driven Development.” In The Poverty Reduction Strategy Sourcebook, vol. 1, edited by Jeni Klugman, 301–31. Washington, DC: World Bank. Easterly, William, and Ross Levine. 1997. “Africa’s Growth Tragedy: Policies and Ethnic Divisions.” Quarterly Journal of Economics 112:1203–50. Engerman, Stanley L., and Kenneth L. Sokoloff. 1997. “Factor Endowments, Institutions, and Differential Paths of Growth among New World Economies: A View from Economic Historians of the United States.” In How Latin America Fell Behind, edited by Stephen Haber. Stanford: Stanford University Press. Fearon, James D. 2003. “Ethnic and Cultural Diversity by Country.” Journal of Economic Growth 8 (2): 195–222. Foster, Elizabeth, and Rachel Glennerster. 2009. “Impact of Decentralization on Public Services: Evidence to Date.” In Decentralization, Democracy, and Development: Recent Experience from Sierra Leone, edited by Yongmei Zhou. Washington, DC: World Bank. Freedom House, Inc. 2009. “Country Report: Sierra Leone.” Accessed July 8, 2010. http:// www.freedomhouse.org/ template.cfm?page=22&year=2009&country =7699. Glennerster, Rachel, Edward Miguel, and Alexander Rothenberg. 2013. “Collective Action in Diverse Sierra Leone Communities.” Economic Journal 123 (568): 285– 316. Gugerty, Mary Kay, and Michael Kremer. 2008. “Outside Funding and the Dynamics of Participation in Community Associations.” American Journal of Political Science 52 (3): 585–602. Habyarimana, James P., Macartan Humphreys, Daniel Posner, and Jeremy Weinstein. 2007. “Why Does Ethnic Diversity Undermine Public Goods Provision?” American Political Science Review 101 (4): 709–25.

32

Katherine Casey, Rachel Glennerster, and Edward Miguel

———. 2009. Coethnicity: Diversity and the Dilemmas of Collective Action. New York: Russell Sage. Human Rights Watch. 1999. Sierra Leone: Getting Away with Murder, Mutilation, and Rape. New York: Human Rights Watch. Humphries, Macartan, and Jeremy M. Weinstein. 2006. “Handling and Manhandling Civilians in Civil War: Determinants of the Strategies of Warring Factions.” American Political Science Review 100 (3): 429–47. Kandeh, Jimmy D. 1992. “Politicization of Ethnic Identities in Sierra Leone.” African Studies Review 35:81–99. Keen, David. 2005. Conflict and Collusion in Sierra Leone. London: James Currey. Kling, Jeffrey R., Jeffrey Liebman, and Lawrence Katz. 2007. “Experimental Analysis of Neighborhood Effects.” Econometrica 75 (1): 83–119. Larcinese, Valentino, James M. Snyder, and Cecilia Testa. 2013. “Testing Models of Distributive Politics Using Exit Polls to Measure Voter Preferences and Partisanship.” British Journal of Political Science 43 (4): 847-75. Lindbeck, Assar, and Jorgen W. Weibull. 1987. “Balanced-Budget Redistribution as the Outcome of Political Competition.” Public Choice 52:273–97. Mansuri, Ghazala, and Vijaynedra Rao. 2004. “Community-Based and -Driven Development: A Critical Review.” World Bank Research Observer 19 (1): 1–39. ———. 2012. Localizing Development: Does Participation Work? Washington, DC: World Bank. Miguel, Edward. 2004. “Tribe or Nation? Nation Building and Public Goods in Kenya versus Tanzania.” World Politics 56 (3): 327–62. ———. 2009. Africa’s Turn? Cambridge, MA: MIT Press. Miguel, Edward, and Mary Kay Gugerty. 2005. “Ethnic Diversity, Social Sanctions, and Public Goods in Kenya.” Journal of Public Economics 89 (11–12): 2325–68. Ngugi wa Thiong’o. 2009. Something Torn and New: An African Renaissance. New York: Nation Book. Oates, Wallace E. 1999. “An Essay on Fiscal Federalism.” Journal of Economic Literature 37 (3): 1120–49. Olson, Mancur. 1982. The Rise and Decline of Nations. New Haven, CT: Yale University Press. Reno, William. 1995. Corruption and State Politics in Sierra Leone. New York: Cambridge University Press. Richards, Paul. 1996. Fighting for the Rainforest: War, Youth and Resources in Sierra Leone. London: James Currey. Srivastava, Vivek, and Marco Larizza. 2011. “Decentralization in Postconflict Sierra Leone: The Genie is out of the Bottle.” In Yes Africa Can: Success Stories from a Dynamic Continent, edited by Putnam Chuhan-Pole and Manka Angwafo. Washington, DC: World Bank. United Nations. 1993. Human Development Report 1993. New York: United Nations Development Program, Oxford University Press. ———. 2009. Human Development Report 2009. New York: United Nations Development Program, Oxford University Press.

2

The Political Economy of Government Revenues in Postconflict Resource-Rich Africa Liberia and Sierra Leone Victor A. B. Davies and Sylvain Dessy

2.1

Introduction

Liberia and Sierra Leone have been hailed as success stories of political transition from civil war in Africa, having emerged from it as budding democracies. Sierra Leone has had two peaceful postwar elections, the second of which was won by the opposition, while Liberia has elected Africa’s first female president. However, at the end of the war, the two countries faced the daunting challenge of resuscitating collapsed government revenues to help finance the peacebuilding and reconstruction process. Government revenues plummeted in Liberia from around US$240 million just before the onset of war in 1989 to about US$46 million at the onset of peace in 2003, and in Sierra Leone from 17 percent of gross domestic product (GDP) in the 1970s to around 12 percent at the end of the 1991–2001 civil war. Natural resources were a key element of the revenue challenges facing Liberia and Sierra Leone. Both countries are endowed with a variety of natural resources with considerable scope for revenue generation. However, Victor A. B. Davies is senior economist in the African Department of the International Monetary Fund. Sylvain Dessy is professor of economics at Université Laval. We gratefully acknowledge funding from the NBER Africa Project for this research. We appreciate the intellectual support of Simon Johnson, the lead co-chair of the project responsible for reviewing this research, and that of Sebastian Edwards and David Weil, the other lead co-chairs. We have benefited from comments on the chapter by participants at the NBER Africa Project conference in Zanzibar, Tanzania, in August 2011. Research assistance from Alvin Johnson in Liberia, Alimamy Bangura in Sierra Leone, Francis Andriarinason and Safa Ragued from Université Laval, Canada, is gratefully acknowledged. Elisa Pepe, coordinator of the project, has provided diligent and timely support. The views expressed herein are those of the authors and should not be attributed to the IMF, its executive board, its management, or the NBER. For acknowledgments, sources of research support, and disclosure of the authors’ material financial relationships, if any, please see http://www.nber.org/chapters/c13389.ack.

33

34

Victor A. B. Davies and Sylvain Dessy

the resources sometimes played a pernicious role in the economy before the war and helped to fuel the conflict in the two countries. In wartime Liberia, the warlord Charles Taylor presided over commerce in gold, timber, and rubber (Reno 1998). Taylor godfathered an arms-for-diamonds trade with the Sierra Leone rebels (UN 2000), prompting United Nations sanctions on export of diamonds and timber from Liberia.1 In prewar Sierra Leone, diamonds were at the epicenter of a thriving underground economy with the bulk of the diamonds smuggled abroad. The diamond mines attracted thousands of young quasi-criminal illicit diggers, generating a recruitment base and financing for the rebellion (Reno 1998; Smillie, Gberie, and Hazleton 2000). Thus, natural resources posed risks to the peace process while offering considerable scope for revenue generation in postwar Liberia and Sierra Leone. The objective of this chapter, therefore, is to examine and draw lessons from the strategies adopted by Liberia and Sierra Leone to generate revenues from their natural resources. As part of the analysis, we develop an analytical model to illuminate the revenue challenges facing a resource-rich country emerging from conflict. We find that after the war, despite weak resource management capacity, the governments of Liberia and Sierra Leone initiated the process of exploiting their natural resources, offering large fiscal concessions. In Sierra Leone, the government argued that generous fiscal concessions were necessary to attract foreign investors to a high-risk environment. However, the fiscal concessions that both countries offered were often perceived to be overly generous, while in Liberia, the government revealed a preference for initial payments seemingly at the expense of longer-term benefits. Concerns endure that, despite revisions, some current exploitation contracts still offer overly generous fiscal concessions to exploiting companies. Moreover, fiscal terms vary, creating an uneven playing field that could discourage investment and distort incentives toward lobbying for better contract terms. Against this backdrop, our analytical model illuminates the revenue challenges facing a government emerging from conflict in a resource-rich country like Liberia or Sierra Leone. We assume the government has a choice between two revenue rehabilitation strategies. The first, the impatience strategy, is to contract a multinational mining company to exploit the resources right away and share the profit with the company. However, the mining cost is the company’s private information, providing an incentive for the company to cheat by inflating it. The second strategy, patience, is to defer exploitation of the resource, and rely on alternative tax instruments while developing resource management capacity. The model analyzes the challenges associated with the impatience strategy, which the government is likely to pursue 1. Charles Taylor is currently standing trial in The Hague for war crimes and crimes against humanity for his role in the Sierra Leone civil war.

The Political Economy of Government Revenues in Postconflict Africa

35

given the urgent need to generate revenues to consolidate the peace process. The findings call attention to the potential role of the international community in developing postconflict countries’ natural resource and revenue institutional capacity, as well as transparent corporate and government institutions for resource management. While both countries initially adopted an impatience revenue strategy, Liberia appears to have made more progress toward developing tax and resource management capacity. According to the World Bank 2011 Doing Business Index, the process of paying taxes is more efficient in Liberia. Time spent to pay taxes per year is 158 hours in Liberia and 357 hours in Sierra Leone. Liberia has made efforts to develop a fiscal framework for natural resources with the help of the international financial institutions. Both countries have taken measures to introduce transparency in the management of natural resources and reduce the risk of the resources fueling conflict. They have enacted legislation setting aside a share of the revenues for local communities, and have joined the Extractive Industries Transparency Initiative (EITI). Liberia has gone much further, becoming Africa’s first EITI compliant country in 2009. Sierra Leone has yet to reach that stage. Liberia also appears to have made more efforts to fight corruption in public management. Between 2005 and 2010, it moved from a rank of 137 out of 158 countries to 87 out of 178 countries on Transparency International’s Corruption Perception Index, while Sierra Leone’s rank deteriorated from 125 to 134. Unsurprisingly, Liberia’s revenue-to-GDP ratio has been higher: 30 percent in 2010 compared to Sierra Leone’s 13 percent. That national income statistics have not been compiled in Liberia for many years suggests caution in the interpretation of this ratio, however. The rest of the chapter is structured as follows. Section 2.2 provides a context for the study with an overview of political and economic developments in Liberia and Sierra Leone. Section 2.3 analyzes the two countries’ postconflict strategy to raise revenues from their natural resources. Section 2.4 discusses international initiatives to support revenue rehabilitation in postconflict resource-rich countries. Section 2.5 presents the model. Section 2.6 concludes. 2.2

The Postconflict Context in Liberia and Sierra Leone

In this section we present a brief outline of the socioeconomic and political context characterizing postconflict Liberia and Sierra Leone. The two countries share a border and many features. The civil wars in the two countries have been blamed on similar factors: state failure in the 1980s, induced by a sharp decline in economic and political governance (Reno 1998). In Liberia, a military coup by Sergeant Samuel Doe in 1980 ended the 130-year aristocracy of the Americo-Liberian ethnic group—descendants of resettled freed slaves from the United States—but only magnified the

36

Table 2.1

Year 1970–79 1980–84 1985–89 1990–94 1995–99 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

Victor A. B. Davies and Sylvain Dessy

Liberia: Economic indicators GDP (current million US$)

Real GDP growth (annual %)

580 940 930 250 280 560 543 560 410 460 530 610 735 840 880 1,000

3.0 –2.5 –6.4 –30 33 26 3 4 –31 3 5 8 9 7 5 6

GDP per capita (constant 2000 US$)

Population (million)

800 650 522 130 114 199 195 196 131 131 133 138 144 148 148

1.6 2.0 2.2 2.0 2.3 2.8 3.0 3.1 3.1 3.2 3.3 3.5 3.6 3.8 4.0

External debt stocks (% of GNI) 42 106 230 — 830 723 747 721 1,028 1,027 940 937 669 465 257 12

Total external debt stock (current million US$) 600 900 1,670 2,190 2,480 2,810 3,000 3,270 3,600 3,830 3,920 4,160 3,790 3,130 1,670 115

Source: World Development Indicators (2010).

underlying problems of corruption, nepotism, and repression (Government of Liberia 2005). Charles Taylor launched the rebellion on Christmas Eve 1989, with the declared goal of remedying the country’s myriad problems. That year Liberia’s per capita income was US$400, less than half of its level in the early 1970s, while its external debt stock was 319 percent of gross national income (GNI), compared with about 30 percent in the mid-1970s (table 2.1). Splinter groups emerged in the rebellion. The Economic Community of West African States Monitoring Group (ECOMOG), a force assembled by the Economic Community of West African States (ECOWAS), intervened in 1990, using bases in Sierra Leone. The intervention prevented Taylor from taking complete power but left him in control of much of the country outside of the capital, Monrovia. In 1997, ECOMOG conducted presidential elections, which Taylor won. However, Taylor failed to transition from warlord to statesman. He sought to eliminate political opponents, godfathered an arms-for-diamonds trade with the Sierra Leone rebels, and kept most of the trade in Liberia’s natural resources off budget (Economist Intelligence Unit 2007). The country soon returned to war. In 2003, with rebels closing in on Monrovia, Taylor agreed to relinquish power to an interim government and to go into exile. In 2005, elections were conducted, which were won by Mrs. Ellen Johnson Sirleaf. Turning to Sierra Leone, the roots of the 1991–2001 civil war have been

The Political Economy of Government Revenues in Postconflict Africa

37

linked to the political leadership of the preceding years that emphasized informalization and control of markets and their reward, and replacement of political competition with a struggle for political favor (Reno 1998). There was extensive intervention in key markets for rice (the staple food), foreign exchange, diamonds, and agricultural exports while political and civil opposition was curtailed, culminating in a one-party state in 1978. By the mid-1980s, burgeoning black markets, dysfunctional infrastructure, and high unemployment were the norm. About 80 percent of the population was estimated to live below the poverty line of US$1 a day. The inflation rate sometimes exceeded 100 percent while government revenues were about 8 percent of GDP, down from 17 percent in the 1970s (Davies 2007). Real GDP growth was negative (table 2.2). The rebellion was launched in 1991 with help from Charles Taylor’s rebel movement in Liberia. Like in Liberia, the declared aim was to redress the economic and political malaise afflicting the country. The rebels pursued their campaign despite a military coup in 1992. The military transferred power to an elected government in 1996, which was overthrown by the military in alliance with the rebel movement in 1997. The West African force, ECOMOG, intervened to reinstate the elected government in 1997. The war finally ended in 2002. The war in Liberia claimed 270,000 lives (Government of Liberia 2005) from a prewar population of about three million, compared with 50,000 deaths in Sierra Leone, from a prewar population of about 4.5 million. In Table 2.2

Year 1970–79 1980–84 1985–89 1990–94 1995–99 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

Sierra Leone: Economic indicators

GDP (current US$) 658 112 807 758 801 636 806 936 991 1,096 1,240 1,422 1,664 1,955 1,942

Real GDP growth (annual %)

GDP per capita (constant 2000 US$)

2.7 2.9 –0.6 –2.8 –5.7 3.8 18.2 27.5 9.3 7.5 7.2 7.3 6.4 5.5 4.0 4.9

Source: World Development Indicators (2010).

280 290 260 230 180 150 172 210 221 228 236 245 254 261 265

Population 2.9 3.3 3.8 4.1 4.0 4.2 4.4 4.5 4.7 4.9 5.1 5.3 5.4 5.6 5.7 5.7

Total external debt stock (% of GNI)

Total external debt stock (current million US$)

31 54 131 201 164 193 154 150 158 157 131 93 19 21 23

211 589 942 1,371 1,228 1,190 1,200 1,355 1,520 1,630 1,540 1,270 312 399 444 770

38

Victor A. B. Davies and Sylvain Dessy

Liberia, much of the fighting took place in Monrovia, damaging much of the infrastructure, unlike in Sierra Leone where much of the fighting was outside of Freetown. At the end of the war, Monrovia had no running water or public electric supply, unlike Freetown. The UN has staged two of its largest peacekeeping operations in the two countries to help end the civil war. It deployed 17,000 peacekeepers in Sierra Leone and a similar number in Liberia. The two countries have emerged from civil war as budding democracies. Sierra Leone has conducted two peaceful postwar elections, with the opposition winning the second. Liberia has also conducted two peaceful postwar elections. It elected Mrs. Ellen Johnson Sirleaf, Africa’s first female president, in the first elections in 2005 and reelected her in the second in November 2011. Liberia and Sierra Leone are both resource-rich coastal economies. Diamonds have played a prominent role in Sierra Leone for good and ill. They accounted for 90 percent of exports and some 15 percent of GDP in the 1960s and 1970s and about a quarter of revenues. Iron ore, timber, and rubber (mainly from large-scale plantation agriculture) have historically been Liberia’s leading exports. Sierra Leone has discovered oil (though the commercial viability is yet to be confirmed), while iron ore is set to become its leading export beginning in 2012. Liberia is also actively prospecting for oil. Agriculture is the dominant economic activity, accounting for 60 percent of GDP in Liberia, and 50 percent in Sierra Leone. At the end of the civil war, Liberia and Sierra Leone embarked on International Monetary Fund (IMF)-supported economic reform programs. In Liberia, a key conditionality was the operation of a cash-based balanced budget. The Liberian economy is highly dollarized with a US dollar component of over 70 percent of broad money. The two countries have benefited from forgiveness of almost all their debts to the multilateral financial institutions after reaching the completion point of the Highly Indebted Poor Countries (HIPC) initiative. As a result, the external debt stock has fallen in Liberia from over 1,000 percent of GNI in 2003 to 12 percent in 2010, and in Sierra Leone from 193 percent of GNI in 2000 to 23 percent in 2009. 2.3

Revenue Measures and Outcomes

In this section we focus on the strategies adopted by Liberia and Sierra Leone to generate revenues from their natural resources. To provide a context for the analysis, we start with a review of the institutional framework for revenue administration. 2.3.1

Revenue Administration, Policy, and Outcomes

Liberia and Sierra Leone implemented a number of measures to rehabilitate and reform their collapsed revenue infrastructure. In 2003, Sierra Leone set up the semiautonomous National Revenue Authority to supervise

The Political Economy of Government Revenues in Postconflict Africa

39

“all aspects of revenue collection.” Hitherto, customs duties and income taxes were collected independently, while other revenues like licence fees for mobile telephone operators and fishing vessels were collected by the line ministries. Other African countries such as Ghana and Uganda have set up a similar revenue authority, seeking to improve coordination of revenue collection activities, reduce duplication, and insulate revenue administration from undue external influence. However, unlike the pattern observed in other countries, following the setting up of the National Revenue Authority, Sierra Leone’s revenue to GDP ratio fell from 12.3 percent in 2003 to 11.9 percent in 2005 and to 11.7 percent in 2009. A 2007 report on the Authority highlighted the following problems: too many departments and vehicles, weak internal control over revenue collection, and political nepotism (Sierra Leone 2008). The Sierra Leone experience suggests that the fundamental governance problems facing revenue institutions should be addressed prior to the setting up of an independent revenue authority. In Liberia, customs duties and income taxes are collected by two departments in the ministry of finance. Plans are underway to set up an independent revenue authority. Both countries have moved toward automating customs revenue collection by introducing the Automated System for Data Analysis (ASYCUDA) developed by the United Nations Conference on Trade and Development (UNCTAD). Automation could eliminate a major source of private revenues for corrupt customs officials and reduce the scope for duty evasion. Unsurprisingly, it has met with major obstacles in both countries. Nevertheless, Liberia was able to operationalize within eighteen months the more sophisticated, Internet-based ASYCUDA World in the Freeport of Monrovia, which handles 85 percent of the country’s trade. In Sierra Leone, efforts are still ongoing to operationalize the less sophisticated ASYCUDA++.2 In the area of tax reform, Sierra Leone introduced a value-added tax (VAT) in 2010 to replace the domestic and import sales tax. The aim has been to replace inefficient production and sales tax with a less distortionary tax, to increase revenue, and improve tax efficiency. Plans are underway to implement the VAT in Liberia by 2013. Table 2.3 examines indicators of the ease of paying taxes from the World Bank Doing Business Index. The table suggests that prior to 2012—the indicators for any year are based on statistics as of June the previous year—paying taxes was easier and more efficiently organized in Liberia than in Sierra Leone. For instance Liberia ranked 85 and 84 out of 183 countries in the 2010 and 2011 ease of paying taxes index, compared with Sierra Leone’s 160 and 159. In 2012, Sierra Leone’s ranking improved considerably to 76 out of 183 countries, while Liberia’s deterio2. Liberia has faced some logistical challenges in the implementation of ASYCUDA: The Free Port of Monrovia has no electric power supply. A 20 KVA generator was purchased and is operated for ten hours every day. The ASYCUDA system is operated during those ten hours. Limited band width for Internet access is another constraint.

40

Victor A. B. Davies and Sylvain Dessy

Table 2.3

Liberia

Sierra Leone

The ease of paying taxes in Liberia and Sierra Leone

Year

Country rank (ease of paying taxes)

Number of payments per year

Time (hours per year) needed to pay taxes

Total tax payable (% of gross profit)

2008 2009 2010 2011 2012 2006 2007 2008 2009 2010 2011 2012

119/178 59/181 85/183 84/183 98/183 — — 145/178 160/181 160/183 159/183 76/183

37 32 32 32 33 20 20 22 28 29 29 29

158 158 158 158 158 399 399 399 399 357 357 357

81.6 35.8 43.7 43.7 43.7 163.9 277 233.5 233.5 235.6 235.6 32.1

Source: Doing Business Report (various years): International Finance Corporation. Note: Total tax payable is the sum of all taxes and mandatory payments like employer contribution to employee social security.

rated to 98. Liberia’s tax administration has been much more efficient in terms of minimizing the time spent to pay taxes: an average of 158 hours a year from 2008 to 2012, compared with 357 to 399 hours in Sierra Leone. However, the number of payments required per year in Sierra Leone has been slightly fewer, twenty-nine, compared with thirty-two in Liberia. Total taxes payable by businesses—the sum of all taxes and mandatory payments like employer contribution to employee social security—appears to have been inordinate in Sierra Leone prior to 2012: about 236 percent compared to about 44 percent in Liberia. Such a high rate of taxation would have been a recipe for tax evasion. As of writing, details were not available about the reforms undertaken in Sierra Leone that has resulted in the improvement in its 2012 ranking on the overall ease of paying taxes and the reduction in total taxes payable. Tables 2.4 and 2.5 provide statistics on government revenues in Sierra Leone and Liberia. The tables indicate that revenues from natural resources have been modest in Sierra Leone, with mining revenues (including from diamonds) accounting for less than 4 percent of total revenues (less than 0.4 percent of GDP) postwar. This figure excludes payroll taxes and personal income taxes of employees in mining companies. When these are included, mining revenues were about US$28 million in 2010 (12 percent of total revenues and 1.5 percent of GDP). The country has been unable to generate large revenues from the exploitation of alluvial diamonds despite exports of over one hundred million dollars in recent years (table 2.3) due to the difficulty of policing the resource. The deposits are widely dispersed and can be mined with simple handheld tools, facilitating illicit mining and smuggling

The Political Economy of Government Revenues in Postconflict Africa

41

to evade taxes. Thus, export taxes have been kept low—3 percent initially and then 6 percent in 2010—to reduce the incentive for smuggling. Production of diamonds from kimberlite deposits began in 2003 by the Koidu Holdings company. Kimberlite deposits are spatially concentrated and require capitalintensive mining techniques, precluding illicit mining and hence offering better revenue prospects. In Liberia, revenues from natural resources have apparently also been modest in the postwar period. An exception is one-off initial payments totaling US$80 million made between 2009 and 2011 by four iron ore-mining firms upon signing of mining concessions. It is not clear under which category these payments have been recorded in table 2.5. Iron ore production was scheduled to start by the end of 2011. While rubber has been the main export, accounting for 40–95 percent of total exports (table 2.4), details of its fiscal contribution are lacking. The export of timber, a major prewar source of revenue, was banned by the UN till 2007. Subsequently, stumpage fees (a percentage of the free on board [FOB] price of timber) and land rental fees accounted for 5 percent of total revenues in 2010. Sierra Leone and Liberia have relied heavily on international trade taxes over the years. These have accounted for 50–60 percent of total revenues in Sierra Leone and 30–40 percent in Liberia. Petroleum imports have been taxed heavily, generating import duties of over 10 percent of total revenues

Table 2.4

Government revenues in Sierra Leone Composition of total revenues (%) International trade taxes

Year

Total revenues (million USD)

Total revenues (% GDP)

Income tax

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

73 105 114 123 132 144 167 180 222 223 253

11.4 13 12.2 12.3 11.9 11.9 11.8 10.8 11.4 11.7 13.3

26 26 26 25 26 28 28 27 30 28 30

Excise duty on petroleum import

Other international trade taxes

Roaduser charges

Mining revenues

Other nontax revenues

12 10 12 13 12 11 13 12 8 13

50 50 14 12 47 42 40 42 42 40

3 3 3 3 5 6 8 6 7 7 5

2.5 1.9 1.8 2 3.6 2.5 3 3.5 2.8 2.7 2.4

8 5 6 6 6 5

Source: Government of Sierra Leone. Notes: Other nontax revenues include license fees paid by fishing vessels and by mobile phone companies, fines, rent of government lands and buildings, and fees and charges. Mining revenues include mining license fees, prospecting rights, mining leases, and royalties.

79 65 73 46 68 77 11 175 201 211 275

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

13 11 15 15 19 24 25 25 30

Total revenue (% GDP)

Source: Government of Liberia.

Total revenue (million USD)

11 17 14

12 9 9 9 17 15

Personal income tax

Government revenues in Liberia

Year

Table 2.5

6 7 10 4 11 22 28 29 10 12 9

Corporate profit tax 29 29 23 37 40 38 44 42 39 38 32

Taxes on international trade 9 9 8 5 1 1 3 2 3 4 4

Petroleum sales tax

23 17 18 29 20 12 11 8 7 6 6

Maritime revenue

Key components of total revenues (%)

8 13 18 4 0 0 0 0 0 1 5

Stumpage/ land rent

13 16 14 12 11 22 14 19 30 22 30

Other

The Political Economy of Government Revenues in Postconflict Africa

43

in Sierra Leone. In Liberia, maritime revenues—registration fees from foreign ships flying the Liberian flag—have been a major source of revenues, especially before the end of the conflict in 2005, accounting for 18–29 percent of total revenues. 2.3.2

Natural Resource Revenue Strategies

With the onset of peace the governments of Liberia and Sierra Leone quickly embarked on the exploitation of natural resources. However, the institutional capacity to negotiate and implement contracts was weak while a defined fiscal regime for natural resources was lacking. Moreover, there were existing contracts that wartime governments had hastily signed. Unsurprisingly, the fiscal terms initially agreed to were perceived to give less than full value to the two countries for their natural resources. The rush to exploit natural resources was to be expected in Liberia in particular. It had a large, unelected transitional government from 2003 when the war ended to 2006 comprising members of various warring factions. In Sierra Leone, the government in power in 2002 at the end of the conflict was elected in 1996, and reelected in 2002. In Liberia Mrs. Johnson Sirleaf’s elected government, upon taking office in January 2006, decided to renegotiate all natural resource agreements entered into by the transitional government. The new government also subjected the 300-plus members of the transitional government to a special audit, many of whom, including the leader, were subsequently convicted of corruption (Economist Intelligence Unit 2007). The new government went on to sign iron ore-mining agreements with Arcelor Mittal, the leading global steel producer, in 2006, and later with three other mining companies. Following widespread concerns about generous fiscal concessions, the government renegotiated the contract with Arcelor Mittal. In Sierra Leone Koidu Holdings began to mine kimberlite diamonds in 2003, having purchased the mining rights from Branch Energy, based on an agreement signed during the war. The government signed an agreement for the mining of rutile with Sierra Rutile in 2001, and bauxite with Sierra Minerals Holdings Limited in 2004. The government of President Koroma, elected from the opposition in 2007, signed iron ore mining contracts with London Mining in 2009, and with African Minerals in 2010. Like in Liberia, widespread concerns about generous fiscal concessions led to renegotiation of contracts. Petroleum exploration agreements, signed from 2001, were the only exception. The prospects of petroleum discovery were seemingly poor at the time. Despite the sometimes repeated revisions, concerns endure that current exploitation contracts still offer overly generous fiscal concessions as well as considerable scope for fiscal abuse. Notably, in Sierra Leone, the government bridged a loan from the European Union to provide start-up capital for Sierra Rutile, bearing much of the production risk, and set the royalty

44

Victor A. B. Davies and Sylvain Dessy

rate at a meagre half a percent. The government argued that generous fiscal concessions were necessary to attract foreign investors at a time when the country was perceived as a high-risk environment. African Minerals, an iron ore-mining company in Sierra Leone, enjoys “duty and excise-free import of all equipment and consumables for company and contractors throughout mine life.” The norm in the industry in other countries is to limit duty-free imports to the start of production. Moreover, the difficulty of distinguishing contractor’s imports intended for use by African Minerals from those intended for other purposes represents a potential source of revenue loss for the government. Another concern is that fiscal terms vary within a sector, creating an uneven playing field. This could discourage investment by signaling unpredictability and lack of policy consistency, and could also produce an incentive for investors to lobby for better contract terms. Tables 2.6 and 2.7, which summarize key fiscal terms in iron ore contracts in Liberia and Sierra Leone, show that in Liberia in particular, there are differences in the tax rates, royalties, and exemptions on import duties. This has been the result of case-bycase negotiations to resolve overlaps and ambiguities caused by the existence of a number of laws governing the fiscal terms of natural resource contracts—the New Minerals and Mining Law, the Liberian Revenue Code, and the Investment Incentive Code of 1973. Some terms of individual contracts are inconsistent with existing legislation. For instance, the income tax rate agreed to with BHP Billiton and China Union in Liberia is 25 percent compared with 30 percent in the Liberian Revenue Code.

Table 2.6

Liberia: Comparative iron ore fiscal terms

Fiscal terms

BHP Billiton

China Union

Arcelor Mittal

Royalty

3%–4.25%, depending on price 25%

3.25%–4.5%, depending on price 25%

4.5%

5 Exempt 12 years from production

5 Exempt 10 years

Income tax rate Depreciation of development costs (years) Import duties

Upfront payment (million US$) Government equity participation

15 None

Source: Government of Liberia.

40 None

Putu 4.5%

Liberia Revenue Code stipulation 4.5%

30%

30%

30%

15 $400,000 for five years

5 Exempt till production

5 Exempt till production

10 No

– –

15 15% minimum

The Political Economy of Government Revenues in Postconflict Africa Table 2.7

45

Sierra Leone: Comparative iron ore fiscal terms

Fiscal terms

African Minerals

Royalty Income tax rate

3% on gross sales price of iron ore 25% over mine life

Import duties

Duty and excise-free import of all equipment and consumables for company and contractors throughout mine life None

Upfront payment (million US$) Government equity participation Explicit review process

10 percent “free” share in one of the parties in the project Yes

London Mining 3% on gross sales 6% for first three years, then 25% 20% of prevailing rate (in effect 1%)

None None No

Source: Government of Liberia.

In Liberia, the government revealed a strong preference for initial payments. The preference for initial payment, the main biddable item in iron ore contracts, led to major fiscal concessions and to renegotiation of the Bong mines iron ore contract already awarded by competitive bidding. The fiscal terms (other than initial payment) for the contract were initially fixed and declared not open to bidding or negotiations. China Union won the bid with a signature bonus of US$40 million. However, before the contract was signed, following the onset of the global financial crisis, China Union argued that, in the face of falling commodity prices, the signature bonus could only be maintained if major fiscal concessions were awarded. The government agreed to this, rather than turn to the second-ranked bidder in the auction that had offered a much smaller signature bonus. Table 2.5 summarizes natural resource contract award mechanisms in Liberia and Sierra Leone. In Sierra Leone there has been no clear institutional framework for awarding contracts unlike in Liberia, which has a centralized framework coordinated by an investment commission and involving line ministries. Liberia has moved toward competitive bidding for the award of contracts in forestry, iron ore mining, and petroleum exploration. However, as noted above, the competitive bidding process has sometimes been accompanied by direct negotiation with the successful bidder, resulting in a revision of bid terms. There have been considerable delays in payments and the start of production in Liberia. In the iron ore sector, China Union delayed for over a year in payment of the signature bonus while production by Arcelor Mittal was delayed by the global recession, which created cash flow problems for the company. Only one of the seven forest management contractors (awarded between October 2008 and September 2009) and three of the nine timber sales contractors (awarded between June 2008 and July 2010) had started

46

Victor A. B. Davies and Sylvain Dessy

exporting by early 2011. Two forest management contractors had still not paid tax arrears for the previous tax year, while none had paid area fees and the land rental. 2.4

Revenue-Rehabilitation Initiatives by the International Community

In recent years, the international community has perceived revenue rehabilitation in postconflict environments as a key element of its approach to building a legitimate and effective state as the basis for durable peace. This section highlights key initiatives that have been undertaken to rehabilitate revenues in resource-rich countries emerging from conflict and to address other related challenges. 2.4.1

Preventing the Use of Natural Resources to Fuel Conflict

The UN imposed sanctions on diamond exports from Sierra Leone, Angola, and Liberia during the civil war in these countries in an effort to stop the use of diamonds to finance conflict. It then adopted in 2000 an international certification scheme for rough diamonds. The scheme was succeeded by the Kimberley Diamond Certification Process launched in 2003 as “a joint government, industry and civil society initiative to stem the flow of conflict diamonds to finance wars against legitimate governments.” The scheme requires member governments to certify shipments of rough diamonds as “conflict-free.” As of January 2012, the Kimberley Process had fifty participants representing seventy-six countries—including Liberia and Sierra Leone and other postconflict countries—with the European Union and its member states counting as a single participant. Member countries account for nearly 100 percent of the global production of rough diamonds. The international diamond industry and civil society organizations also participate in the process. However, the benefits of the scheme for countries that are no longer in conflict are likely to be limited, given that the initiative was designed primarily for countries in conflict. 2.4.2

Increasing Transparency and Accountability in the Management of Natural Resources

The Extractive Industry Transparency Initiative (EITI) aims to strengthen governance by improving transparency and accountability in the extractives sector in resource-rich countries in general (including nonconflict countries). The EITI supports improved governance in resource-rich countries through the verification and full publication of company payments and government revenues from oil, gas, and mining. The EITI is a coalition of governments, companies, civil society groups, investors, and international organizations. It is a global standard that promotes revenue transparency and provides a procedure for monitoring and reconciling company payments and government revenues at the country level. Each implementing country creates its own EITI process, which is overseen by participants from

The Political Economy of Government Revenues in Postconflict Africa

47

the government, companies, and national civil society. In October 2009, Liberia became the first country in Africa, and the second in the world, to attain EITI compliant status—a stage in which a country is considered to have met all EITI requirements. Fourteen countries are now EITI compliant while twenty-one, including Sierra Leone, are at the initial “candidate status” stage. Other countries have signaled their intent to implement the EITI. Another initiative has been “Publish What You Pay”—a global network of civil society organizations that campaigns for transparency in the extractive industries. The initiative calls for companies to “publish what you pay” and for governments to “publish what you earn.” It was motivated by a report published by the advocacy organization Global Witness on the role of the oil and banking industries in the misappropriation of state assets in Angola during the civil war. Publish What You Pay membership currently spans sixty countries, including Liberia and Sierra Leone. 2.4.3

Strengthening Revenue Collection and Administration

The World Bank, IMF, African Development Bank, and bilateral donors have provided technical and financial assistance to help countries emerging from conflict rebuild their revenue collection and administration capacity. Notably, donors have funded the setting up of a semiautonomous independent revenue authority in Uganda and Sierra Leone, and the automation of customs revenue collection in Liberia and Sierra Leone that involved the setting up of the ASYCUDA software system. Donors have sometimes used conditionality to address concerns that aid might substitute, rather than develop, domestic revenue capacity. The European Union conditioned its 2002 budget support to Mozambique—which experienced civil conflict from 1977 to 1992—on increases in domestic revenue (Boyce and O’Donnell 2007). One of the benchmarks in the Afghanistan Compact of 2006, which sets out the framework for international assistance, was an increase in the revenue/GDP ratio from 4.5 percent in 2004–05 to 8 percent in 2010–11 (Boyce and O’Donnell 2007). 2.4.4

Preventing Illegal Exploitation of Natural Resources

Liberia has adopted a Chain of Custody certification scheme for timber and timber products, which partly financed the civil war. The Chain of Custody scheme, a requirement by the UN for lifting sanctions on timber exports, is a traceability system that confirms that taxes have been paid and certifies that the timber product comes from an ethically well-managed source. The Switzerland-based Société Générale de Surveillance (SGS) was hired in October 2007 to build, implement, and eventually transfer the Chain of Custody system to Liberia’s Forestry Development Authority (FDA). Also, the government of Liberia has signed Voluntary Partnership Agreements (VPA) with the European Union to ensure that all timber products exported to the European Union are of legal origin. The VPA is a bilateral agreement between the European Union and wood exporting countries that

48

Victor A. B. Davies and Sylvain Dessy

aims to improve forest governance and ensure that the wood imported into the European Union has complied with the legal requirements of the partner country. There is no obligation for any country to enter into a VPA with the European Union. However, once agreed, VPAs are legally binding on both parties, committing them to trading only in wood products that can be verified as legal. 2.4.5

Managing the Wealth from Natural Resources

In 2011 the IMF launched the multidonor Topical Trust Fund on Managing Natural Resource Wealth to finance technical assistance to low-income and lower-middle-income countries endowed with oil, gas, and minerals to help them deal with the associated economic policy challenges. The trust fund seeks to help build macroeconomic policy capacities and assist countries to get a fair share of their natural resource wealth, and invest and spend it wisely. The Netherlands, Norway, and Switzerland have made contributions to the trust fund, which concentrates on capacity building in five areas:

• extractive industries fiscal regime; • extractive industries revenue administration; • fiscal policies and public financial management specific to resource-rich countries;

• natural resources-related financial asset and liability management; and • statistics for natural resources. The trust fund will serve fifteen to twenty countries from a group of fifty eligible countries that have substantial current or prospective extractive industry revenue. Country commitment to reform will be a key selection criterion. 2.4.6

Enhancing the Benefits of Natural Resources to Host Communities

Such initiatives have been prompted by widespread concern that host communities benefit little from the exploitation of natural resources. One such initiative is the Peace Diamond Alliance in Sierra Leone sponsored by the United States Agency for International Development (USAID). In 2002, USAID brought together producers, buyers, advocates, Sierra Leone government officials, and heavyweight industry players such as DeBeers and Rappaport to form an alliance for the first time to monitor diamond royalties and fees, inform miners of the value of stones, start addressing environmental degradation, and reduce exploitation of miners, especially children. Training was provided to help miners assess the true value of rough diamonds before making a sale. The USAID also began a pilot program to give miners access to credit at reasonable terms to purchase mining inputs and avoid getting into serious debt. These strategies were designed to help miners sell their stones to the highest bidder.

The Political Economy of Government Revenues in Postconflict Africa

49

Generally, donors and domestic and international advocacy groups have advocated for measures to distribute some of the revenues from natural resources to the host communities. As a result, in Liberia the law stipulates the allocation of revenues from timber as follows: 10 percent of log export fees and of stumpage fees to the Protected Forest Areas Network, 30 percent of land rental fees to forest communities, 30 percent to counties, and 40 percent to general revenues. In Sierra Leone, the Mines and Minerals Act imposes a 1 percent expenditure tax on mining companies for community development. Also, half of the 20 percent profit-sharing tax on Koidu Holdings, a company mining diamonds from kimberlite sources, is paid directly to the mining community. With regard to alluvial diamonds, the government initially levied a 3 percent tax on exports—subsequently increased to 6 percent—and distributed half a percentage point to the diamond mining communities. (The half a percentage point was the rate when the export tax was 3 percent. We lack information about whether this rate may have increased at the higher export tax rate of 6 percent.) 2.5

A Model of Postconflict Revenue Rehabilitation

In this section, we provide a framework for analyzing the revenue rehabilitation challenges confronting postwar Liberia and Sierra Leone. Consider a benevolent government aiming to raise revenue to finance the peace consolidation process in the aftermath of a civil war. The country is endowed with a stock, θ, of a given natural resource, say, iron ore, which requires a large capital outlay to exploit. The government has a choice between two revenue strategies. The first is to contract a multinational mining company to mine the resource right away—the impatience strategy. The other strategy is to delay exploiting the resource, and in the meantime, rely on alternative nonresource tax instruments while developing the public management institutions that will ensure the government gets good value for the resource—the patience strategy. But in the aftermath of a devastating conflict, as in the case of Liberia and Sierra Leone, weak administrative capacity and a low nonresource tax base may limit the scope for raising nonresource tax revenues. Therefore, the government may have no choice but to adopt the impatience strategy. In what follows, we highlight the challenges confronting an impatient government in a postconflict environment. 2.5.1

The Payoff to Impatience

Suppose the government considers a contract requiring that the mining company pays a fraction λ ∈ [0,1] of the gross receipts, less the mining cost. Denote gross receipts by π, and let the reported mining cost, k, be restricted to the closed interval [k0, π]. This implies that the reported mining cost cannot be less than the true cost, k0, and cannot exceed total receipts, π < +∞.

50

Victor A. B. Davies and Sylvain Dessy

Under this contract, the government will receive a share λ(π − k) of the profit, with the remaining share accruing to the company, (1 − λ)(π − k). An important feature of the model is that the mining cost, k, is the company’s private information, providing an incentive for the company to cheat by inflating it. Aware of this, the government may need to invest a level of public funds, θ, to uncover the scheme used by the multinational firm to misrepresent the level of the mining cost. Transfer pricing is one possible misrepresentation scheme that the company could use. The sequence of events under this revenue rehabilitation strategy is as follows. First, the government signs a contract with the mining company to mine the resource. Next the mining company decides whether or not to cheat by inflating the mining cost, k. If it decides to cheat, it declares a mining cost k, which in turn gives rise to a net government revenue, (1)

R = l (p − k) − u,

if cheating is not detected and (2)

R = l (p − k0) − u

if it is. In the latter case, the multinational firm is assumed to repay all the money owed the government. After that, nature moves to reveal whether or not misrepresentation of the mining costs is detected. We denote as ρ the probability that misrepresentation of the mining cost is uncovered. Next, nature moves again to reveal whether or not the peace process is consolidated. Finally, the economy ends. Let (3)

s = k − k0

denote the extent of mining cost misrepresentation by the multinational firm, that is, the fictitious amount by which the true mining cost, k0, is inflated. We make the following assumption. The probability, ρ, that cheating is detected is given by: (4)

r = ff (u,s) ,

with (a) fθ > 0, (b) fσ > 0, and (c) fθθ> 0, where ϕ ∈ (0,1) denotes a scalar measuring the productivity of the detection technology available to the government. Furthermore, (5)

´s < 1,

where ´s = fs

s f (u,s)

denotes the elasticity of the probability that cheating is detected with respect to the size of the payoff from cheating.

The Political Economy of Government Revenues in Postconflict Africa

51

Assumption A.1 implies that poor institutional capacity is an issue not only when the postconflict government chooses to adopt a patience strategy, but also when it elects to follows the impatience strategy, because of the need to mitigate perverse incentives to the multinational firm. Indeed, a low ϕ implies a poor institutional environment that impedes government capacity to fight corruption in mining. This may also happen if accountability mechanisms are weak. Condition (5), on the other hand, states that the probability that cheating is detected is not too sensitive to the size of the payoff from cheating. This can happen when a large proportion of overseas transactions affecting the cost of mining occurs between the multinational and its parent company, thereby maximizing the secrecy of the mining operations. Another important feature of the model is that government revenue is an input in the peace process (for instance, it could be used to enhance job creation for former combatants). We thus add the following assumption: The probability that the country returns to war, p, is a decreasing function of the level of government revenue, R, invested in the peace process: (6)

p = 1 − gR

where γ ∈ (0,1) is a scalar. Assumption A.2 highlights the political salience of revenue rehabilitation in the aftermath of civil war and other devastating conflicts. It implies that government ability to rehabilitate state revenue is crucial to lasting peace, which is necessary for sustaining rebuilding efforts. Indeed, the lower the state revenue, R, the higher the probability that the peace process will collapse, leading to a recurrence of the conflict. We define the expected payoff from following the impatience strategy in the process of revenue rehabilitation to be the expected value of the probability that the peace process is consolidated, 1 − p: V = E(1 − p). The objective of the postconflict government is to maximize the expected value of this probability. This expected value can be written as follows using the definition of p (7)

V = g[rR + (1 − r) R].

From equation (7), substituting in equations (1), (2), (3), and (4), rearranging yields a reformulation of this expected payoff as follows: (8)

V (u,l,f,s) = (l (p − k0) − u − ls [1 − ff (u,s)]) g.

To fully highlight the challenges facing a postconflict government that adopts the impatience strategy, we need to specify the government choice of the level of public funds invested in mitigating corruption. Government Response to Cheating Taking the level of cheating σ = k – k0 as a proxy for the incidence of corruption in this postconflict country, the implications of the impatience

52

Victor A. B. Davies and Sylvain Dessy

strategy for government revenue may arguably depend on (a) the productivity of the cheating detection technology, ϕ, (b) the level of corruption, σ, and (c) the share of mining revenues accruing to the government, λ. Therefore, a benevolent government’s problem is to choose θ so as to solve: maxV (u,l,f,s). u

Since by assumption A.1 the function f is strictly concave in θ, the optimal choice, θ*, must therefore satisfy the following necessary and sufficient condition: ffu (u,s) ls − 1 = 0.

(9)

Applying the Implicit Function Theorem to equation (9) yields the government optimal choice of anticorruption action as follows: u* = Q (f,l,s)

(10) where

(i) Qf > 0, (ii) Ql > 0,

(iii) Qs > 0, as an implication of Assumption A.1. In other words, government effort to thwart corruption in mining extraction is higher (a) the more productive the detection technology, (b) the higher its claim to the mining revenues, and (c) the higher the firm’s payoff from cheating. 2.5.2

The Determinants of the Effectiveness of the Impatience Strategy

We now discuss factors that bear on the effectiveness of a postconflict revenue rehabilitation strategy based on impatience. With the above result, we can now reformulate the optimal government payoff from following the impatience strategy as follows using equation (8): V * (f,l,s) = (l (p − k0) − Q (f,l,s) − ls [1 − ff [Q (f,l,s) ,s]]) g. Hence the following proposition, which is obtained by a straightforward application of the Envelope Theorem: Let Assumptions A.1 and A.2 hold. Then, in a postconflict environment, (a) access to a more productive cheating detection technology raises the expected payoff from adopting an impatience revenue rehabilitation strategy (i.e., ∂V * / ∂f > 0), (b) as does government’s ability to secure a favorable mining deal (i.e., ∂V * / ∂l > 0 ). Furthermore, (c) this expected payoff decreases with the size of the payoff from cheating (i.e., ∂V * / ∂s < 0). Proposition 1 highlights the challenges facing a postconflict government pursuing the impatience strategy for rehabilitating state revenue. First, detec-

The Political Economy of Government Revenues in Postconflict Africa

53

tion of cheating is difficult because of weak institutional capacity for monitoring mining activity. Second, the government is often in a weak bargaining position relative to mining companies: it often lacks the technical capacity and international experience to secure a good deal for its natural resources. In these circumstances, as demonstrated in section 2.4, the international community could play a critical role to help ensure the country gets a fair share of the natural resource benefits. First, it could use aid to raise ϕ—the productivity of the cheating detection technology. The international community could also use aid, especially technical assistance, to strengthen the government’s negotiating capacity, raising λ, the government share of mining revenues. Third, the international community could pressure both the government and exploiting companies to be transparent in their actions. 2.5.3

Synthesis

We have developed a model to highlight the revenue rehabilitation challenges facing Liberia and Sierra Leone at the onset of peace. A key conclusion of the analysis is that prioritizing the development of public management institutions prior to the exploitation of the natural resources is required for a country to get full value for its natural resources. This conclusion concurs with Stiglitz (2007), who discusses the problems facing developing country governments in dealing with multinational oil exploitation companies. However, our analysis qualifies this conclusion: in the absence of alternative fiscal instruments, even a benevolent government might be forced to exploit the resources prior to developing sound public management institutions. Although Liberia and Sierra Leone seemingly rushed to exploit their natural resources at the onset of peace, by several measures Liberia appears to have pursued a more patient revenue rehabilitation strategy over time. It has developed a fiscal framework for the natural resource sector, with the help of the international financial institutions, and has also realized much more improvement in fighting corruption. Between 2005 and 2010, it moved from a rank of 137 out of 158 countries to 87 out of 178 countries on Transparency International’s Corruption Perception Index, while Sierra Leone’s rank deteriorated from 125 to 134. Liberia’s revenue-to-GDP ratio of 30 percent in 2010 is much higher than Sierra Leone’s 13 percent. However, we hesitate to advance this as evidence that Liberia’s patience has paid off because national income statistics have not been compiled in Liberia until recently, raising doubts about the reliability of GDP estimates. The model is based on a number of simplifying assumptions that can be relaxed, such as a benevolent government. With a not-so-benevolent government or a government with a short-time horizon, the tendency toward impatience would only be stronger. For simplicity, the government rehabilitation strategy was presented as a stark choice between patience and impatience. In reality, the choice of strategy would lie in between and could change over time. Lastly, we assumed that under the patience strategy, with time, strong

54

Victor A. B. Davies and Sylvain Dessy

public management institutions can be developed. However, we ignored the question of how long that may take, if ever. 2.6

Conclusion

At the end of their respective civil conflicts, Liberia and Sierra Leone initially adopted an impatience strategy to generate revenues from their natural resources, opting to initiate right away the process of exploiting the resources despite weak public management capacity. Unsurprisingly, the generous fiscal concessions often granted exploiting companies raised concerns about whether the two countries were getting good value for their resources. Such concerns endure despite the corrective measures that both countries have taken with the revision of mining contracts and efforts to develop public management capacity. A key lesson from the two countries’ experience and our analytical model is that resource-rich countries emerging from conflict face a difficult tradeoff between relatively large longer-term gains from their natural resources and smaller short-term revenues. With a narrow nonresource fiscal base, the need to consolidate the peace process may make adoption of the impatience strategy inevitable. Moreover, other factors, notably elections that generate huge spending pressures, may also induce impatience.The international community can mitigate the potential costs of impatience in revenue generation by providing assistance to enhance the capacity of countries emerging from conflict to manage their natural resources, especially with regard to the negotiation and implementation of contracts with exploiting companies. Home countries of exploiting companies can set and enforce ethical standards for exploiting companies in their activities abroad. Similarly, as demonstrated by ongoing initiatives, the international civil society can advocate for such standards and monitor the activities of both the government and resource exploiting companies.

References Boyce, J. K., and O’Donnell. 2007. “Policy Implications: The Economics of Postwar Statebuilding.” In Peace and the Public Purse, edited by J. K. Boyce and M. O’Donnell. Boulder, CO: Lynne Rienner Publishers. Davies, V. A. B. 2007. “Sierra Leone’s Economic Growth Performance, 1961–2000.” In The Political Economy of Economic Growth in Africa 1960–2000, vol. 2: Country Case Studies, edited by B. Ndulu et al., Cambridge: Cambridge University Press. Economist Intelligence Unit. 2007. “Liberia.” Country Report, Economist Intelligence Unit, London.

The Political Economy of Government Revenues in Postconflict Africa

55

Government of Liberia. 2005. “Liberia Poverty Reduction Strategy 2005.” Monrovia, Liberia. Reno, W. 1998. Warlord Politics and African States. Boulder, CO: Lynne Rienner Publishers. Sierra Leone. 2008. Presidential Transition Team Report. Freetown: Government of Sierra Leone. Smillie I., L. Gberie, and R. Hazleton. 2000. The Heart of the Matter: Sierra Leone Diamonds and Human Security. Ottawa: Partnership Africa Canada. Stiglitz, Joeseph. 2007. “What is the Role of the State?” In Escaping the Resource Curse, edited by Macartan Humphreys, Jeffrey D. Sachs, and Joseph E. Stiglitz, 23–52. New York: Columbia University Press. United Nations (UN). 2000. Report of Panel of Experts Appointed Pursuant to UN Security Council Resolution 1306 (2000), Paragraph 19, in Relation to Sierra Leone (S/2000/1195). New York: UN Security Council.

3

Does Decentralization Facilitate Access to Poverty-Related Services? Evidence from Benin Emilie Caldeira, Martial Foucault, and Grégoire Rota-Graziosi

3.1

Introduction

Over the past two decades, decentralization has been implemented by many developing countries, becoming a key element of public sector reform. By bringing decision makers closer to citizens, decentralization should alleviate information asymmetries, improve accountability, and increase the efficiency of public goods provision. In developing countries, decentralizaEmilie Caldeira is associate professor of economics at CERDI-CNRS at the University of Auvergne. Martial Foucault is professor of political science at Sciences Po Paris. Grégoire Rota-Graziosi is professor of economics at CERDI-CNRS at the University of Auvergne. We thank the National Bureau of Economic Research (NBER), which has funded this project since 2009. We are grateful to the Municipal Development Partnership (MDP) in Cotonou, with special thanks to Hervé Agossou, for their warm welcome and valuable assistance in collecting data, as well as for their fruitful comments and discussions. We also thank the Benin National Institute of Statistics and Economic Analysis, especially Cosmé Vodounou and Damien Mededji, for allowing us access to EMICoV surveys. We thank Elias Potek (University of Montreal, Geography Dept.) for his outstanding work in creating geographical maps in record time. We thank Simon Johnson (MIT), who acted not only as a scientific mentor throughout this research effort, but also as a valuable advisor. We warmly thank Michael Hiscox (Harvard University) for the valuable comments that helped make the chapter’s final version stronger and Antoinette Sayeh (IMF) for handling the chapter’s policy concerns when it was presented at the NBER conference in Zanzibar. We thank Odd-Helge FjeldStad (International Centre for Tax and Development) and François Vaillancourt (University of Montreal) for all of their constructive suggestions. We are grateful to Leonard Wantchekon (Princeton University) and the participants at the IREEP (Institut de Recherche Empirique en Economic Politique) conference, the CERDI (Centre d’Etudes et de Recherche sur le Developpement International) seminar and the CIRANO workshop, where a preliminary draft of this chapter was presented in November 2011. Finally, we acknowledge financial support from the NBER Program on African Successes, especially Elisa Pepe for her amazing support throughout this project. Any remaining errors are ours. For acknowledgments, sources of research support, and disclosure of the authors’ material financial relationships, if any, please see http://www .nber.org/chapters/c13388.ack.

57

58

Emilie Caldeira, Martial Foucault, and Grégoire Rota-Graziosi

tion is one of the main institutional reforms on international organizations’ and donors’ agendas to enhance public governance and ultimately reduce poverty. This strategy has been in place for a number of years, but has not undergone a systematic evaluation of its impact on well-being and local governance. Therefore, now that several years have elapsed, it seems an appropriate time to examine the success of decentralization in the struggle against poverty in sub-Saharan Africa. With this objective in mind, we analyze the effect of decentralization in Benin on access to some poverty-related services, namely water, sanitation, refuse and sewage disposal, and primary education. Poverty is a multidimensional issue and basic health and education services are fundamental human rights.1 Decentralization is, by definition, a transfer of competencies to local governments, especially in the education and health sectors. These services do not correspond exactly to the Samuelsonian definition of pure public good (nonexcludability and nonrivalry). However, local and central governments share the responsibility for meeting fundamental rights in education and health. Whatever the means of producing such basic services, or nature of relationships with providers, local decision makers ultimately remain politically responsible (World Bank 2004) for achieving improvements in access to drinking water, sanitation, and primary education. In a sense, our argument rests on how decentralization facilitates access to high-quality services rather than on an investigation of how well publicly provided local goods are delivered. In regard to its democratization and decentralization processes, Benin is representative of African French-speaking countries. An ethnically fragmented country that has been politically stable only since 2001, Benin began a transfer of competencies to seventy-seven local governments, called communes, in 1998. The decentralization process definitively took off with local elections in 2002. Our analysis focuses on the 2006–2007 period, which corresponds to a crucial time for democracy in Benin, with the 2006 national elections bringing Yayi Boni to power in place of Mathieu Kerekou, who had ruled the country for twenty-nine years.2 By analyzing panel data from seventy-seven Beninese communes for 2006 and 2007, we aim to shed light on the following issues: (a) To what extent does decentralization, measured as the share of local own-revenue in total local revenue for each commune, modify access to basic public services? Is this effect monotone with the degree of decentralization? (b) Does the decentralization effect vary between communes according to their wealth? To answer these questions, we compiled several databases: original public finance panel data, which concerns all Beninese local governments, and the 2006 and 2007 Integrated Modular Survey on Household Living Conditions 1. Articles 25 and 26 of the Universal Declaration of Human Rights. 2. In the spring of 2011, President Yayi Boni was reelected for his second and last mandate.

Does Decentralization Facilitate Access to Poverty-Related Services?

59

(EMICoV), which covers a sample of 18,000 Beninese households throughout the entire national territory and is representative at the commune level. We have developed a consistent econometric method, taking into account potential endogeneity in the degree of decentralization, the heterogeneity of local governments, and inefficiency in estimating the effects of variables having little within variance. Our analysis suggests that, on average, decentralization increases access to basic public services. However, this effect is not only nonmonotone, following an inverted U-shaped curve, but its impact is also heterogeneous between poor and nonpoor communes. Decentralization’s effect on access to povertyrelated services is positive for sufficiently wealthy communes (measured by higher quintiles of an asset-based measure of wealth), and becomes negative for the poorest ones. Therefore, although decentralization succeeds in reducing nonmonetary poverty in Benin by improving access to some basic services, the risk of creating higher inequalities between communes remains. A second important result is that communes seem to prioritize basic services. Distinguishing local jurisdictions by their wealth allows us to shed light on significant differences in local governments’ behaviors. While the poorest jurisdictions neglect primary education, focusing more on access to drinking water, the richest ones pay less attention to sewage services, since they are already provided at a sufficiently high level. Although the latter is not an issue, the former casts some doubt on the efficiency of decentralization. The chapter is structured as follows: Section 3.2 presents a review of the literature on the impact of decentralization on service delivery and human development indicators in developing countries. Section 3.3 portrays the process of decentralization in Benin. Section 3.4 describes our econometric framework. Section 3.5 presents our results. Section 3.6 concludes. 3.2

The Impact of Decentralization on Services Delivery and Human Development Indicators: A Review of the Literature

A huge portion of economic literature focuses on decentralization in developing countries. A brief review of this literature addresses the strengths and weaknesses of decentralization in reducing poverty, or at least in increasing the efficiency of public goods provision. Many benefits of decentralization are claimed in the fiscal federalism literature, most of them related to the fact that decentralization brings decisions closer to citizens, alleviating information asymmetries, and improving local governments’ accountability. The fiscal federalism literature has largely stressed the economic efficiency of intergovernmental competition in providing local public goods. If such a normative prescription seems to fit well with developed countries, this issue remains more complex for developing countries where the “voting with your feet” mechanism is not so relevant. Thus, the logic of decentralization raises some intriguing issues in developing countries that we can summarize from

60

Emilie Caldeira, Martial Foucault, and Grégoire Rota-Graziosi

two perspectives: (a) Why does decentralization entail a better provision of local public goods? and (b) What are the limits of decentralization in such countries, given their institutional and geographical constraints?3 A demand-side argument in favor of decentralization is derived from the existence of information asymmetries. Indeed, the seminal idea that decentralization may improve the provision of public services when local governments have an informational advantage goes back at least as far as Hayek (1948) and Oates (1972). Since local decision makers have a better knowledge of local preferences, decentralization is expected to improve the level and quality of public services. This informational gain may induce a better targeting of the poorest populations in a country, as indicated by the research conducted by Alderman (2002) in Albania, Bardhan and Mookherjee (2005) in West Bengal, and Galasso and Ravallion (2005) in Bangladesh. On the supply side, decentralization should enhance the accountability of policymakers. Decentralization allows for a better provision of public goods and a better match between public policies and local needs. Several authors have established such a link: Bird and Rodriguez (1999) in the Philippines (health, primary education, housing, and infrastructure); Faguet (2004) in Bolivia (education and social services); Galiani, Gertler, and Schargrodsky (2008) in Argentina (education); Robalino, Picazo, and Voetberg (2001) on a panel discussing low- and high-income countries from 1970 to 1995 (mortality rate); and Enikolopov and Zhuravskaya (2007) on seventy-five developing and transition countries for twenty-five years (DPT immunization,4 infant mortality, illiteracy rate, and pupil-to-teacher ratio). Other studies mitigated the impact of decentralization. For instance, Azfar and Livingston (2010) find little evidence of better provisions in government services by local governments in Uganda; for Winkler and Rounds (1996), the transfer of education competencies in Chile reduced the scores of cognitive tests. Beyond improving the matching of public policies with local preferences, decentralization is also considered as an essential support of democratization. Thus, the governance of local public goods is expected to strengthen accountability under the strong assumption of well-informed voters, mobility of citizens, and participation in the political market. Seabright (1996) considers allocations of power within local and central governments as alternative means of motivating governments to act in the interests of citizens. This author shows that, although centralization entails benefits from policy coordination, it also induces some costs in terms of diminishing account3. Important literature has been devoted to analyzing the benefits of decentralization on human development indicators in the context of the Millennium Objectives. The final impact of decentralization on growth has been studied, for instance, by Zhang and Zou (1998), Wollera and Phillips (1998), Xie, Zou, and Davoodi (1999), Lin and Liu (2000), Akai and Sakata (2002), and Martinez-Vazquez and McNab (2003). 4. Diphtheria, pertussis, and tetanus.

Does Decentralization Facilitate Access to Poverty-Related Services?

61

ability. Moreover, interjurisdictional competition may enhance accountability: local citizens encourage incumbents to increase the efficiency of public spending through a “vote with their feet” (Tiebout 1956) or a “yardstick competition” (Salmon 1987; Besley and Case 1995).5 Few studies have examined the relevance of this phenomenon in developing countries: Arze, Martinez-Vasquez, and Puwanti (2008) suggest a process of yardstick competition between local governments in Indonesia; Caldeira, Foucault, and Rota-Graziosi (2015) establish the existence of strategic complementarities of local public goods among Beninese communes. However, by expanding the decision space of local governments, decentralization may increase corruption. Bardhan and Mookherjee (2000) point out the theoretical ambiguity of the importance of relative capture at the local and national levels. Huther and Shah (1998), Barenstein and de Mello (2001), and Fisman and Gatti (2002) find a negative relationship between fiscal decentralization and corruption for several countries.6 In contrast, Reinikka and Svensson (2004) highlight the capture of school grants by local officials in Uganda. At the macroeconomic level, Treisman (2000) and Fan, Lin, and Treisman (2009) conclude that federal states are more corrupt. Using data on 154 countries, Treisman (2000) also suggests that more tiers of government induce higher perceived corruption, less effective provision of public health services, and lower adult literacy, especially in developing countries. Prud’homme (1995) stresses several additional pitfalls of decentralization in developing countries, namely the increase in interjurisdictional disparities, the jeopardizing of macroeconomic stability, the ethnic bias of local elections, and weak capacities of local bureaucracies. Another supply-side argument against decentralization concerns the risk of diseconomies of scale, or at least a loss of scale economies. However, many of the public goods in question are community and site specific, and it is often possible to exclude nonresidents. Rural communities of poor countries, in particular, are often face to face, and social norms sharply distinguish “outsiders” from “insiders,” especially with respect to entitlement to community services (Bardhan 2002). Finally, decentralization is generally viewed as a trade-off between autonomy and accountability, between costs of coordination and better provision of public goods, and between preference matching and externalities. 5. Citizens can “vote with their feet,” that is, move to a nearby jurisdiction to obtain the public service tax package they prefer so that local governments compete to attract people and increase their tax bases. Even in the absence of population mobility, in the context of informational asymmetries between voters and politicians, voters can use the performance cues of other governments as benchmarks for judging whether or not their representative wastes resources and, consequently, whether or not he/she deserves to remain in office. Thus, an action chosen by a politician in one jurisdiction affects the informational set of imperfectly informed voters in other jurisdictions, forcing neighboring politicians to compete in order to avoid being signaled as bad incumbents, so that they might remain in office. 6. Fisman and Gatti (2002) use legal origin as an instrument for decentralization.

62

Emilie Caldeira, Martial Foucault, and Grégoire Rota-Graziosi

Besley and Coate (2003) and Lockwood (2002) confirm Oates’ insights by showing that the relative performance of centralized and decentralized provisions of public goods depends upon spillovers and differences in tastes for public spending between jurisdictions.7 To our knowledge, no attention has been paid to the consequences of decentralization on well-being conditions in French-speaking African countries. Our chapter fills this gap by focusing on Benin, where microdata (household surveys) and macrodata (local public finance) have been combined for the first time. 3.3 3.3.1

An Overview of Benin The Democratization Process

In regard to its democratization and decentralization processes, Benin belongs to the group of French-speaking African countries, which include twenty countries and around 243 million inhabitants (2009). Benin is also a lower-income country with an estimated per capita income of US$740 in 2011 and a ranking of 134 out of 169 countries in the 2010 Human Development Index. Benin’s population (8.93 million inhabitants in 2009) is fragmented into forty-two different ethnic groups, the most prominent being the Fon and the Adjas in the south, the Baribas and the Sombas in the north, and the Yorubas in the southeast. Since its independence on August 1, 1960, the political history of Benin has been chaotic. A succession of military governments ended in 1972, with the last military coup—led by Mathieu Kerekou—and the establishment of a government based on Marxist-Leninist principles. A move to democracy began in 1989. Two years later, as a result of free elections, the former prime minister, Nicephore Soglo, became president. Kerekou regained power in 1996. With the political support of the north of the country (Alibori, Atacora, Borgou, and Donga), he won subsequent elections in 2001. Having served two terms and being over seventy years old, he was ineligible to run in the presidential election of 2006. He was succeeded by Thomas Yayi Boni, an independent political outsider. In March 2007, President Yayi Boni strengthened his position after the legislative elections in which his coalition, Force Cauris pour un Benin Emergent (FCBE), won the largest number of seats (thirty-five out of eighty-three) and negotiated a progovernment majoritarian coalition in parliament with seven minor parties. With a strong electoral basis from the northern communes, he was reelected in 2011 with 7. Competition among jurisdictions to attract mobile capital is a way to discipline governments, motivating them to invest more in infrastructure, reduce waste and corruption, and spend less on nonproductive public goods. But Cai and Treisman (2005) emphasize that the required assumptions (perfect mobility, perfect local autonomy, etc.) are often unrealistic, and capital mobility may even weaken the discipline of the poorly endowed units.

Does Decentralization Facilitate Access to Poverty-Related Services?

63

the pivotal support of the southern part of the country (Atlantique, Collines, and Mono). 3.3.2

The Decentralization Process

The decentralization process in Benin began in 1998 through the transfer of several competencies to local Beninese jurisdictions, called communes. While an elected local government manages each commune, a representative of the central government is in charge of the départments to which the communes belong. Local elections were held in 2002 and 2007. Benin has seventy-seven communes in twelve départements.8 As in many Frenchspeaking African countries,9 the territorial shape of Beninese communes results from colonial history and not from any economic consideration with regard to efficiency in public goods provision. For instance, Tanguieta stretches out across more than 5,460 square kilometers for a population of 62,321 inhabitants in 2008 (11.4 inhabitants per square kilometer), while Akro-Misserete contains 98,961 inhabitants on only 79 square kilometers (1,252 inhabitants per square kilometer). In January 1999, Law No. 97-029 defined the competencies transferred from the central government to the seventy-seven communes. Their scope was large, ranging from elementary schooling to economic development, and including transport infrastructure, environmental programs, health services, social goods, tourism, security, and cultural activities. We may distinguish four kinds of competencies: exclusive local competencies, shared competencies, delegated competencies, and specific competencies. For delegated competencies, local jurisdictions act as representatives of the central state. Specific competencies concern some communes that have particular statuses (Cotonou, Porto-Novo, and Parakou). Table 3.1 summarizes these competencies. The distinction between shared and exclusive local competencies is largely subjective, linked to our interpretation of the relevant law and of observed practices in Benin. First, the transfer of competencies is obviously progressive and may take some time. For instance, the effective role of communes in water and sanitation is limited. The SONEB (Societé Nationale des Eaux du Benin) is a public enterprise still in charge of drinking water supply and sewage disposal in urban areas. A deconcentrated service, the General Direction of Water, remains essential in rural areas. Second, some competencies, such as primary education, require some technical and financial support from the central government. Usually, a transfer of competencies implies a transfer of financial resources. Table 3.2 presents Beninese communes’ revenues, distinguishing local own-revenue (tax and nontax) and other local revenue (central con8. Communes are themselves divided into 546 districts. 9. Burkina Faso counts 351 communes for 16.2 million inhabitants, while Mali has 703 communes for 15 million inhabitants.

64

Emilie Caldeira, Martial Foucault, and Grégoire Rota-Graziosi

Table 3.1

Beninese communes’ competencies

Type of competencies Exclusive local competencies Transport infrastructure: Maintenance of local roads, public lightings Shared competencies Hygiene conditions: Sewage and refuse disposal (latrines, septic tanks, etc.), drinking water Education: Construction and maintenance of public primary schools, adult literacy, cultural public infrastructures, sports, and leisure Delegated competencies Public records office, security, publication and application of laws Specific competencies Secondary schools, security, communication Source: Law No. 97-029 of Benin Republic, January 15, 1999.

ditional and unconditional grants, external transfers, loans, and advances), over the period 2006–2007. A local representative of the central tax administration (Directions Departementales des Impots) collects local taxes, mainly property and patent taxes.10 By contrast, local governments support collection costs of nontax own-revenue, essentially revenue related to occupations in the public domain (market stalls, parking tolls, kiosks, hoardings, etc.), as well as to some administrative services. Central conditional grants represent about 25 percent of local revenue with some huge disparities: less than 3 percent for Atlantic and more than 30 percent for Oueme. Unconditional transfer is another source of Beninese communes’ revenue. It corresponds to a retroceded tax, the road tax, collected by customs on exports (0.85 percent of the value of exported goods).11 Generated revenue is shared among communes following a fixed rule: 80 percent is allocated to three “special” communes (Cotonou, 60 percent; Porto-Novo, 24 percent; and Parakou, 16 percent). The rest is distributed among the seventy-four other communes according to their respective demographic weight. Beninese communes are characterized by a low average level of per capita revenue with about CFAF 2,200 (US$4.7) over the period studied (2006– 2007). Moreover, important inequalities exist among communes: the revenue per capita of the twenty poorest communes represents only 50 percent of the revenue of the five richest ones. Local governments’ revenues also differ in their composition. For instance, Parakou and Porto-Novo, despite having similar per capita revenues (around CFAF 6,500), have 50 and 35 percent of local own-revenue, respectively. Regarding our criteria of financial autonomy, Benin is characterized by 10. Beninese local governments can also tax mining, advertisements, and taxi drivers, and they have the opportunity to collect a tax on local development (see Chambas [2010] for a detailed analysis of local fiscal resources in sub-Saharan Africa, in particular in Benin). 11. The authorities abolished this tax in 2009 for transit goods being exported to landlocked countries such as Niger and Burkina Faso.

Does Decentralization Facilitate Access to Poverty-Related Services? Table 3.2

65

Average composition of Beninese communes’ per capita revenue (CFAF)

Total local revenue Own-revenue Local nontax own-revenue Local tax own-revenue Other local revenue Unconditional central grants Conditional central grants External transfers Loans and advances

Average level

Percentage of total resources

2,175 1,137 623 514 1,038 225 571 225 17

100 52 29 23 48 11 26 10 1

Source: Beninese Ministry of Finance and Economy.

strong geographical disparities (see figure 3.1) where south and northeast communes collect more own-revenue. 3.3.3

Local Public Goods Provision and Poverty

Since 1999, Benin has been involved in a national strategy aimed at reducing poverty for human sustainable development through its successive Growth and Poverty Reduction Strategy (GPRS). The main objective of the latest GPRS for the 2011–2015 period is the improvement of the living conditions of the population with specific attention to water, basic sanitation, primary education, and primary health care sectors in line with the Millennium Development Goals (MDGs). Poverty, which is a general state of deprivation, is multidimensional. It is usually associated with conditions under which people live. Poverty may be viewed in either absolute or relative terms. Absolute poverty is a situation in which a person or group of persons is unable to satisfy the most basic and elementary requirements of human survival in terms of good nutrition, sanitation, transport, health, education, and recreation. Several approaches exist that enable us to appreciate the level of poverty in Benin. Based on the monetary approach, the proportion of poor people in Benin in 2009 is estimated at 35.21 percent, which means that more than one of every three persons is living below the subsistence level (2011–2015 GPRS). An alternative approach involves looking at nonmonetary poverty, based on a composite index, including variables of household living conditions and property or assets. This measure stated that 30.84 percent of the Beninese population was poor in terms of subsistence and property in 2009. A geographical cleavage between rural and urban communes seems to matter. Indeed, urban communes located in the Littoral, Collines, and Oueme departments display a rate of poverty of 13, 17, and 19 percent respectively, namely two times less that the Beninese average. A final and crucial dimension of poverty in developing countries concerns the dynamic trend of poverty mobility. Availability of data does not allow for a robust

Fig. 3.1

Share of local own-revenue by commune (2006 and 2007)

Source: http://wwwgadm.org/country; authors’ calculations.

Fig. 3.1

(cont.)

68

Emilie Caldeira, Martial Foucault, and Grégoire Rota-Graziosi

discussion regarding the extent to which the implementation of national and local public policies has positively affected the reduction of poverty. The current situation remains ambiguous. For instance, between 2006 and 2007, income poverty fell by roughly 4 percentage points, versus 2.4 points in the case of nonincome poverty. On the other hand, between 2007 and 2009, income poverty rose by 1.9 percentage points. This increase in income poverty between 2007 and 2009 is the result of the effects of the economic and financial crises, which caused household consumer spending to fall. Nonincome poverty registered a substantial decline of 9 percentage points, falling from 39.6 percent in 2007 to 30.85 percent in 2009. This decline is the result of various actions taken by the government, during the period 2007– 2009, to improve access to basic social services. In particular, these actions involved the construction of water points and school infrastructure. Such policies have not only been implemented at the central level, but are also the responsibility of local governments when they have legal competencies for providing local poverty-related goods. In order to tackle the methodological problems resulting from an overly broad definition of poverty, we have chosen an approach that confines poverty-related issues to five main basic services: toilet facilities, drinking water, sewage, garbage, and primary education. In this way, we are able to assess the impact of decentralization on major dimensions of poverty issues. 3.4

Econometric Framework

In this section, we present our empirical strategy. We first test the average effect of decentralization on access to poverty-related services. We then assess its distributional effect between jurisdictions by distinguishing communes according to their wealth. 3.4.1

Data

We use several sources of information. The Beninese Ministry of Finances and Economy provided us with the communes’ accounts. The 2006 and 2007 Integrated Modular Surveys on Household Living Conditions (EMICoV) contain information concerning individual education level, household consumption and wealth, and access to several local public goods. They cover a sample of 18,000 Beninese households across the entire national territory. The sample includes 7,440 urban households and 10,560 rural households.12 The major uniqueness of these surveys lies in their representative character at the commune level, allowing us to measure aggregated and distributional indicators at the study level as described below. Data concerning population, urbanization rate, and ethnic fragmentation are drawn from the General Population and Housing Census in Benin (1992 and 2002) and seventy-seven 12. This sample is a stratified one, selected in two stages: stratification was achieved by separating every commune into urban and rural areas.

Does Decentralization Facilitate Access to Poverty-Related Services?

69

communes’ monographs provided by the European Union (Programme d’Appui au Démarrage des Communes). Testing the Average Effect of Decentralization on Access to Basic Services Our empirical analysis focuses on universal basic needs, setting aside any normative considerations in terms of welfare. It appears more relevant to study actual access to public services than ultimate effects on individual well-being, which may depend on many factors outside local governments’ control. We consider several basic services that have been assessed through the EMICoV surveys: toilet facilities, water access, refuse and sewage disposal, and primary education. Table 3.3 gives the detailed list of indicators, denoted by Yit , for each kind of service. These indicators are all measured at the household level except for education indicators, which require individual data (level of education of the respondent). To assess public services access we use two indicators: the first measures the share of households or individuals having access to the service (quantity) and the second reflects the qualitative scale of the service provided (quality). By using these two measures, we are able to capture, in a comprehensive way, how decentralization has or has not facilitated access to poverty-related services. Reasoning only on the quantity would be fallacious, as such a measure does not reveal to what extent local citizens have benefited from an improvement in the quality of local public goods. To better understand how quantitative and qualitative variables have been computed, let us describe the first indicator, namely toilet facilities. The Table 3.3

Indicators of basic services access

Basic services, Yit

Indicators

Toilet facilities

Share of households having access to a toilet or latrine facility, SToilit Type of toilet facility (no facility, bucket/pan, latrine with composting, suspension latrine, nonflagged pit latrine, nonventilated pit latrine, ventilated pit latrine, own flush toilet, flush toilet), TToilit

Water access

Share of households having access to drinking water, SWatit Source of drinking water (rainwater, rainwater in tanker truck, river, pond, protected spring, nonprotected well, protected well, borehole with manual pump, borehole with automatic pump, public tap, piped somewhere, piped into residence), TWatit

Refuse disposal

Share of households having access to refuse disposal facilities, SGarbit Type of refuse disposal (nature, courtyard, burning, burying, rubbish dump, collection truck [NGO], collection truck [public]), TGarbit

Sewage disposal

Share of households having access to sewage disposal facilities, SSewit Type of sewage disposal (nature, courtyard, well, grid/downstream, open pipe waste, covered pipe waste, draining), TSewit

Primary education

Primary school enrollment for children ages six to eleven, SEit

Source: EMICoV surveys, 2006 and 2007.

70

Emilie Caldeira, Martial Foucault, and Grégoire Rota-Graziosi

Table 3.4

Descriptive statistics

Basic services, Yit Degree of decentralization, Dit Public spending per capita, Git Per capita consumption, Cit Population size, Poit Population density, Deit Urbanization rate, Uit Ethnic fragmentation, Fit Partisan affiliation, Pait Toilet facility SToilit TToilit Water access SWatit TWatit Refuse disposal SGarbit TGarbit Sewage disposal SSewit TSewit Primary education, SEit

Mean

Std. dev.

Min.

Max.

0.484 7.218 11.754 11.356 338.084 0.278 0.357 0.305

0.235 0.778 0.464 0.529 1,050.57 0.232 0.232 0.461

0.066 4.521 10.513 10.250 7.382 0 0.013 0

0.986 9.436 12.970 13.500 9,235.63 1 0.822 1

0.237 2.836

0.208 1.541

0 1

0.306 7.214

0.249 1.060

0 4.748

0.033 1.255

0.102 0.466

0 1

0.684 3.785

0.009 1.138 0.876

0.026 0.156 0.149

0 1 0.236

0.184 1.785 1

0.969 7.958 0.994 10.559

EMICoV survey provides the share of households having access to a toilet. On average, 23.7 percent of Beninese households claim to have a toilet facility (table 3.4). The quality of the toilet measured by the scale in table 3.3 takes the value 1 for no facility to 9 for a flush toilet. Using responses from EMICoV respondents, we compute an average index at the commune level that indicates that households in only one commune (Toucountouna) have no toilet facilities, and 10 percent of the population has at least nonflagged pit latrines. As depicted in figure 3.2, only six communes out of seventy-seven converge toward the best quality of toilets with a score superior to 6, that is, those including either nonventilated pit latrines, ventilated pit latrines, or flush toilets. As local governments are in charge of sanitation facilities, decentralization should produce more efficient and equitable service delivery by making better use of knowledge of local needs. The same coding procedure applies for the four other indicators. Sewage and garbage facilities are respectively depicted in figures 3.3 and 3.4. We observe a small variance for every basic service among jurisdictions, with the exception of Segbana, which displays the highest level of sewage and garbage disposal and drinking water in the region (Alibori). One explanation for this discrepancy is linked to the development of hydraulic plans (DED and PADEAR-DANIDA projects). There exist sixty-seven drillings and fifty-four modern shafts that allow the center to be served by the water supply network (SONEB). Figure 3.5 illustrates the diversity of quality for sources of drinking water.

Fig. 3.2

Toilet quality

Source: http://www.gadm.org/country; authors’ calculations.

Fig. 3.3

Sewage quality

Source: http://www.gadm.org/country; authors’ calculations.

Fig. 3.4

Garbage quality

Source: http://www.gadm.org/country; authors’ calculations.

Fig. 3.5

Drinking water quality

Source: http://www.gadm.org/country; authors’ calculations.

Does Decentralization Facilitate Access to Poverty-Related Services?

75

In sum, combining consolidated household data on access to services, and the nature of locally provided public good quality to local public finance offers a new avenue for evaluating the impact of decentralization in both dimensions. 3.4.2

Empirical Models

The degree of decentralization, denoted by Dit, is the share of local ownrevenue in a given commune’s total revenue. This measure is used in the literature as an indicator of financial autonomy, and also allows us to approximate the accountability of local governments. Indeed, while central transfers are often opaque to taxpayers, who are then unable to judge the efficiency of local policies, the link between local taxes and local public services provided is more immediate and may provide an incentive for local officials to improve their efficiency. We add several control variables. Time dummies, denoted by tt, serve as controls for omitted explanatory variables that vary over time, but remain constant between communes, and can influence the share of local governments’ own-revenue. We also control for explanatory variables that may be correlated with the degree of decentralization, and that vary across both communes and time. Since we consider the effect of local revenues’ composition, and not the impact of local public spending itself, we introduce communes’ per capita public spending, denoted by Git. We are then able to see if a higher degree of decentralization affects the efficiency of local policies, given the level of local public spending. This control variable is essential because the communes’ public spending affects the level of received transfers, the measured degree of decentralization, and access to basic services.13 For similar reasons, we introduce per capita consumption, (measured by an index of about 1,200 commodities and services).14 Studying jurisdiction population size (Poit) and population density (Deit) allows us to capture, respectively, overrepresentation of smaller jurisdictions and some scale economies in the provision of public goods studied. We also consider urbanization rates, denoted by Uit, since urban areas generally offer better access to basic services and have higher fiscal capacities, especially in terms of the property tax base. Finally, ethnic fragmentation, denoted by Fit,15 may be correlated with the degree of decentralization and affects the provision of public goods in quantity and quality (Alesina and Ferrara 2005). Table 3.4 provides some descriptive statistics. The main independent vari13. Although the pursuit of an equitable allocation of resources would lead one to expect a pro-poor allocation of transfers across jurisdictions, most empirical studies (Wallis 1998; Meyer and Naka 1999; Alm and Boex 2002) find that wealthier local governments receive greater intergovernmental transfers, indicating that political considerations outweigh those of equity. 14. Provided by the EMICoV surveys. 15. Ethnic fragmentation in commune i on year t is defined as the probability that two individuals randomly drawn from one commune are from different ethnic groups.

76

Emilie Caldeira, Martial Foucault, and Grégoire Rota-Graziosi

able (degree of decentralization, Dit) is quite normally distributed with a median value and a mean of 0.48. Nevertheless, the mode of distribution indicates that most communes (around 22 percent) have collected about 15 to 22 percent of own-revenue. Conversely, only eight communes located in the southern part of the country perform very well in taxes, having raised more than 85 percent of own-revenue. Among control variables the average value of Cit is CFCA 142,598, the median is CFCA 123,042, and the ninety-fifth percentile is CFCA 299,798. To put these numbers in perspective, note that in strongly urbanized communes the average per capita consumption (CFCA 197,645) is higher than the national average due to the better situation of the first quintile, which entails a higher median value (CFCA 223,688). Another wealth measure is provided by Wit, a score based on the Demographic and Health Survey’s (DHS) wealth index, which provides each household’s position on an index of asset wealth at the national level using principal components analysis (PCA) weights.16 This variable fluctuates between –1.72 (poorest households) and 4.54 (wealthier households). Despite a significant correlation between Cit and Wit, the latter is a good proxy for permanent wealth, whereas the former is more sensitive to the economic conjuncture. Finally, ethnic fragmentation, measured by probability that two randomly selected individuals belong to the same ethnicity, indicates that the Beninese are strongly fragmented with an average value of 0.36. Such a cultural pattern is expected to affect preferences for public goods provision in the sense that ethnically heterogeneous communities may express contrasted needs or define different priorities for basic services delivery. We start with the following simplest regression, which assesses the average impact of decentralization on access to basic services:17 (1)

Yit = bDit + uGit + gCit + rPoit + τDeit + vUit + fFit + tt + ´it.

We also consider a nonmonotone effect of the degree of decentralization by introducing its quadratic term (Dit2): (2) Yit = b1Dit + b2Dit2 + uGit + gCit + rPoit + τDeit + vUit + fFit + tt + ´it. A Heterogeneous Effect between Communes In addition to the average impact of decentralization on access to public services, we study its effect by distinguishing communes by their respective wealth. This analysis allows us to assess the overall impact of decentralization on intercommune inequalities in terms of access to basic services. We obtain the following regression: 16. The general methodology used to calculate the wealth index is given in Filmer and Pritchett (2001). The specific approach used in the DHS is described in Rutstein and Johnson (2004). 17. Population, per capita public spending, and per capita consumption are given in logarithmic terms.

Does Decentralization Facilitate Access to Poverty-Related Services?

(3)

77

Yit = b1(Dit *QPit ) + b2(Dit *(1 − QPit )) + fQPit + uGit + gCit + rPoit + τDeit + vUit + ϕFit + tt + ´it,

where QPit is a dummy variable taking the value 1 if the commune i belongs to the first quintile of poor communes and zero otherwise. Following Filmer and Pritchett (2001) we define an asset-based measure of wealth, denoted by Wit for each commune using the EMICoV.18 However, the DHS index underestimates the wealth of rural areas since urban populations own many valuable assets. Following Rutstein (2008), we compute a national-level composite index from wealth indexes that have been separately constructed for urban and rural areas. We then consider the average score by communes and divide the latter into quintiles to distinguish the poor from the nonpoor. The same procedure applies for the first quintile of wealthier communes in order to control how decentralization may lead local governments to define priorities in delivering poverty-related services. In so doing, we assume that all local public goods are not provided according to the same economic and political determinants, that is, poor communes are expected to make greater efforts in facilitating access to drinking water than in organizing high-quality systems of waste disposal or sewage facilities. Econometric Issues and Identification Strategy Given the small number of time series with respect to cross-sectional observations and the fact that some variables have little within variance, we first estimate pooled ordinary least squares (OLS) regressions with year dummies. This estimation method increases the degree of freedom and allows for inquiry into variables that have low variability. However, it assumes that control variables capture all relevant communes’ characteristics. This estimate may be biased by unobserved heterogeneity between communes. Our panel data allows for controlling a large number of unobserved explanatory variables by using the fixed effects (FE) estimator. However, the traditional FE method fails in estimating the effects of variables that have little within variance, a problem worth considering when analyzing two successive years of observations. To assess coefficients of time-invariant variables and to control for commune-specific effects, we use the Fixed Effects Vector Decomposition estimator (FEVD) developed by Plümper 18. Due to the abundance of household survey data on asset ownership and the considerable bias measurement error associated with reported income or consumption, a substantial body of literature has developed an asset-based measure of wealth. Filmer and Pritchett (2001) conclude that the DHS wealth index actually performed better than the traditional consumption or expenditure index in explaining differences in economic statuses. From the EMICoV, we built such a DHS measure based on a myriad of assets (cars, canoes, hi-fi systems, refrigerators, iron, beds, phones, motorcycles/scooters, radios, VCRs, DVD players, ovens/stoves, washing machines, chairs, sewing machines, cell phones, bicycles, televisions, video recorders/ VCRs, fans, foam mattresses, computers, Internet access, land, home ownership, types of fuel, building materials, etc.).

78

Emilie Caldeira, Martial Foucault, and Grégoire Rota-Graziosi

and Troeger (2007).19 Through a three-step procedure, this estimator allows for a decomposition of the unit fixed effect into two parts: an explained part by time-invariant variables and an unexplained part.20 To correct for other potential endogeneity biases in the estimation of the causal effect of decentralization on access to basic services, we instrument the degree of decentralization using a dummy variable, denoted by PAit, taking the value 1 if the commune i has the same political affiliation as the president in office. This dummy variable differs between 2006 and 2007 since Yayi Boni was elected in April 2006, succeeding Mathieu Kerekou. Partisan affiliation is a good instrument of decentralization in a regression involving access to public services. In the relevant literature, a jurisdiction that has greater political support from the central government receives more transfers from the latter (see Cox and McCubbins [1986] for a theoretical argument, Case [2001] for the Albanian case, and Miguel and Zaidi [2003] for the Ghanaian case). 3.5

Estimation Results

This section presents our empirical results using panel data from seventyseven Beninese communes for 2006 and 2007. 3.5.1

The Average Effect of Decentralization on Access to Basic Services

Figures 3.6, 3.7, 3.8, 3.9, and 3.10 confirm our expectations that a higher degree of decentralization is positively correlated to better access to povertyrelated services.21 However, the most decentralized communes are the richest, most populated, and most urbanized (table 3.5). These variables are also associated with easier access to basic public services (table 3.6). This confirms the important role of our control variables in avoiding endogeneity bias. To test the average effect of decentralization on access to basic services (equation [1]), we first run the pooled OLS regressions with year dummies, introducing our control variables progressively (columns [1] to [7]). Considering potential unobserved heterogeneity between communes, we then use the FEVD estimator (column [8]). Finally, we instrument for the degree of 19. Based on Monte Carlo simulations, Plümper and Troeger (2007) compare the vector decomposition model with the FE model, the random effects (RE) model, pooled OLS, and the Hausman-Taylor procedure and find that, while the FE model does not compute coefficients for the time-invariant variables, the vector decomposition model performs far better than other procedures. 20. First, the unit fixed effect is estimated by running a fixed effect estimate of the model. Second, the latter is split into its two parts by regressing it on the time-invariant variables of the model. The unexplained part corresponds to the residuals of this equation, h!i . Third, the estimation of the full model is implemented by including the time-invariant variables and the unexplained part of the fixed effect vector estimated in the second step. 21. The relation is relatively weak for primary school enrollment (figure 3.10).

79

Does Decentralization Facilitate Access to Poverty-Related Services? Table 3.5

Correlations of our key variables

Variables

Dit

Git

Cit

Poit

Deit

Uit

Fit

1 0.3294* 0.3128* 0.3095* 0.2431* 0.2513* 0.0258

1 0.5646* 0.5025* 0.4656* 0.4117* 0.2696*

1 0.5801* 0.7571* 0.5505* 0.0817

1 0.8080* 0.5379* 0.2895*

1 0.4089* 0.0153

1 0.3330*

1

Degree of decentralization, Dit Public spending per capita, Git Per capita consumption, Cit Population size, Poit Population density, Deit Urbanization rate, Uit Ethnic fragmentation, Fit

*Correlation coefficient significant at the 10 percent level.

Table 3.6

Correlations of our key variables

Variables Type of toilet facility, TToilit Source of drinking water, TWatit Type of sewage disposal, TSewit Type refuse disposal, SGarbit Primary school enrollment, SEit

Git

Cit

Poit

Deit

Uit

Fit

0.5155* 0.5221* 0.3826* 0.2321* 0.2286*

0.2760* 0.3493* 0.1831* 0.2987* –0.0638

0.4030* 0.3555* 0.4420* 0.5511* 0.0461

0.4274* 0.3902* 0.4618* 0.6045* 0.1058

0.4108* 0.2823* 0.3018* 0.3771* 0.0962

0.0018 0.1693* 0.0911 0.0533 –0.1412

*Correlation coefficient significant at the 10 percent level.

decentralization with the partisan affiliation PAit in column (9). Table 3.7 reports the relevance of our instrument.22 Moreover, the Sargan overidentifying restriction test23 indicates that we cannot reject the hypothesis that there is no correlation between the instrument and the error term in the regression stating that the partisan affiliation variable is a valid instrument. In tables 3.8, 3.9, 3.10, and 3.11, we highlight the fact that a higher degree of decentralization is consistently associated with improved access to water sources and sanitation systems. Table 3.8 indicates that the coefficient associated with the degree of decentralization is significantly different from zero and could be interpreted as follows: the impact of a 10 percentage point increase in decentralization represents an extra 3.7 percent in people having access to toilets or latrines. In other words, the standard deviation of the degree of decentralization (23.5 percent) implies an 8.69 percentage point increase for one standard deviation change. When we turn our attention to the quality of basic services, for instance, we find that once controlled for endogeneity bias, such services undergo a 10 percentage point increase 22. As in most empirical studies, political considerations outweigh those of equity: wealthier, smaller, or ethnically fragmented jurisdictions receive more intergovernmental transfers and are less autonomous. 23. We use the dummy variable indicating whether a commune has the same dominant ethnic affiliation as the president in office as another instrument to compute the Sargan test.

Fig. 3.6

Share of local own-resources and access to toilet facility

Fig. 3.7

Share of local own-resources and access to water

Fig. 3.8

Share of local own-resources and access to sewage disposal

Fig. 3.9

Share of local own-resources and access to refuse disposal

Fig. 3.10 Table 3.7

Share of local own-resources and access to primary school enrollment Validity of our instrumental variable Dependent variable: Dit Partisan affiliation, PAit Public spending per capita, Git Per capita consumption, Cit Population size, Pit Population density, Deit Urbanization rate, Uit Ethnic fragmentation, Fit Constant Number of observations Adjusted R2 F-statistic Fixed effect Year dummies Sargan test (p -value)

–0.026*** –0.027*** –0.053*** 0.130*** 0.002*** 0.155*** –0.104*** –0.182

(0.000) (0.002) (0.004) (0.029) (0.000) (0.017) (0.029) (0.31)

145 0.68 54,680 Yes Yes 0.519

Note: Controls for serial correlation of the error term, AR(1) Coccrane-Orcutt transformation. Robust standard errors are in brackets. ***Coefficient significant at the 1 percent level. **Coefficient significant at the 5 percent level. *Coefficient significant at the 10 percent level.

150 0.17 20.55 no yes

Number of observations Adjusted R2 F-statistic Fixed effect Year dummies

149 0.35 21.86 no yes

–0.779*** (0.22) 149 0.41 20.71 no yes

–2.165*** (0.46)

0.184*** (0.06) 0.101*** (0.03) 0.133** (0.04)

(3)

149 0.46 25.54 no yes

–2.893*** (0.52)

0.116** (0.05) 0.109*** (0.03) 0.094*** (0.03) 0.102*** (0.02)

(4)

149 0.49 34.43 no yes

–2.253*** (0.71)

0.129** (0.05) 0.090*** (0.03) 0.086** (0.03) 0.064* (0.03) 0.003** (0.001)

(5)

149 0.96 30.62 no yes

–1.641*** (0.58)

0.140** (0.05) 0.066* (0.03) 0.086** (0.03) 0.021 (0.03) 0.003** (0.01) 0.191*** (0.08)

(6)

145 0.54 46.74 no yes

0.092* (0.08) 0.071** (0.03) 0.099*** (0.03) 0.049 (0.03) 0.002** (0.001) 0.234*** (0.08) 0.188*** (0.05) –2.090*** (0.65)

(7)

(8)

145 0.92 715.01 yes yes

0.171*** (0.025) 0.030*** (0.004) 0.001 (0.01) 0.021 (0.01) 0.005*** (0.00) 0.301*** (0.02) 0.131*** (0.04) –0.397* (0.20)

Note: Controls for serial correlation of the error term, AR(1) Coccrane-Orcutt transformation. Robust standard errors are in brackets. ***Coefficient significant at the 1 percent level. **Coefficient significant at the 5 percent level. *Coefficient significant at the 10 percent level.

0.049 (0.03)

0.249*** (0.06) 0.126*** (0.03)

0.370*** (0.06)

Constant

Ethnic fragmentation, Fit

Urbanization rate, Uit

Population density, Deit

Population size, Poit

Per capita consumption, Cit

Public spending per capita, G

(2)

(1)

Estimation results: The average effect of decentralization on the access to toilet facility (quantity)

Degree of decentralization, Dit

Dep. var.: SToilit

Table 3.8

145 0.96 622.75 yes yes

0.715*** (0.01) 0.009 (0.01) –0.314* (0.21) –0.057*** (0.01) 0.004*** (0.001) 0.242*** (0.02) –0.052* (0.04) –11.72*** (1.06)

(9)

150 0.15 11.88 no yes

Number of observations Adjusted R2 F-statistic Fixed effect Year dummies

149 0.26 12.61 no yes

–3.202** (1.27) 149 0.26 29.43 no yes

–3.940 (3.94)

1.871*** (0.56) 0.705*** (0.20) 0.071 (0.36)

(3)

149 0.46 25.54 no yes

–7.483 (4.97)

1.537*** (0.56) 0.749*** (0.20) –0.119 (0.34) 0.498* (0.25)

(4)

149 0.32 57.82 no yes

–1.405 (5.54)

1.660*** (0.54) 0.565*** (0.21) –0.197 (0.34) 0.142 (0.30) 0.003*** (0.001)

(5)

149 0.35 49.10 no yes

–2.706 (5.28)

1.734*** (0.55) 0.402* (0.22) 0.199 (0.35) –0.150 (0.27) 0.003*** (0.001) 1.284* (0.68)

(6)

145 0.34 64.70 no yes

1.553*** (0.56) 0.464** (0.21) –0.249 (0.37) –0.345 (0.31) 0.004*** (0.001) 1.123* (0.73) 0.031 (0.51) –5.178 (5.92)

(7) 1.471*** (0.027) 0.273*** (0.019) 0.189*** (0.006) –0.411*** (0.008) 0.004*** (0.00) 1.298*** (0.01) –0.229 (0.39) –0.397* (0.20)

(8)

145 0.89 10,608.02 yes yes

Note: Controls for serial correlation of the error term, AR(1) Coccrane-Orcutt transformation. Robust standard errors are in brackets. ***Coefficient significant at the 1 percent level. **Coefficient significant at the 5 percent level. *Coefficient significant a the 10 percent level.

1.542*** (0.25)

1.905*** (0.53) 0.718*** (0.19)

2.531*** (0.54)

Constant

Ethnic fragmentation, Fit

Urbanization rate, Uit

Population density, Deit

Population size, Poit

Per capita consumption, Cit

Public spending per capita, Git

(2)

(1)

Estimation results: The average effect of decentralization on the access to toilet facility (quality)

Degree of decentralization, Dit

Dep. var.: TToilit

Table 3.9

145 0.87 992.44 yes yes

0.692*** (0.02) 0.352*** (0.02) 0.189*** (0.007) –0.260*** (0.008) 0.003*** (0.00) 1.264*** (0.01) –0.342 (0.41) 0.043 (0.11)

(9)

150 0.12 13.51 no yes

Number of observations Adjusted R2 F-statistic Fixed effect Year dummies

149 0.24 15.49 no yes

–0.660** (0.19) 149 0.25 13.44 no yes

–1.505*** (0.47)

0.220*** (0.08) 0.101*** (0.02) 0.081** (0.04)

(3)

149 0.26 12.62 no yes

–1.801*** (0.51)

0.192*** (0.07) 0.105*** (0.02) 0.065 (0.04) 0.041 (0.04)

(4)

149 0.29 30.13 no yes

–0.836 (0.62)

0.212*** (0.07) 0.076*** (0.03) 0.053 (0.04) –0.148 (0.04) 0.005*** (0.001)

(5)

149 0.30 25.31 no yes

–1.080* (0.65)

0.207** (0.08) 0.085*** (0.03) 0.053 (0.04) 0.002 (0.04) 0.005*** (0.001) –0.076 (0.10)

(6)

145 0.34 64.70 no yes

0.173** (0.08) 0.087*** (0.03) 0.072 (0.04) 0.022 (0.05) 0.004*** (0.001) 0.058 (0.11) –0.145* (0.08) –1.480 ** (0.71)

(7) 0.230*** (0.028) 0.047*** (0.003) 0.021 (0.014) –0.032 (0.019) 0.006*** (0.00) 0.002 (0.003) –0.115*** (0.01) –0.378* (0.21)

(8)

145 0.88 1,537.87 yes yes

Note: Controls for serial correlation of the error term, AR(1) Coccrane-Orcutt transformation. Robust standard errors are in brackets. ***Coefficient significant at the 1 percent level. **Coefficient significant at the 5 percent level. *Coefficient significant at the 10 percent level.

0.110*** (0.04)

0.260*** (0.07) 0.116*** (0.02)

0.365*** (0.07)

Constant

Ethnic fragmentation, Fit

Urbanization rate, Uit

Population density, Deit

Population size, Poit

Per capita consumption, Cit

Public spending per capita, Git

(2)

(1)

Estimation results: The average effect of decentralization on water access (quantity)

Degree of decentralization, Dit

Dep. var.: SWatit

Table 3.10

145 0.87 992.44 yes yes

0.654*** (0.01) 0.047*** (0.003) 0.021 (0.15) –0.064** (0.018) 0.006*** (0.00) 1.264*** (0.01) –0.053*** (0.012) 0.110 (0.20)

(9)

86

Emilie Caldeira, Martial Foucault, and Grégoire Rota-Graziosi

in the share of own-revenue, entailing an extra 0.236 point on the quality index of water access in communes (table 3.11). However, while the effect of decentralization on access to refuse disposal facilities is less robust (tables 3.12 and 3.13), decentralization is not found to have a significant average effect on access to sewage disposal facilities and communes’ primary school enrollment (tables 3.14, 3.15, and 3.16). In table 3.17, we consider the nonmonotone effect of the degree of decentralization by introducing its quadratic term (equation [2]). We find a positive coefficient associated with the degree of decentralization and a negative sign for its squared value. The impact of decentralization is then nonmonotone: the relationship between decentralization and access to basic services may be described by an inverted U-shaped curve. Even if we cannot calculate the average optimal decentralization degree due to a combination of different scaled criteria for basic services, we are able to determine it individually. Defined as the ratio of local own-revenue over total revenue (given by −b1 / 2b2, equation [2]), the optimal degree of decentralization reaches a 55 percent value for access to toilet facilities, 65 percent for refuse disposal facilities (columns [1] to [3]), and a lower value for sewage disposal facilities (49 percent) and primary school enrollment (52 percent, columns [4] and [5]). We observe that the effect of decentralization is monotone for drinking water access since the optimal level is above 1 (exactly 103 percent). 3.5.2

The Nonlinear Effect of Decentralization between Communes

We now consider the heterogeneous effects of decentralization across communes according to their wealth (equation [3]). Table 3.18 reports that this effect is generally lower for 20 percent of the poorest communes. While toilet, garbage, and drinking water facilities are increasing in quality with decentralization, there is no impact on average on sewage and primary school enrollment. For the latter, it actually has a positive effect on wealthier communes and a negative one on the poorest communes.24 As a robustness check, we interact a continuous variable, the DHS wealth index scores Wit, with the degree of decentralization (see table 3.19). Estimated results confirm that the positive effect of decentralization is contingent on a minimum wealth in communes. Only the effect of decentralization on access to drinking water does not seem to depend on wealth. The coefficient associated with the degree of decentralization measures the impact of decentralization in the absence of any wealth. Its negative sign indicates that a commune with zero wealth would suffer from decentralization. Finally, we highlight the point that communes may prioritize basic services despite a uniform decentralization process. Such a hierarchy results from their autonomy, and should also be considered by the central government 24. We complete our analysis with Wald tests to ascertain that coefficients for poor communes are significantly different from those in other communes.

150 0.18 17.44 no yes

Number of observations Adjusted R2 F-statistic Fixed effect Year dummies

149 0.24 14.71 no yes

4.186*** (0.83) 149 0.26 11.33 no yes

0.021 (0.60)

1.490*** (0.35) 0.250** (0.12) 0.402* (0.21)

(3)

149 0.27 10.42 no yes

–1.363 (3.09)

1.360*** (0.37) 0.268** (0.12) 0.327 (0.22) 0.194 (0.20)

(4)

149 0.30 51.08 no yes

2.649 (3.29)

1.441*** (035) 0.147 (0.12) 0.276 (0.22) –0.040 (0.21) 0.002*** (0.001)

(5)

149 0.30 45.16 no yes

3.382 (3.40)

1.454*** (0.36) 0.117 (0.14) 0.275 (0.23) –0.092 (0.21) 0.002*** (0.001) 0.228 (0.44)

(6)

145 0.32 82.58 no yes

1.591*** (0.37) 0.133 (0.13) 0.143 (0.24) –0.340 (0.24) 0.003*** (0.001) 0.001 (0.44) –0.934*** (0.34) 7.297* (3.84)

(7) 1.714*** (0.22) –0.028 (0.034) 0.185* (0.11) –0.434*** (0.10) 0.003*** (0.00) 0.216* (0.13) –0.933*** (0.02) 8.872*** (1.29)

(8)

145 0.86 142.652 yes yes

Note: Controls for serial correlation of the error term, AR(1) Coccrane-Orcutt transformation. Robust standard errors are in brackets. ***Coefficient significant at the 1 percent level. **Coefficient significant at the 5 percent level. *Coefficient significant the 10 percent level.

6.364*** (0.18)

1.685*** (0.34) 0.327*** (0.12)

1.935*** (0.32)

Constant

Ethnic fragmentation, Fit

Urbanization rate, Uit

Population density, Deit

Population size, Poit

Per capita consumption, Cit

Public spending per capita, Git

(2)

(1)

Estimation results: The average effect of decentralization on water access (quality)

Degree of decentralization, Dit

Dep. var.: TWatit

Table 3.11

145 0.84 188.52 yes yes

2.361*** (0.27) –0.028 (0.037) 0.185* (0.10) –0.528*** (0.10) 0.003*** (0.00) 0.147 (0.14) 1.028*** (0.03) 9.616*** (1.33)

(9)

150 0.10 4.26 no yes

Number of observations Adjusted R2 F-statistic Fixed effect Year dummies

149 0.21 4.26 no yes

–0.357** (0.14) 149 0.22 3.30 no yes

–0.595* (0.37)

0.081** (0.03) 0.044** (0.01) 0.022 (0.02)

(3)

149 0.40 4.50 no yes

–1.258*** (0.007)

0.018 (0.04) 0.053*** (0.01) –0.012 (0.02) 0.019*** (0.02)

(4)

149 0.55 67.65 no yes

–0.424 (0.36)

0.035 (0.03) 0.027** (0.01) –0.023 (0.02) 0.044** (0.01) 0.005*** (0.001)

(5)

149 0.57 56.65 no yes

–0.424 (0.36)

0.040 (0.03) 0.017* (0.01) –0.023 (0.02) 0.025** (0.01) 0.005*** (0.001) 0.084** (0.03)

(6)

145 0.58 79.65 no yes

0.048 (0.03) 0.019* (0.01) –0.036 (0.02) 0.002 (0.01) 0.005*** (0.001) 0.065* (0.03) 0.074*** (0.02) 0.205 (0.29)

(7)

Note: Controls for serial correlation of the error term, AR(1) Coccrane-Orcutt transformation. Robust standard errors are in brackets. ***Coefficient significant at the 1 percent level. **Coefficient significant at the 5 percent level. *Coefficient significant at the 10 percent level.

–0.033** (0.01)

0.092*** (0.03) 0.049** (0.02)

0.137*** (0.04)

Constant

Ethnic fragmentation, Fit

Urbanization rate, Uit

Population size, Poit Population density, Deit

Per capita consumption, Cit

Public spending per capita, Git

(2)

(1)

Estimation results: The average effect of decentralization on the access to refuse disposal facility (quantity)

Degree of decentralization, Dit

Dep. var.: SGarbit

Table 3.12.

145 0.96 89.10 yes yes

0.038*** (0.007) 0.013*** (0.001) –0.006* (0.003) 0.001 (0.003) 0.005*** (0.000) 0.069*** (0.06) 0.060*** (0.006) –0.081** (0.03)

(8)

145 0.95 79.30 yes yes

0.077*** (0.01) –0.024*** (0.001) –0.006 (0.004) –0.012** (0.005) 0.005*** (0.000) 0.089*** (0.009) 0.065*** (0.001) 0.161*** (0.05)

(9)

150 0.07 4.83 no yes

Number of observations Adjusted R2 F-statistic Fixed effect Year dummies

149 0.18 4.41 no yes

–0.466 (0.56) 149 0.19 3.40 no yes

–1.590 (1.58)

0.260 (0.17) 0.198*** (0.01) 0.108* (0.09)

(3)

149 0.25 3.90 no yes

–3.308* (2.05)

0.098 (0.19) 0.219*** (0.07) 0.016 (0.11) 0.241* (0.12)

(4)

149 0.41 68.32 no yes

0.509 (1.72)

0.176 (0.17) 0.104* (0.06) –0.032 (0.11) 0.018 (0.10) 0.002*** (0.000)

(5)

149 0.41 56.66 no yes

1.249 (1.30)

0.189 (0.17) 0.074* (0.05) –0.033 (0.11) –0.034 (0.08) 0.002*** (0.000) 0.023** (0.21)

(6)

145 0.42 58.22 no yes

0.176 (0.17) 0.089* (0.05) –0.067 (0.11) –0.122 (0.08) 0.002*** (0.000) 0.154 (0.21) 0.163 (0.11) 2.499* (1.37)

(7)

(8)

145 0.83 119.60 yes yes

0.225*** (0.05) 0.044** (0.02) –0.122*** (0.03) –0.144* (0.08) 0.002*** (0.000) 0.213*** (0.03) 0.196*** (0.01) 3.568*** (0.92)

Note: Controls for serial correlation of the error term, AR(1) Coccrane-Orcutt transformation. Robust standard errors are in brackets. ***Coefficient significant at the 1 percent level. **Coefficient significant at the 5 percent level. *Coefficient significant at the 10 percent level.

0.973*** (0.07)

0.313** (0.15) 0.218*** (0.08)

0.510*** (0.18)

Constant

Ethnic fragmentation, Fit

Urbanization rate, Uit

Population density, Deit

Population size, Poit

Per capita consumption, Cit

Public spending per capita, Git

(2)

(1)

Estimation results: The average effect of decentralization on the access to refuse disposal facility (quality)

Degree of decentralization, Dit

Dep. var.: TGarbit

Table 3.13

145 0.81 1,317.32 yes yes

1.345*** (0.12) –0.078*** (0.008) –0.122** (0.05) –0.365*** (0.11) 0.003*** (0.000) 0.276*** (0.02) 0.359*** (0.01) 6.354*** (1.30)

(9)

150 0.05 4.00 no yes

Number of observations Adjusted R2 F-statistic Fixed effect Year dummies

149 0.17 3.17 no yes

–0.852** (0.38) 149 0.18 3.38 no yes

–1.675* (0.001)

0.105 (0.07) 0.110** (0.04) 0.078 (0.06)

(3)

149 0.37 3.25 no yes

–3.444*** (1.29)

–0.061 (0.10) 0.132*** (0.04) –0.016 (0.05) 0.248*** (0.07)

(4)

149 0.52 21.06 no yes

–1.34 (1.21)

–0.018 (0.09) 0.068* (0.04) –0.043 (0.05) 0.125** (0.05) 0.001*** (0.000)

(5)

149 0.55 18.75 no yes

–0.587 (0.85)

–0.005 (0.09) 0.038* (0.002) –0.044 (0.05) 0.072* (0.03) 0.001*** (0.000) 0.235* (0.13)

(6)

145 0.57 19.07 no yes

0.037 (0.08) 0.034 (0.02) –0.061 (0.05) 0.054 (0.04) 0.001*** (0.000) 0.218* (0.13) 0.157** (0.07) –0.221 (0.98)

(7) 0.013 (0.009) 0.020*** (0.002) 0.006 (0.008) 0.051*** (0.01) 0.001*** (0.000) 0.226*** (0.05) 0.124** (0.05) –0.897*** (0.17)

(8)

145 0.91 235.90 yes yes

Note: Controls for serial correlation of the error term, AR(1) Coccrane-Orcutt transformation. Robust standard errors are in brackets. ***Coefficient significant at the 1 percent level **Coefficient significant at the 5 percent level. *Coefficient significant at the 10 percent level.

–0.038 (0.03)

0.143** (0.06) 0.125** (0.05)

0.258*** (0.09)

Constant

Ethnic fragmentation, Fit

Urbanization rate, Uit

Population density, Deit

Population size, Poit

Per capita consumption, Cit

Public spending per capita, Git

(2)

(1)

Estimation results: The average effect of decentralization on the access to sewage disposal (quantity)

Degree of decentralization, Dit

Dep. var.: SSewit

Table 3.14

145 0.91 350.65 yes yes

0.024*** (0.002) 0.009*** (0.001) –0.001 (0.001) 0.004*** (0.000) 0.001*** (0.000) 0.034*** (0.01) 0.012 (0.008) –0.069*** (0.003)

(9)

150 0.06 6.22 no yes

Number of observations Adjusted R2 F-statistic Fixed effect Year dummies

149 0.10 4.61 no yes

7.909*** (1.83) 149 0.10 3.47 no yes

5.93 (5.07)

1.008* (0.61) 0.352 (0.25) 0.190 (0.41)

(3)

149 0.18 3.68 no yes

–0.634 (6.16)

0.390 (0.69) 0.433* (0.24) –0.163 (0.38) 0.924** (0.36)

(4)

149 0.26 47.77 no yes

8.682 (6.19)

0.578 (0.66) 0.152 (0.25) –0.282 (0.38) 0.378 (0.34) 0.005*** (0.000)

(5)

149 0.26 40.32 no yes

10.643** (4.92)

0.613 (0.65) 0.074 (0.22) –0.283 (0.39) 0.238 (0.30) 0.005*** (0.000) 0.312* (0.79)

(6)

145 0.28 49.20 no yes

0.680 (0.67) 0.079 (0.22) –0.357 (0.40) 0.254 (0.34) 0.005*** (0.000) 0.656 (0.79) 0.274 (0.55) 11.138* (5.66)

(7)

Note: Controls for serial correlation of the error term, AR(1) Coccrane-Orcutt transformation. Robust standard errors are in brackets. ***Coefficient significant at the 1 percent level. **Coefficient significant at the 5 percent level. *Coefficient significant at the 10 percent level.

–1.48*** (0.23)

1.101** (0.51) 0.388 (0.26)

1.412** (0.57)

Constant

Ethnic fragmentation, Fit

Urbanization rate, Uit

Population density, Deit

Population size, Poit

Per capita consumption, Cit

Public spending per capita, Git

(2)

(1)

Estimation results: The average effect of decentralization on the access to sewage disposal (quality)

Degree of decentralization, Dit

Dep. var.: TSewit

Table 3.15

145 0.64 35.87 yes yes

0.142 (0.35) 0.049** (0.02) 0.595** (0.06) 0.348 (0.36) 0.004*** (0.000) 0.503* (0.29) –0.204 (0.33) –1.089 (4.13)

(8)

145 0.84 140.55 yes

–2.81*** (0.306) 0.049*** (0.01) 0.595*** (0.06) 0.778** (0.38) 0.004*** (0.000) 0.819*** (0.30) –0.637* (0.37) –4.494 (0.003)

(9)

150 0.35 43.21 no yes

Number of observations Adjusted R2 F-statistic Fixed effect Year dummies

149 0.37 30.10 no yes

0.573*** (0.11) 149 0.10 3.47 no yes

0.446* (0.22)

–0.016 (0.04) 0.028* (0.09) 0.122 (0.02)

(3)

149 0.37 20.60 no yes

0.464** (0.25)

–0.015 (0.04) 0.028* (0.01) 0.013 (0.02) –0.002 (0.01)

(4)

149 0.38 16.95 no yes

0.652** (0.35)

–0.011 (0.04) 0.022 (0.01) 0.010 (0.02) –0.013 (0.02) 0.001*** (0.000)

(5)

149 0.38 14.45 no yes

0.776* (0.40)

–0.009 (0.04) 0.017 (0.01) 0.010 (0.02) –0.022 (0.02) 0.001*** (0.000) 0.038* (0.04)

(6)

145 0.41 12.77 no yes

–0.039 (0.04) 0.020 (0.01) 0.023 (0.02) 0.001 (0.03) 0.002*** (0.000) 0.063 (0.05) –0.131** (0.06) 0.3898* (0.45)

(7)

Note: Controls for serial correlation of the error term, AR(1) Coccrane-Orcutt transformation. Robust standard errors are in brackets. ***Coefficient significant at the 1 percent level. **Coefficient significant at the 5 percent level. *Coefficient significant at the 10 percent level.

0.777*** (0.02)

–0.010 (0.04) 0.031** (0.01)

0.019 (0.04)

Constant

Ethnic fragmentation, Fit

Urbanization rate, Uit

Population density, Deit

Population size, Poit

Per capita consumption, Cit

Public spending per capita, Git

(2)

(1)

Estimation results: The average effect of decentralization on the access to primary education

Degree of decentralization, Dit

Dep. var.: SEit

Table 3.16

145 0.58 17.26 yes yes

0.003 (0.03) 0.063** (0.02) 0.134*** (0.04) 0.013 (0.01) 0.001*** (0.000) 0.030*** (0.009) –0.056 (0.08) 1.809*** (0.26)

(8)

145 0.52 14.19 yes yes

0.242** (0.11) 0.031* (0.01) 0.072** (0.02) –0.036*** (0.007) 0.001*** (0.000) 0.040*** (0.01) –0.049 (0.08) 1.765*** (0.32)

(9)

Does Decentralization Facilitate Access to Poverty-Related Services? Table 3.17

93

Estimation results: A nonmonotone effect of decentralization on the access to basic services

Dep. var.:

(1) TToilit

(2) TWatit

(3) TGarbit

(4) TSewit

(5) SEit

Degree of decentralization, Dit

2.190*** (0.05) –1.96*** (0.07) 0.030*** (0.004) –0.008 (0.01) –0.002 (0.01) 0.005*** (0.000) 0.359*** (0.02) –0.132** (0.05) –0.552** (0.22)

3.234*** (0.34) –1.56*** (0.47) –0.026 (0.03) 0.182* (0.12) –0.440*** (0.11) 0.003*** (0.000) 0.270* (0.16) 0.921*** (0.03) 8.667*** (1.32)

1.700*** (0.26) –1.31*** (0.23) –0.079*** (0.01) –0.121** (0.05) –0.237** (0.09) 0.003*** (0.000) 0.422*** (0.01) 0.2131*** (0.007) 5.106*** (1.15)

4.332*** (0.24) –4.44*** (0.24) 0.003* (0.001) 0.062*** (0.006) 0.088** (0.03) 0.003*** (0.000) –0.077** (0.03) –0.021 (0.03) 0.212 (0.43)

6.866*** (2.21) –6.60*** (2.11) –0.009 (0.01) 0.138*** (0.04) –0.112*** (0.02) 0.004*** (0.000) 0.329*** (0.10) –0.047 (0.09) 2.302*** (0.13)

145 0.84 96.49 yes yes yes

145 0.80 87.06 yes yes yes

D2it Public spending per capita, Git Per capita consumption, Cit Population size, Poit Population density, Deit Urbanization rate, Uit Ethnic fragmentation, Fit Constant Number of observations Adjusted R2 F-statistic Fixed effect Year dummies Instrumental variable

145 0.91 577.64 yes yes yes

145 0.59 149.11 yes yes yes

145 0.53 19.60 yes yes yes

Note: Controls for serial correlation of the error term, AR(1) Coccrane-Orcutt transformation. Robust standard errors are in brackets. ***Coefficient significant at the 1 percent level. **Coefficient significant at the 5 percent level. *Coefficient significant at the 10 percent level.

and donors in the struggle against poverty. The absence of a normalized scale for every basic service prevents concluding immediately on such a prioritization. However, through tables 3.18 and 3.19, we pinpoint some significant differences among local governments’ behaviors in relation to their wealth. The poorest communes are characterized by the negative impact of decentralization on access to primary education. This suggests that these governments pay less attention to education than they do to both drinking water access and toilet facilities (table 3.18). Table 3.20 focuses on the top 20 percent wealthier communes. For these communes the effect of decentralization on sewage access is negative and significant. In accordance with the reading of the seventy-seven detailed communes’ monographs,25 we may 25. Provided by the European Union through the Programme d’Appui au Démarrage des Communes.

94

Emilie Caldeira, Martial Foucault, and Grégoire Rota-Graziosi

Table 3.18

Estimation results: A differentiated effect of decentralization between communes (quintiles)

Dep. var.:

(1) TToilit

(2) TWatit

(3) TGarbit

(4) TSewit

(5) SEit

Dit * QPit

0.562* (0.28) 2.544*** (0.03) –0.006 (0.005) 0.190*** (0.02) –0.233** (0.10) –0.755*** (0.01) 0.004*** (0.000) 1.400*** (0.01) –0.602 (0.48) 7.575*** (0.35)

2.120*** (0.29) 2.355*** (0.27) –0.063 (0.03) 0.189* (0.10) –0.278*** (0.08) –0.587*** (0.10) 0.003*** (0.000) 0.147 (0.13) 0.843*** (0.04) 10.64*** (1.23)

0.416** (0.19) 1.162*** (0.08) 0.088*** (0.01) –0.121** (0.05) –0.126*** (0.01) –0.369*** (0.12) 0.003*** (0.000) 0.231*** (0.03) 0.115** (0.04) 6.719*** (1.39)

0.231 (1.24) 0.139 (0.41) 0.013 (0.14) 0.598** (0.24) –0.124 (0.53) 0.313 (0.25) 0.004*** (0.000) 0.551* (0.46) –0.247 (0.44) –0.434 (4.03)

–0.24*** (0.03) 0.656** (0.21) –0.010 (0.01) –0.13*** (0.04) 0.133*** (0.10) –0.106*** (0.03) 0.002*** (0.000) 0.009** (0.004) –0.088 (0.11) 3.465*** (0.31)

Dit * (1 − QPit) Public spending per capita, Git Per capita consumption, Cit Belong to the first quintile, QPit Population size, Poit Population density, Deit Urbanization rate, Uit Ethnic fragmentation, Fit Constant Number of observations Adjusted R2 Fixed effect Year dummies Instrumental variable Wald test: p-value

145 0.88 yes yes yes 0.000

145 0.84 yes yes yes 0.494

145 0.80 yes yes yes 0.011

145 0.63 yes yes yes —

145 0.50 yes yes yes 0.000

Note: Controls for serial correlation of the error term, AR(1) Coccrane-Orcutt transformation. Robust standard errors are in brackets. ***Coefficient significant at the 1 percent level. **Coefficient significant at the 5 percent level. *Coefficient significant at the 10 percent level.

deduce that the richest local governments, having already reached a certain level of quality in sanitation, choose to redirect their financial resources to other public facilities. Table 3.21 sums up our empirical results considering the effect of decentralization on the qualitative indicators. 3.6

Conclusion

Benin is a young democracy that has experienced a decentralization process since the end of the 1990s. The main objective of this institutional reform was to improve public policy governance and reduce poverty. Our

Does Decentralization Facilitate Access to Poverty-Related Services? Table 3.19

95

Estimation results: A differentiated effect of decentralization between communes (wealth)

Dep. var.:

(1) TToilit

(2) TWatit

(3) TGarbit

(4) TSewit

(5) SEit

Dit

–1.70*** (0.07) 1.597*** (0.04) 0.017** (0.008) 0.407*** (0.02) 0.274*** (0.03) –0.482*** (0.04) 0.001*** (0.000) –0.048 (0.03) –1.265*** (0.29) 2.889*** (0.64)

1.365*** (0.24) –0.01 (0.06) –0.065* (0.03) 0.052 (0.12) 0.247*** (0.06) –0.451*** (0.13) 0.003*** (0.000) –0.069 (0.15) 0.780*** (0.06) 10.93*** (1.58)

–0.94*** (0.16) 0.762*** (0.05) –0.084*** (0.01) –0.187*** (0.05) 0.114** (0.02) –0.177** (0.08) 0.001*** (0.000) 0.034 (0.04) –0.013 (0.05) 6.024*** (1.05)

–3.99*** (0.73) 2.312*** (0.25) 0.096*** (0.02) 0.652*** (0.06) 0.722*** (0.16) 0.621 (0.39) 0.004 (0.000) –0.158 (0.12) –0.784*** (0.28) –3.350 (4.38)

–0.64*** (0.18) 0.399*** (0.11) –0.010* (0.005) –0.148*** (0.04) 0.064*** (0.02) –0.001 (0.006) 0.006** (0.002) –0.066 (0.04) –0.176 (0.14) 2.898*** (0.07)

Dit * Wit Public spending per capita, Git Per capita consumption, Cit DHS wealth index scores, Wit Population size, Poit Population density, Deit Urbanization rate, Uit Ethnic fragmentation, Fit Constant Number of observations Adjusted R2 Fixed effect Year dummies Instrumental variable Wald test: p-value

145 0.88 yes yes yes 0.000

145 0.85 yes yes yes —

145 0.81 yes yes yes 0.000

145 0.57 yes yes yes 0.000

145 0.50 yes yes yes 0.000

Note: Controls for serial correlation of the error term, AR(1) Coccrane-Orcutt transformation. Robust standard errors are in brackets. ***Coefficient significant at the 1 percent level. **Coefficient significant at the 5 percent level. *Coefficient significant at the 10 percent level.

analysis focuses on the average and distributional effects of decentralization on access to poverty-related services. An original compilation of data sets concerning the well-being of households and local public finance allows us to study the ultimate effects of decentralization on the Beninese population. This study suggests that decentralization has an unambiguous positive overall effect on access to drinking water and sanitation systems. Beyond this average pattern, however, decentralization yields some distributional outcomes: its impact is nonlinear and heterogeneous. First, the effect of decentralization on access to basic services follows an inverted U-shaped curve with an optimal degree of decentralization (at 67 percent

96

Emilie Caldeira, Martial Foucault, and Grégoire Rota-Graziosi

Table 3.20

Estimation results: A differentiated effect of decentralization between communes (quintiles)

Dep. var.:

(1) TToilit

(2) TWatit

(3) TGarbit

(4) TSewit

(5) SEit

Dit * QRit

4.073*** (0.08) 2.414*** (0.05) –0.019** (0.007) 0.184*** (0.01) 0.185***

1.643*** (0.23) 1.569*** (0.25) –0.051 (0.03) 0.164* (0.12) 0.498***

1.126*** (0.09) 0.963*** (0.08) 0.088*** (0.01) –0.133** (0.05) 0.218***

–0.149*** (0.13) 0.103* (0.05) 0.001 (0.003) 0.054** (0.008) 0.088***

0.680*** (0.24) 0.261** (0.10) –0.010 (0.006) –0.133*** (0.04) –0.017***

(0.06) –0.643*** (0.02) 0.003*** (0.000) 0.823*** (0.06) –0.145 (0.39) 6.143*** (0.35)

(0.05) –0.387*** (0.12) 0.003*** (0.000) 0.141 (0.15) 0.877*** (0.03) 8.865*** (1.44)

(0.05) –0.296** (0.11) 0.002*** (0.000) 0.100*** (0.02) 0.285*** (0.01) 5.974*** (1.28)

(0.02) 0.021 (0.04) 0.005*** (0.000) 0.081*** (0.01) –0.007 (0.03) 0.079 (0.49)

(0.03) –0.052*** (0.01) 0.002*** (0.000) 0.042* (0.02) –0.027 (0.08) 2.936*** (0.09)

Dit * (1 − QRit) Public spending per capita, Git Per capita consumption, Cit Belong to the fifth quintile, QRit Population size, Poit Population density, Deit Urbanization rate, Uit Ethnic fragmentation, Fit Constant Number of observations Adjusted R2 Fixed effect Year dummies Instrumental variable Wald test: p-value

145 0.87 yes yes yes 0.000

145 0.84 yes yes yes 0.255

145 0.80 yes yes yes 0.000

145 0.58 yes yes yes 0.000

145 0.49 yes yes yes 0.005

Note: Controls for serial correlation of the error term, AR(1) Coccrane-Orcutt transformation. Robust standard errors are in brackets. ***Coefficient significant at the 1 percent level. **Coefficient significant at the 5 percent level. *Coefficient significant at the 10 percent level.

on average), showing that a minimum level of central transfers is still beneficial. Second, decentralization affects service access differently according to the communes’ individual wealth, having a positive effect on any nonmonetary poverty indicators, and a negative effect on the poorest communes. These results are consistent with those of Galiani, Gertler, and Schargrodsky (2008), who conclude that decentralization improves public services only in wealthier areas that have the ability to voice their preferences. Hence, if decentralization is adopted as a policy for improving overall access to basic services, it is essential for the central state to maintain a minimum level

Does Decentralization Facilitate Access to Poverty-Related Services? Table 3.21

Main empirical results

Average effect Toilet facility Water access Refuse disposal Sewage disposal Primary educ.

97

0.692*** 2.361*** 1.345*** NR NR

Nonmonotone average effect

Between communes

Dit

D

Poor

Nonpoor

Rich

Nonrich

2.190*** 3.234*** 1.700*** 4.332*** 6.866***

–1.96*** –1.56*** –1.31*** –4.44*** –6.60***

0.562* 2.120*** 0.416** 0.231 –0.24***

2.544*** 2.355*** 1.162*** 0.139 0.656**

4.073*** 1.643*** 1.126*** –0.14*** 0.680***

2.414*** 1.569*** 0.963*** 0.103*** 0.261***

2 it

Note: NR = Nonrobust. ***Coefficient significant at the 1 percent level. **Coefficient significant at the 5 percent level. *Coefficient significant at the 10 percent level.

of central transfers, in particular for the poorest communes, to avoid an increase in interjurisdictional inequalities. Finaly, the decentralization process in Benin has reduced poverty by improving access to some basic services, but increasing inequalities between communes are the counterpart of this process. Decentralization gives control over decisions and resources to local governments, whose aim is to target the poorest households better. In so doing, the central government treats the poor and local democratic institutions as assets and partners in the development process. Our results suggest that the patterns of decentralization in Benin describe improved access to primary services, but raise some issues about the design of transfers in both financial resources and competences. Indeed, certain basic services, mostly in education, have not been delivered to the degree expected. Controlling for different geographical and socioeconomic variables, poor communes do not improve primary education. A potential explanation rests on the idea that these local governments allocate their available resources for other basic services rather than education, which are considered more urgent, such as drinking water access and, to a lesser extent, toilet facilities.

98

Emilie Caldeira, Martial Foucault, and Grégoire Rota-Graziosi

Appendix Table 3A.1 Communes Banikoara Gogounou Kandi Karimama Malanville Segbana Bembereke Tchaourou Kalale N dali Nikki Parakou Perere Sinende Boukoumbe Cobly Kerou Kouande Materi Natitingou Pehunco Tanguieta Toucountouna Bassila Copargo Djougou Ouake Abomey–Calavi Allada Kpomasse Ouidah Toffo Torri–Bossito So–Ava Ze Cotonou Aplahoue Djakotomey Dogbo

List of communes (numbered) Number

Region

Communes

Number

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39

ALIBORI ALIBORI ALIBORI ALIBORI ALIBORI ALIBORI BORGOU BORGOU BORGOU BORGOU BORGOU BORGOU BORGOU BORGOU ATACORA ATACORA ATACORA ATACORA ATACORA ATACORA ATACORA ATACORA ATACORA DONGA DONGA DONGA DONGA ATLANTIQUE ATLANTIQUE ATLANTIQUE ATLANTIQUE ATLANTIQUE ATLANTIQUE ATLANTIQUE ATLANTIQUE LITTORAL COUFFO COUFFO COUFFO

Klouekanme Lalo Toviklin Athieme Bopa Come Grand–popo Houeyogbe Kolossa Adjarra Adjohoun Aguegues Akpro–Misserete Avrankou Bonou Dangbo Porto–Novo Seme–Kpodji Adja–Ouere Ifangni Pobe Ketou Sakete Bante Dassa–Zoume Glazoue Ouesse Savalou Save Abomey Agbangnizoun Bohicon Cove Djidja Ouinhi Za–Kpota Zagnanado Zogbodomey

40 41 42 43 44 45 46 47 49 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77

Region COUFFO COUFFO COUFFO MONO MONO MONO MONO MONO MONO OUEME OUEME OUEME OUEME OUEME OUEME OUEME OUEME OUEME PLATEAU PLATEAU PLATEAU PLATEAU PLATEAU COLLINES COLLINES COLLINES COLLINES COLLINES COLLINES ZOU ZOU ZOU ZOU ZOU ZOU ZOU ZOU ZOU

Fig. 3A.1

School enrollment

Source: http://www.gadm.org/country; authors’ calculations.

100

Emilie Caldeira, Martial Foucault, and Grégoire Rota-Graziosi

References Akai, N., and M. Sakata. 2002. “Fiscal Decentralization Contributes to Economic Growth: Evidence from State-Level Cross-Section Data for the United States.” Journal of Urban Economics 52 (1): 93–108. Alderman, H. 2002. “Do Local Officials Know Something We Don’t? Decentralization of Targeted Transfers in Albania.” Journal of Public Economics 83 (3): 375– 404. Alesina, A., and E. L. Ferrara. 2005. “Ethnic Diversity and Economic Performance.” Journal of Economic Literature 43 (3): 762–800. Alm, J., and J. Boex. 2002. “An Overview of Intergovernmental Fiscal Relations and Subnational Public Finance in Nigeria.” Discussion Paper no. 0201, International Studies Program, Andrew Young School of Policy Studies, Georgia State University. Arze, J., J. Martinez-Vasquez, and R. Puwanti. 2008. “Local Government Fiscal Competition in Developing Countries: The Case of Indonesia.” Urban Public Economics Review 8:13–45. Azfar, O., and J. A. Livingston. 2010 “Federalist Disciplines or Local Capture? An Empirical Analysis of Decentralization in Uganda.” Discussion Paper no. 00/12, IRIS, University of Maryland. Bardhan, P. 2002. “Decentralization of Governance and Development.” Journal of Economic Perspectives 16 (4): 185–205. Bardhan, P. K., and D. Mookherjee. 2000. “Capture and Governance at Local and National Levels.” American Economic Review 90 (2): 135–39. ———. 2005. “Decentralizing Antipoverty Program Delivery in Developing Countries.” Journal of Public Economics 89 (4): 675–704. Barenstein, M., and L. de Mello. 2001. “Fiscal Decentralization and Governance: A Cross-Country Analysis.” IMF Working Paper no. 01/71, Washington, DC, International Monetary Fund. Besley, T., and A. Case. 1995. “Incumbent Behavior: Vote-Seeking, Tax-Setting, and Yardstick Competition.” American Economic Review 85 (1): 25–45. Besley, T., and S. Coate. 2003. “Centralized versus Decentralized Provision of Local Public Goods: A Political Economy Approach.” Journal of Public Economics 87 (12): 2611–37. Bird, R. M., and E. Rodriguez. 1999. “Decentralization and Poverty Alleviation. International Experience and the Case of the Philippines.” Public Administration and Development 19:299–319. Cai, H., and D. Treisman. 2005. “Does Competition for Capital Discipline Governments? Decentralization, Globalization, and Public Policy.” American Economic Review 95 (3): 817–30. Caldeira, E., M. Foucault, and G. Rota-Graziosi. 2015. “Decentralization in Africa and the Nature of Local Governments’ Competition: Evidence from Benin.” International Tax Public Finance 22:1048–76. Case, A. 2001. “Election Goals and Income Redistribution: Recent Evidence from Albania.” European Economic Review 45 (3): 405–23. Chambas, G. 2010. Mobiliser des Ressources Locales en Afrique Subsaharienne. Paris: Economica. Cox, G. W., and M. McCubbins. 1986. “Electoral Politics as a Redistributive Game.” Journal of Politics 48 (2): 370–89. Enikolopov, R., and E. Zhuravskaya. 2007. “Decentralization and Political Institutions.” Journal of Public Economics 91 (11–12): 2261–90.

Does Decentralization Facilitate Access to Poverty-Related Services?

101

Faguet, J.-P. 2004. “Does Decentralization Increase Government Responsiveness to Local Needs? Evidence from Bolivia.” Journal of Public Economics 88 (3–4): 867– 93. Fan, C. S., C. Lin, and D. Treisman. 2009. “Political Decentralization and Corruption: Evidence from around the World.” Journal of Public Economics 93 (1–2): 14–34. Filmer, D., and L. H. Pritchett. 2001. “Estimating Wealth Effects without Expenditure Data or Tears: An Application to Educational Enrollments in States of India.” Demography 38 (1): 115–32. Fisman, R., and R. Gatti. 2002. “Decentralization and Corruption: Evidence across Countries.” Journal of Public Economics 83 (3): 325–45. Galasso, E., and M. Ravallion. 2005. “Decentralized Targeting of an Antipoverty Program.” Journal of Public Economics 89 (4): 705–27. Galiani, S., P. Gertler, and E. Schargrodsky. 2008. “School Decentralization: Helping the Good Get Better, but Leaving the Poor Behind.” Journal of Public Economics 92 (10–11): 2106–20. Hayek, F. A. 1948. Individualism and Economic Order. Chicago: University of Chicago Press. Huther, J., and A. Shah. 1998. “Applying a Simple Measure of Good Governance to the Debate on Fiscal Decentralization.” Policy Research Working Paper no. 1894, Washington, DC, World Bank. Lin, J. Y., and Z. Liu. 2000. “Fiscal Decentralization and Economic Growth in China.” Economic Development and Cultural Change 49 (1): 1–21. Lockwood, B. 2002. “Distributive Politics and the Costs of Centralization.” Review of Economic Studies 69 (2): 313–37. Martinez-Vazquez, J., and R. M. McNab. 2003. “Fiscal Decentralization and Economic Growth.” World Development 31 (9): 1597–616. Meyer, S. A., and S. Naka. 1999. “The Determinants of Japanese Local-Benefit Seeking.” Contemporary Economic Policy 17 (1): 87–96. Miguel, E., and F. Zaidi. 2003. “Do Politicians Reward Their Supporters? Regression Discontinuity Evidence from Ghana.” Discussion Paper, University of California, Berkeley. Oates, W. E. 1972. Fiscal Federalism. New York: Harcourt Brace Jovanovich. Plümper, T., and V. E. Troeger. 2007. “Efficient Estimation of Time-Invariant and Rarely Changing Variables in Finite Sample Panel Analyses with Unit Fixed Effects.” Political Analysis 15 (2): 124–39. Prud’homme, R. 1995. “The Dangers of Decentralization.” World Bank Research Observer 10 (2): 201–20. Reinikka, R., and J. Svensson. 2004. “Local Capture: Evidence from a Central Government Transfer Program in Uganda.” Quarterly Journal of Economics 119 (2): 678–704. Robalino, D. A., O. F. Picazo, and A. Voetberg. 2001. “Does Fiscal Decentralization Improve Health Outcomes? Evidence from a Cross-Country Analysis.” Policy Research Working Paper no. 2565, Washington, DC, World Bank. Rutstein, S. O. 2008. “The DHS Wealth Index: Approaches for Rural and Urban Areas.” DHS Comparative Reports no. 60, United States Agency for International Development. Rutstein, S. O., and K. Johnson. 2004. “The DHS Wealth Index.” DHS Comparative Reports no. 6, United States Agency for International Development. Salmon, P. 1987. “Decentralisation as an Incentive Scheme.” Oxford Review of Economic Policy 3 (2): 24–43.

102

Emilie Caldeira, Martial Foucault, and Grégoire Rota-Graziosi

Seabright, P. 1996. “Accountability and Decentralisation in Government: An Incomplete Contracts Model.” European Economic Review 40 (1): 61–89. Tiebout, C. M. 1956. “A Pure Theory of Local Expenditures.” Journal of Political Economy 64:416. Treisman, D. 2000. “The Causes of Corruption: A Cross-National Study.” Journal of Public Economics 76 (3): 399–457. Wallis, J. J. 1998. “The Political Economy of New Deal Spending Revisited, Again: With and Without Nevada.” Explorations in Economic History 35 (2): 140–70. Winkler, D. R., and T. Rounds. 1996. “Municipal and Private Sector Response to Decentralization and School Choice.” Economics of Education Review 15 (4): 365– 76. Wollera, G. M., and K. Phillips. 1998. “Fiscal Decentralisation and IDC Economic Growth: An Empirical Investigation.” Journal of Development Studies 34 (4): 139– 48. World Bank. 2004. World Development Report: Making Services Work for Poor People. New York: Oxford University Press. Xie, D., H.-F. Zou, and H. Davoodi. 1999. “Fiscal Decentralization and Economic Growth in the United States.” Journal of Urban Economics 45 (2): 228–39. Zhang, T., and H.-F. Zou. 1998. “Fiscal Decentralization, Public Spending, and Economic Growth in China.” Journal of Public Economics 67:221–40.

4

Demographic Pressure and Institutional Change Village-Level Response to Rural Population Growth in Burkina Faso Margaret S. McMillan, William A. Masters, and Harounan Kazianga

4.1

Introduction and Motivation

An unusual factor in Africa’s twentieth-century agricultural development was a relatively low initial level of average population density coupled with unusually high rates of rural population growth over the last thirty years. As shown in figure 4.1, Africa’s year-to-year rate of rural population growth rose above that of Asia around 1975, peaked in 1990, and only recently has fallen below the highest levels ever seen in other regions. All regions have seen a rise and then fall in their annual rates of rural population growth, but in the post-1975 period Africa’s growth rate rose more recently and reached a higher level for a longer time than that of other regions. This project investigates the link between rural population growth and the local institutions and infrastructure needed for market development in agriculture. We use spatial differences in migration exposure to test how village societies have responded to population pressure. Our central hypothesis is that recent increases in rural population densities are associated with a Margaret S. McMillan is associate professor of economics at Tufts University and a research associate of the National Bureau of Economic Research. William A. Masters is a professor at Tufts University in the Friedman School of Nutrition with a secondary appointment in the Department of Economics. Harounan Kazianga is associate professor of economics at Oklahoma State University. We are grateful to the NBER African Successes Project for financial support, and thank Moussa Kabore and colleagues in the Direction de la Prospective et des Statistiques Agricoles et Alimentaires (DPSAA) of Burkina Faso for survey implementation, Jose Castillo for research assistance, and Jun Folledo of IFPRI for GIS calculations. The chapter was much improved by valuable suggestions from Will Dow and other participants at the Zanzibar Conference of the NBER African Successes Project, August 3–5, 2011, as well as comments from Rohini Pande and Remi Jedwab. For acknowledgments, sources of research support, and disclosure of the authors’ material financial relationships, if any, please see http://www.nber.org/chapters /c13446.ack.

103

104

Margaret S. McMillan, William A. Masters, and Harounan Kazianga

A

B

Fig. 4.1 Past and projected rural population growth, by region and country (1950– 2050): A, Regional aggregates; B, Burkina Faso Source: Calculated from UN Population Projections (esa.un.org/unpp).

wider spread of rural public services, infrastructure, and local marketplaces; a transition from open access to regulated land use, including stronger individual property rights; and more reliance on the rule of law to adjudicate disputes. Our data come from Burkina Faso, a landlocked West African country of about 13 million people. As shown in figure 4.1, from 1950 to 2005 Burkina Faso’s rural population growth rate rose even more dramatically than that of Africa as a whole, to a peak above 2.5 percent per year. Burkina’s rural population growth rate is projected to decline rapidly in the coming decades, but

Demographic Pressure and Institutional Change

105

will remain well above zero until the absolute size of the urban population becomes large enough for its annual growth to absorb each year’s increase in the country’s entire population. Figure 4.1 shows that rapid growth in Burkina Faso’s rural population was not uniform in time, with a temporary reversal in the 1980s that may have been associated with migration to Côte d’Ivoire or other factors, followed by a burst of catch-up growth and downward projections until urbanization is sufficient to achieve zero rural population growth around 2050. Historically, Burkina Faso has had large movements of rural people to its own cities and a large migration to coastal Côte d’Ivoire after colonization and particularly through the 1980s. A large number of those migrants were then forcibly repatriated following civil unrest in Côte d’Ivoire starting in the late 1990s. In addition, the donor-funded Onchocerciasis Control Program quickly eradicated river blindness starting in the 1970s, leading to large population movements into river valleys. These demographic shocks affected villages across Burkina Faso in different ways depending on their location, offering two different exogenous shocks to rural population density with which to study the impact of rural demography on local institutions and infrastructural investments. We hypothesize that changes in rural population growth change the payoffs from collective action, making it relatively more urgent to develop market infrastructure and institutions. This hypothesis follows Boserup (1965), who argued that rising rural population densities create incentives not only for farm-level adoption of more input-intensive techniques and “induced invention” of new technologies in response to factor scarcity as suggested by Hicks (1932), but also induced institutional changes to allocate newly scarce natural resources more efficiently. A link between rural population density and rural public goods could also be due to political pressures or indivisibilities and scale effects in the provision of infrastructure and institutions. Both relative price and scale effects could be subject to time lags, leading rural population growth to have a Malthusian effect in the short run, even as it facilitates the institutional and technological innovations needed for later agricultural productivity growth. Modern analyses of how population density and factor scarcity affect agricultural development were pioneered by Hayami and Ruttan (1971) for the United States and Japan, and tested in a large subsequent literature such as Olmstead and Rhode (1993). Only a few of these papers (e.g., Lin 1995) focus on the emergence and adoption of institutions; most ask how institutions affect technology adoption, such as Kazianga and Masters (2002, 2006). Focusing on rural demography also expands on our other previous work regarding the role of environmental factors in economic growth (Masters and McMillan 2001) and African policy choices (McMillan 2001; McMillan and Masters 2003). Here, we focus on changes in village-level institutions, testing how the governance of local resources and market infrastructure has responded to demographic change among local households.

106

Margaret S. McMillan, William A. Masters, and Harounan Kazianga

Our focus on the specific challenge of rural population growth for agricultural development follows Johnston and Kilby (1975), among others. Most of the development economics literature concerned with demography has focused either on demographic transition in the population as a whole (including the demographic “drag” or “dividend” from age structure emphasized by Bloom and Williamson [1998]), or the structural transformation from farm to nonfarm employment in terms of output and employment shares, including the one-time “growth bonus” associated with shifting from a low productivity to a high productivity sector as in Temple (2005). Focusing on demographic conditions within rural areas addresses a distinctive aspect of Africa’s postindependence economic decline and are grounds for optimism about the future as rural infrastructure and institutions adapt to higher levels of population density and the speed of further demographic slows down. The motivation for our approach begins with an economic view of rural demography. Demographic accounting ensures that each locality’s rural population growth is its natural increase (births minus deaths, which in turn are determined by age structure as well as age-specific mortality and fertility), plus or minus each year’s net migration to urban or other rural areas. From an economic point of view, however, both fertility and migration are choice variables, and mortality may also be influenced by investment in health. Given this endogeneity, identification of a potentially causal effect of population requires an exogenous shock to rural population size that occurs with sufficient speed and magnitude to induce a measurable institutional response. Our study design takes advantage of Burkina Faso’s unusual demographic history, which includes two large waves of exogenous migration into specific rural areas from the 1970s through the early twenty-first century. One wave flowed into river valleys in response to an international campaign of Onchocerciasis eradication, which made those locations newly attractive, and another wave flowed in from Côte d’Ivoire in response to political violence there. We use three rounds of census data in 1985, 1996, and 2006 to capture the resulting variation in village population, and compare that to variance in institutions and infrastructure as recalled by focus group interviews of village elders. Our work contributes to an important gap in the literature on institutions and economic development identified by Pande and Udry (2006), who argue that “the research agenda identified by the institutions and growth literature is best furthered by the analysis of much more microdata than has typically been the norm in this literature.” Specifically, we study the historical evolution of institutions in response to demographic pressure by focusing on diversity across villages in a setting with wide variation in exposure to clearly exogenous demographic shocks. The closest antecedent is probably Grimm and Klasen (2008), who test for endogenous adoption of land titles

Demographic Pressure and Institutional Change

107

at the village level on Sulawesi in Indonesia. Our surveys include land titles and also consider a very wide range of other institutions, public services, and infrastructure used for market exchange. Methodologically, our use of focus groups to obtain village-level recall data on the location and availability of public services follows Chattopadhyay and Duflo (2004), building on a long tradition of participatory surveys in rural areas (e.g., Chambers 1994). This approach allows us to ask about many different types of public services, with access to physical infrastructure measured by its proximity to the village center. Though not the central focus of this particular chapter, our survey data could also be used to analyze causal effects of public services and institutions on economic outcomes. For example, Besley (1995) and others have found evidence that institutions significantly affect investment outcomes in rural Africa (see Pande and Udry [2006] for a summary of these studies). In Burkina Faso, Kazianga and Masters (2002) found that stronger cropland tenure was associated with more intensive soil and water conservation. Our approach to changes in village-level infrastructure and institutions is also relevant to the mechanisms by which large-scale public health interventions influence economic development, as in Acemoglu and Johnson (2007), Bleakley (2007), and Cutler et al. (2010). In the next section, we describe the major exogenous population shifts that might permit identification of how changes in rural population density affect public goods provision. We then turn to our empirical strategy and a description of our data in section 4.3. In section 4.4 we present and discuss our results. Section 4.5 concludes. 4.2

Historical Background

Since independence in 1960, Burkina Faso has experienced two major policy-induced changes in settlement patterns. The first began in 1974 when the Onchocerciasis Control Program was launched by the World Bank to control river blindness in seven West African countries: Benin, Burkina Faso, Côte d’Ivoire, Ghana, Mali, Niger, and Togo. The second occurred from the late 1990s until 2002, when up to one million Burkinabe returned from Côte d’Ivoire to escape violence and a suspension of immigrants’ rights in that country. Since our ability to draw a causal link between population growth and institutional change hinges on the extent to which these two events were exogenous to other influences on village population size, we describe the two shocks in more detail below. 4.2.1

The Onchocerciasis Control Program

The Onchocerciasis Control Program (OCP) was initiated in 1974 to control river blindness in West Africa, and is widely considered to be among the most successful public health programs ever launched in sub-Saharan

108

Margaret S. McMillan, William A. Masters, and Harounan Kazianga

Africa. Onchocerciasis, or “river blindness,” is primarily a rural disease that affects sub-Saharan Africa more than anywhere else in the world. The disease is spread through bites from black flies of the genus Simulium, which transmit the larvae of a filarial worm, Onchocerca volvulus. The worms multiply only in the human body, where they cause debilitating symptoms that include blindness, and are transmitted only by the black fly, which lives in proximity to fast-moving rivers. The OCP was a multilateral effort that covered eleven countries, including Burkina Faso. The program involved weekly aerial treatment and groundlevel treatment of black fly breeding grounds. Annual drug treatments offered immediate relief from the symptoms and elimination of nearly all offspring of the adult worm. Today, the disease is no longer considered a threat in the control zone, which has consequently attracted in-migration from other rural areas (McMillan, Nana, and Savadogo 1992, 1993). To control the anticipated immigration to these newly attractive areas, the government of Burkina Faso created a special national agency—the Volta Valley Authority (AVV)—and gave the agency control of 75 percent of the river basins. Figure 4.2 shows these locations, and the “planned” villages to which it provided financial and institutional support. However, the pace of spontaneous settlement soon outgrew the ability of the AVV to finance

Fig. 4.2 Location of planned settlements associated with Onchocerciasis control, 1973–1984 Source: McMillan, Nana, and Savadogo (1993).

Demographic Pressure and Institutional Change

109

and create sufficient numbers of sponsored settlements. As a result, there were sizable intra- and interregional differences in the rate of new lands settlement documented by McMillan, Nana, and Savadogo (1992), as well as substantial variation in land-use practices and land management institutions described by McMillan, Nana, and Savadogo (1993). 4.2.2

Repatriation from Côte d’Ivoire

For more than three decades after independence from France in 1960, Côte d’Ivoire was an important destination for immigrants from Burkina Faso, offering peaceful stability and economic prosperity, including rural work associated with opening new forests for cocoa production. The death of the autocratic ruler Felix Houphet-Boigny in 1993 ushered in a new era. His successor, Henri Konan Bedie, has been accused of sowing the seeds of ethnic discord by introducing the concept of “Ivorian-ness” in 1995, allegedly to deny Ivorian citizenship to his main political rival, Alassane Ouattara, thereby excluding him from office. Bedie insisted that Ouattara, a Muslim from the north of the country, was actually from Burkina Faso. Subsequently, attacks on people of foreign descent became increasingly widespread (Human Rights Watch 2001). By that time, more than one quarter of Côte d’Ivoire’s population had immigrated to the country since independence, the overwhelming majority of whom had come from Burkina Faso. As shown in figure 4.2, the Côte d’Ivoire census of 1998 identified about 2.25 million Burkinabe living in Côte d’Ivoire, which was close to 20 percent of Burkina’s total population at that time. Peace and stability in Côte d’Ivoire came to an abrupt halt on December 24, 1999, when the military, under the leadership of General Robert Guei, overthrew the elected government of Konan Bedie in the country’s first coup d’état. Although the coup was ostensibly prompted by soldiers’ unhappiness over pay and conditions, it soon became apparent that, like Bedie, General Guei was also ready to incite ethnic and religious rivalries in order to remove political opposition. Continuing the theme of Ivorian-ness, Guei introduced even stricter eligibility requirements for the 2000 presidential elections, once again excluding Alassane Ouattara on the basis of his alleged links with Burkina Faso. Though exact numbers are difficult to come by, it is estimated that between 1999 and 2002 hundreds of thousands of Burkinabe were repatriated as a result of political unrest and worsening economic conditions in Côte d’Ivoire. They returned by rail, road, and on footpaths, often but not always to their original villages. 4.3

Empirical Strategy, Data, and Descriptive Statistics

Our evidence on village-level access to public services, infrastructure, and institutions comes from a novel survey conducted for this project by the

110

Margaret S. McMillan, William A. Masters, and Harounan Kazianga

Burkina Faso Office of Agricultural Statistics in January through June 2010. This survey asked groups of village elders to discuss and describe the history of the facilities around them, recording the date of any changes in the distance to each kind of facility and any changes in property-rights arrangements. From those underlying observations, we construct a time-varying index of the village’s proximity to public services, public infrastructure, religious services, and markets, as well as time-varying indicators of property rights over land. We combine these indexes with population estimates for each village from the Burkina Faso national censuses of 1986, 1996, and 2006 to test whether variance in population size can help explain variance in the provision of public services, infrastructure, and institutions. To overcome endogeneity between a village’s amenities and its population size, we use each village’s straight-line distance to any river from which Onchocerciasis could have been eradicated, as well as distance to the Côte d’Ivoire border from which migrants could have returned as instruments for the village’s population in each survey year. The result is a set of two-stage least squares (2SLS) regressions asking whether population shocks associated with changes in the attractiveness of rivers and of Côte d’Ivoire are correlated with the spread of rural public services, infrastructure, and market institutions. Our chapter does not identify the mechanism by which more populated villages might attract more rural public services, infrastructure, or market institutions: instead, we are testing for reduced-form relationships, exploiting an unusual natural experiment in rural population density. Our sample of villages consists of 747 sites that had previously been selected by the Office of Agricultural Statistics for their nationally representative agricultural survey conducted annually since the early 1990s. In this context, villages are very small, averaging about a thousand people. Their boundaries can change somewhat from decade to decade, as some households split off into new settlements. Our final data set consists of 730 villages whose recorded names are the same across the three censuses and our new survey at a correctly recorded geographic information system (GIS) location. We use year and region fixed effects for each of Burkina’s forty-five provinces in order to focus on spatial variation across villages within relatively small administrative units. The survey instrument is provided in the appendix. It was administered by experienced enumerators employed for Burkina’s annual agricultural survey, whose structure is designed to accommodate new survey modules. The survey began by assembling a focus group of village elders and officials who were asked a series of detailed questions regarding various types of public services, infrastructure, and institutions available to them. For each variable, we typically asked for its distance from the village and other salient characteristics, at present and in previous years, along with the date of any change. For example, the section on property rights poses the following question: Can land be sold in your village? If the group agrees that the answer

Demographic Pressure and Institutional Change

111

to this question is yes, the interviewer then asks: Since when could land be sold in your village? Questions posed in this way allow us to construct timevarying indexes of public amenities from the point of view of the villagers themselves. Our results focus on two kinds of variables: travel distances to public amenities, and categorical indicators of land-use rights, both as reported for each census year. The travel distances to collective amenities are grouped into four categories: (a) Public Services and Utilities, defined as the administrative office used to register births, any savings and loan facility, any fixed-line telephone, or any mobile phone reception; (b) Public Infrastructure, defined as a road that is accessible by truck all year, a road accessible by truck seasonally, a bus stop, a primary school, a secondary school, or a health center; (c) Religious Services, defined as any church, mosque, or temple; and (d) Markets, defined as any market with storage facilities, any livestock market, or a private shop. These are all the distances for which our group-interview technique elicited unambiguous agreement in at least 700 of the 730 villages. Other questions, such as distance to water wells, bridges, and electricity supplies, were less likely to elicit agreement, perhaps because those amenities are less salient to villagers’ lives or their use is more varied among the respondents. The distances to collective amenities were then aggregated in each of three ways. First, we consider the distance one must travel to have access to all the services in a given category, that is, the distance associated with the farthest one. Second, we consider the average distance to all of the services in the group, in other words, the arithmetic mean of each distance. Finally, we consider the distance to any of the listed services, that is, the minimum distance among them. Categorical indicators of land rights address three kinds of land use. First, we ask whether use rights over cropland are undefined or held by individuals, families, or the community. Then we ask whether cropland had ever been rented or sold, which we take to indicate the presence of a land market. Finally, we ask whether villagers recognize a formal authority that regulates access to pasture land, forests, and potentially cropped land. Table 4.1 presents the proportion of all observations with each category of property right, as reconstructed for the census years of 1985, 1996, and 2006. For example, rights over cropland are not defined in 14.4 percent of village-year observations. Descriptive statistics on all variables as used in the regressions are provided in table 4.2, separately for each year to reveal the time trends. Public services become more closely available and property rights are more tightly regulated in more recent years. Also, note that the average population of all surveyed villages grows from 1985 to 1996, but then falls in 2006. There is likely to have been systematic undercounting of the rural population in 2006, which is why the Burkina government is planning a new census several years ahead of its decennial schedule.

112

Margaret S. McMillan, William A. Masters, and Harounan Kazianga

Table 4.1

Property rights and land use across sample villages in Burkina Faso (n = 2,170) Percentage of observations in sample

Land rights Rights over cropland Not defined Communal Familial Individual Existence of sales or rental of cropland None At least one sale or rental has occurred Role of traditional authorities in solving cropland conflict None Some Role of elected authorities involved in solving cropland conflict None Some Demarcation and regulation of pastureland No delimited pastureland Pastureland delimited, access not regulated Pastureland delimited, access regulated by tax or quota Demarcation and regulation of forestland No delimited forestland Forestland delimited, access not regulated Forestland delimited, access regulated by tax or quota

14.4 10.0 59.9 15.7 92.4 7.7 63.8 36.2 81.9 18.1 71.7 80.9 19.1 70.1 15.9 14.0

Source: Authors’ calculations. Notes: Results shown are from village elders’ response to questions asked in local languages, translated by local enumerators from the French questionnaire reproduced in the appendix to this chapter. Items shown are from questionnaire sections VIII (for cropland), IX (for pastureland), and X (for forestland).

4.4

Estimating Equations and Results

Our estimation begins with a set of descriptive ordinary least squares (OLS) regressions showing the correlations between village-level population and public infrastructure or institutions, controlling for year and province fixed effects, using the following specification: (1)

I jkt = a + bPjt + dX j + gt + ´ jkt,

where I is our measure of infrastructure or institution of type k in village j at time t from the survey data, and P is our measure of the total population in village j at year t from the census data, and γ are time dummies. The X controls for province fixed effects, and in robustness tests also controls for the ethnic composition of village population, or more generally for village fixed effects. Our hypothesis is that β > 0, as larger populations facilitate the provision of public goods and market institutions, due either to relative

Individual

0.41 [0.018] 0.423 [0.019] 0.453 [0.018]

1985

9.274 [0.518] 7.465 [0.409] 5.218 [0.331]

Religious services

0.665 [0.018] 0.671 [0.018] 0.669 [0.017]

Familial

0.1 [0.011] 0.099 [0.011] 0.104 [0.011]

Communal

12.832 [0.790] 12.735 [0.741] 11.455 [0.611]

Markets

0.056 [0.009] 0.064 [0.009] 0.107 [0.011]

Land markets

26.915 [0.779] 25.055 [0.708] 19.681 [0.543]

Public services 6.321 [0.360] 4.726 [0.267] 3.036 [0.194]

Religious services

0.152 [0.013] 0.186 [0.015] 0.24 [0.016]

Pasture 1.353 [0.024] 1.44 [0.027] 1.516 [0.029]

Forest

8.585 [0.432] 7.811 [0.384] 6.11 [0.276]

Markets

2.75 [0.033] 2.751 [0.034] 2.786 [0.033]

Crop

Regulated access to

14.662 [0.482] 11.532 [0.374] 8.099 [0.278]

Public infrast.

1.6 [0.058] 1.682 [0.059] 1.396 [0.091]

Population (1,000s)

18.351 [0.607] 15.115 [0.533] 8.596 [0.415]

Public services

4.855 [0.269] 3.788 [0.251] 1.975 [0.176]

Markets

65.986 [1.782] 66.876 [1.818] 66.336 [1.777]

506.904 [8.787] 506.478 [8.984] 509.231 [8.753]

Côte d’Ivoire

Distance (km) to:

3.536 [0.299] 2.328 [0.230] 1.16 [0.138]

Religious services

Nearest river

3.566 [0.308] 1.817 [0.209] 0.501 [0.083]

Public infrast.

Proximity to closest source (km) (distance to closest site in each set)

Source: Authors’ calculations. Notes: Standard deviations are in brackets. Proximity measures refer to travel distances from the village to reach the closest site offering one or more of each set of collective resources: Public Services and Utilities (defined as the administrative office used to register births, any savings and loan facility, any fixed-line telephone, any mobile phone reception); Public Infrastructure (defined as a road that is accessible by truck all year, a road accessible by truck seasonally, a bus stop, a primary school, a secondary school, and a health center); Religious Services (any church, mosque, or temple); and Markets (any open-air food market, livestock market, or private shop). Specific wording of each question is reproduced in the appendix. From the questionnaire as a whole, we retained only those proximity questions that more than 700 of the 730 villages were unable to answer unambiguously. Population is computed from the Burkina Faso national censuses for 1985, 1996, and 2006. Distances to nearest river and to the Côte d’Ivoire border are straight lines calculated from latitude and longitude geocodes.

2006

1996

35.458 [1.239] 28.053 [0.977] 20.955 [0.771]

Public infrast.

Land ownership rights

Year

2006

1996

35.348 [1.206] 35.635 [1.137] 32.151 [1.005]

Public services

Proximity to all sources (km) (average distance to all services)

Mean and standard deviations for all variables (n = 2,121)

Proximity of farthest source (km) (distance to farthest site in each set)

1985

Year

Table 4.2

114

Margaret S. McMillan, William A. Masters, and Harounan Kazianga

scarcities as in Boserup (1965) or to indivisibilities at the relevant scale of population size. Estimates of regression (1) are shown in table 4.3, where X controls only for province fixed effects. In columns (1)–(4) the dependent variable is the maximum distance one must travel to have access to all amenities in each category. In columns (5)–(8) the dependent variable is the average distance one must travel to access any amenity in each category, and in columns (9)–(12) the dependent variable is the minimum distance one must travel to access at least one of them. Both the distances and village population are expressed in natural logs, so that the coefficients can be interpreted as elasticities. We find that larger villages have closer amenities in eleven of the twelve regressions; the one exception is column (3), where only the time trend is significant. Institutions for land use are significantly linked to village population in only two of the seven regressions. Table 4.4 repeats the diagnostic OLS regression with additional controls for the number of ethnic groups and number of clans in the village, as a crude approximation of the village’s social fragmentation that might influence political cooperation and collective action for public goods provision (e.g., Alesina and La Ferrara 2005). The correlation between population size and access to public goods is robust to these controls. Estimated coefficients on population size are somewhat smaller when controlling for ethnic diversity, but contrary to some hypotheses the more diverse villages actually have more public infrastructure than the less diverse ones. In the absence of any clear identification strategy regarding fragmentation, however, for this chapter we focus on the main relationship concerning total population size. Finding significant coefficients in these OLS regressions is not surprising, as people could choose to locate in villages with closer access to public institutions and services, or both could be caused by something else. To overcome endogeneity, we use instrumental variables for population, so that the only variation in village population that we actually use is associated with distance to rivers and distance to the border with Côte d’Ivoire, and changes in these associations over time. The first-stage regression of our 2SLS system is specified as follows: (2)

Pjt = a0 + a1G j + a2TtG j + a3Tt + a4M j + εj

where G is a vector of the logs of geographic distance to rivers and to the border with Côte d’Ivoire, T is year dummies for 1996 and 2006, and M is controls imposed through province fixed effects. When using the resulting predicted village populations in equation (1), our identifying assumption is that a village’s distance to rivers and to Côte d’Ivoire have no other channel of influence on infrastructure and institutions beyond their importance for population size. Some evidence regarding the validity of those exclusion restrictions is provided here using Hansen’s J statistic, but that test is not conclusive. This initial use of our data concerns Burkina Faso as a whole, and

2,146 0.335

Observations R-squared

2,146 0.294

–0.012 [0.010] 0.005 [0.021] 0.001 [0.021] 0.747*** [0.071]

Family (14)

2,161 0.220

2,146 0.139

0.012* [0.007] –0.002 [0.015] 0.007 [0.015] 0.016 [0.051]

2,040 0.265

0.145*** [0.025] 0.167*** [0.054] 0.515*** [0.052] –2.14*** [0.220]

Religion (7)

2,146 0.379

0.010* [0.005] 0.008 [0.010] 0.053*** [0.011] –0.013 [0.037]

Land markets (16)

0.049** [0.021] 0.193*** [0.041] 0.538*** [0.041] –2.71*** [0.175]

Infrastr. (6)

Communal (15)

2,132 0.116

0.041* [0.023] 0.025 [0.045] 0.255*** [0.043] –2.97*** [0.197]

Services (5)

Proximity to all sources (km)

2,146 0.181

0.009 [0.009] 0.036* [0.018] 0.091*** [0.019] 0.087 [0.064]

2,161 0.234

0.116*** [0.020] 0.305*** [0.047] 0.611*** [0.043] –1.40*** [0.167]

Infrastr. (10)

2,146 0.227

–0.021 [0.016] 0.086*** [0.033] 0.158*** [0.034] 1.498*** [0.116]

Forest (18)

Regulated access

2,132 0.230

0.108*** [0.032] 0.191*** [0.056] 1.109*** [0.060] –2.69*** [0.268]

Services (9)

Pasture (17)

2,136 0.225

0.187*** [0.027] 0.034 [0.056] 0.222*** [0.054] –3.03*** [0.292]

Markets (8)

2,146 0.313

–0.011 [0.019] 0.008 [0.039] 0.032 [0.040] 2.828*** [0.135]

Cropland (19)

2,040 0.232

0.125*** [0.022] 0.225*** [0.051] 0.490*** [0.047] –1.50*** [0.195]

Religion (11)

Proximity to closest source (km)

2,136 0.235

0.228*** [0.023] 0.193*** [0.054] 0.628*** [0.051] –2.64*** [0.243]

Markets (12)

Source: Authors’ calculations. Notes: Population and distance measures are in logs, with proximity defined as its additive inverse (–log[distance]), so that coefficients can be read as elasticities and a positive coefficient implies closer facilities. The regression also controls for forty-five province dummies (not shown). Robust standard errors are in brackets. ***Significant at the 1 percent level. **Significant at the 5 percent level. *Significant at the 10 percent level.

Constant

Year = 2006

Year = 1996

0.012 [0.010] 0.018 [0.022] 0.044** [0.022] 0.324*** [0.074]

2,136 0.220

0.182*** [0.031] –0.030 [0.064] 0.072 [0.064] –3.20*** [0.341]

Markets (4)

Land ownership rights

2,040 0.233

0.153*** [0.029] 0.127** [0.063] 0.493*** [0.062] –2.44*** [0.266]

Religion (3)

Individual (13)

2,161 0.147

0.003 [0.027] 0.174*** [0.051] 0.485*** [0.050] –3.18*** [0.209]

Infrastr. (2)

lnpopulation

2,132 0.097

Observ. R-squared

Constant

Y = 2006

Y = 1996

0.045* [0.026] –0.059 [0.051] 0.027 [0.048] –3.29*** [0.223]

Services (1)

Proximity of farthest source (km)

OLS regression results for public infrastructure and institutions on village-level population

Population

Table 4.3

2,132 0.121

Observ. R-squared

Constant

Y = 2006

Y = 1996

Ethnicities

Clans

0.027 [0.025] –0.001 [0.004] 0.063*** [0.013] –0.057 [0.050] 0.025 [0.047] –3.27*** [0.220]

Services (1)

2,161 0.157

–0.009 [0.026] –0.001 [0.003] 0.044*** [0.011] 0.175*** [0.051] 0.483*** [0.050] –3.16*** [0.207]

Infrastr. (2).

2,040 0.273

0.107*** [0.029] 0.017*** [0.003] 0.065*** [0.011] 0.134** [0.062] 0.487*** [0.060] –2.38*** [0.258]

Religion (3)

Markets (4)

2,136 0.243

0.144*** [0.031] 0.013*** [0.004] 0.057*** [0.012] –0.026 [0.064] 0.066 [0.063] –3.15*** [0.338]

Proximity of farthest source (km)

2,132 0.150

0.021 [0.022] –0.000 [0.003] 0.066*** [0.012] 0.028 [0.045] 0.252*** [0.042] –2.94*** [0.193]

Services (5)

2,161 0.241

0.031 [0.021] 0.001 [0.002] 0.050*** [0.008] 0.195*** [0.041] 0.534*** [0.040] –2.68*** [0.172]

Infrastr. (6)

2,040 0.302

0.107*** [0.024] 0.013*** [0.002] 0.058*** [0.009] 0.173*** [0.053] 0.511*** [0.051] –2.09*** [0.215]

Religion (7)

Proximity to all sources (km)

2,136 0.258

0.150*** [0.026] 0.012*** [0.003] 0.061*** [0.010] 0.037 [0.055] 0.216*** [0.053] –2.98*** [0.288]

Markets (8)

2,132 0.268

0.068** [0.031] 0.008** [0.004] 0.087*** [0.012] 0.196*** [0.054] 1.103*** [0.059] –2.63*** [0.262]

Services (9)

OLS regression results for public infrastructure and institutions on village-level population and diversity

Population

Table 4.4

2,161 0.243

0.100*** [0.020] 0.004** [0.002] 0.028*** [0.006] 0.306*** [0.047] 0.608*** [0.043] –1.38*** [0.165]

Infrastr. (10)

2,040 0.247

0.105*** [0.022] 0.006*** [0.002] 0.035*** [0.008] 0.229*** [0.051] 0.488*** [0.047] –1.48*** [0.195]

Religion (11)

Proximity to closest source (km)

2,136 0.271

0.192*** [0.023] 0.011*** [0.002] 0.063*** [0.008] 0.197*** [0.053] 0.622*** [0.050] –2.59*** [0.237]

Markets (12)

2,132 0.335

Observations R-squared

2,132 0.294

–0.015 [0.010] –0.002 [0.004] 0.002* [0.001] 0.005 [0.021] 0.000 [0.021] 0.756*** [0.072]

Family (14)

2,132 0.163

0.004 [0.007] 0.005 [0.003] 0.005*** [0.001] –0.001 [0.015] 0.005 [0.015] 0.005 [0.050]

Communal (15)

2,132 0.383

0.008 [0.005] 0.007** [0.003] –0.000 [0.001] 0.009 [0.010] 0.053*** [0.011] –0.024 [0.036]

Land markets (16)

2,132 0.180

0.011 [0.009] –0.008** [0.003] 0.001 [0.001] 0.037** [0.018] 0.092*** [0.019] 0.096 [0.064]

Pasture (17)

2,132 0.231

–0.030* [0.017] 0.011* [0.006] 0.003 [0.002] 0.087*** [0.033] 0.158*** [0.034] 1.487*** [0.117]

Forest (18)

Regulated access

2,132 0.307

–0.013 [0.019] 0.004 [0.007] –0.001 [0.002] 0.008 [0.039] 0.032 [0.040] 2.834*** [0.136]

Cropland (19)

Source: Authors’ calculations. Notes: Population and distance measures are in logs, with proximity defined as its additive inverse (–log[distance]), so that coefficients can be read as elasticities and a positive coefficient implies closer facilities. The regression also controls for forty-five province dummies (not shown). Robust standard errors in brackets. ***Significant at the 1 percent level. **Significant at the 5 percent level. *Significant at the 10 percent level.

Constant

Year = 2006

Year = 1996

Clans

Ethnicities

0.007 [0.011] 0.009** [0.004] 0.001 [0.001] 0.019 [0.022] 0.044** [0.022] 0.308*** [0.075]

lnpopulation

Individual (13)

Land ownership rights

118

Margaret S. McMillan, William A. Masters, and Harounan Kazianga

to investigate more deeply with stronger identification, future work could focus on specific regions and times when more narrowly defined natural experiments have occurred. First-stage results are shown in table 4.5. Our preferred specification with both distances is in column (3), while columns (1) and (2) show results with only (log) distance to rivers and to Côte d’Ivoire, respectively. Columns (1) and (3) indicate that villages located further from rivers are less populated than other villages, with no significant difference between census years. As documented by McMillan, Nana, and Savadogo (1992), much of the population movement triggered by river blindness control had already occurred by the 1985 census, so this effect is primarily cross-sectional in our data. Table 4.5

First-stage regression results for IV estimation (1)

Excluded instruments: Distance to river Distance to river*1996 Distance to river*2006

Distance to border*1996 Distance to border*2006

Year = 2006 Constant Observations R-squared F-stat. inst.

(3)

–0.181* [0.110] –0.140** [0.067] –0.149** [0.067]

–0.155*** [0.038] 0.034 [0.056] 0.073 [0.053] –0.044 [0.113] –0.143** [0.067] –0.157** [0.067]

0.932** [0.409] 0.689* [0.415] 7.985*** [0.720]

0.813* [0.435] 0.458 [0.448] 7.605*** [0.734]

2,146 0.170 5.688

2,146 0.180 6.831

–0.157*** [0.037] 0.022 [0.056] 0.060 [0.053]

Distance to border

Time trends: Year = 1996

(2)

–0.001 [0.228] –0.446** [0.213] 7.275*** [0.186] 2,146 0.177 9.896

Source: Authors’ calculations. Notes: Dependent variable for all columns is log of village population size; column (3) is our preferred specification. Distance measures are in logs. Proximity to nearest river is straightline distance, to capture flight time needed by the black flies that carry Onchocerciasis from the river to people’s homes. In contrast, proximity to Côte d’Ivoire is travel distance by roads, train, or footpath. The regression also controls for forty-five province dummies (not shown). Robust standard errors are in brackets. ***Significant at the 1 percent level. **Significant at the 5 percent level. *Significant at the 10 percent level.

Demographic Pressure and Institutional Change

119

Repatriation from Côte d’Ivoire occurred later, as shown in columns (2) and (3), where villages further from the border have smaller populations than others in 1996 and 2006. Thus, our preferred first stage (column [3]) has as its significant excluded instruments distance to rivers (in all years) and distance to the border (in 1996 and 2006). Beneath each column, we provide an F-statistic on the joint significance of all excluded instruments. The F-statistic levels indicate that in each case, the null hypothesis that the instruments are jointly irrelevant in the regression can be rejected at the 1 percent level. The F-statistics are, however, smaller than the rule of thumb cut-off suggested by Stock and Yogo (2005), implying that our second-stage estimations may suffer from weak identification in these regressions. Future work could focus on the regions of Burkina Faso where Onchocerciasis control and repatriation from Côte d’Ivoire was concentrated, to strengthen the identification strategy. Table 4.6A reports the instrumental variable (IV) estimates for our preferred specification. In each column, we report the Hansen J statistics and the associated probability. In columns (1), (2), (3), (4), (6), (9), (10), and (17), we cannot reject the null hypothesis that the instruments are wrongly excluded from the second-stage regression. Results should be interpreted with caution, but it is notable that the IV results are stronger than the OLS estimates in table 4.3, with larger estimated coefficients and greater statistical significance. Variation in a village’s population that is linked to being near rivers and to Côte d’Ivoire is positively associated with having more public services, infrastructure, religious facilities, and markets, as well as more individual land rights (as opposed to familial or communal), more land rental or sale transactions, and regulated access to forestland. In tables 4.6B and 4.6C, we test these relationships separately using each of the two kinds of instruments. Table 4.6B shows the IV estimations using only distance to the nearest river, and table 4.6C shows IV results using only distance to the border of Côte d’Ivoire. Both sources of identification produce qualitatively similar results, with somewhat larger point estimates when population is instrumented by distance to the border. The identification is, however, stronger when we use distance to nearest river in table 4.6D. The F-statistic in the first stage is 9.9, and we cannot reject the null hypothesis that the exclusion restrictions do not hold for column (10) only. This contrasts with table 4.6C, where columns (1), (5), (6), (9), (10), (12), and (13) do not pass the overidentification test. The main results presented in table 4.6A use province fixed effects to control for variation in political and economic circumstances across the country’s forty-five administrative regions. This leaves unobserved heterogeneity among villages within each province, and table 4.6D shows results when village fixed effects are used. Two relationships survive these controls: villages with above-trend population increases gain closer proximity to markets and become less likely to use communal property rights over land. The

2,137 17.03 0.00444

0.574*** [0.202] 0.127** [0.059] 0.603*** [0.074]

Infrastr. (2)

2,146 7.879 0.163

Observations Hansen J stat. Prob. HJS

2,112 5.231 0.388

0.649*** [0.239] –0.071 [0.068] 0.173** [0.087]

Markets (4)

2,146 3.175 0.673

–0.246*** [0.082] 0.024 [0.025] –0.049* [0.029]

Family (14)

Land ownership rights

2,016 9.398 0.0942

1.165*** [0.235] 0.055 [0.075] 0.737*** [0.100]

Religion (3)

2,137 18.42 0.00247

0.780*** [0.187] 0.134** [0.054] 0.694*** [0.069]

Infrastr. (6)

2,146 2.912 0.714

0.097* [0.050] –0.009 [0.016] 0.025 [0.019]

Communal (15)

2,108 21.42 0.000674

0.410** [0.166] –0.004 [0.050] 0.331*** [0.058]

Services (5)

2,146 1.780 0.879

0.061* [0.033] 0.004 [0.010] 0.064*** [0.014]

Land markets (16)

2,016 6.483 0.262

0.989*** [0.199] 0.106* [0.064] 0.721*** [0.085]

Religion (7)

Proximity to all sources (km)

2,112 5.490 0.359

2,108 14.14 0.0148

0.953*** [0.246] 0.119* [0.069] 1.286*** [0.088]

Services (9)

2,146 25.70 0.000102

0.102 [0.070] 0.028 [0.020] 0.110*** [0.023]

Pasture (17)

0.718*** [0.215] –0.011 [0.060] 0.340*** [0.079]

Markets (8)

2,016 0.170 0.999

0.629*** [0.174] 0.185*** [0.056] 0.621*** [0.070]

Religion (11)

2,146 2.662 0.752

0.222* [0.125] 0.066* [0.035] 0.210*** [0.045]

Forest (18)

Regulated access

2,137 20.33 0.00108

0.303* [0.171] 0.285*** [0.050] 0.643*** [0.059]

Infrastr. (10)

2,112 5.693 0.337

0.933*** [0.211] 0.135** [0.061] 0.791*** [0.080]

Markets (12)

2,146 1.781 0.878

–0.231 [0.151] 0.025 [0.042] –0.015 [0.052]

Cropland (19)

Proximity to closest source (km)

Source: Authors’ calculations. Notes: First-stage results are shown in column (3) of table 4.5. Population and proximity measures are in logs. All regressions control for forty-five province dummies (not shown). Robust standard errors are in brackets. ***Significant at the 1 percent level. **Significant at the 5 percent level. *Significant at the 10 percent level. Notes: First stage results are shown in Column 3 of Table 5. Population and proximity measures are in logs. All regressions control for 45 province dummies (not shown). Robust standard errors in brackets, and asterisks indicate significance levels at *** p 0 (i.e., after the reform date of February 2004), and zero otherwise. This simple specification assumes that the trend in matatu accident rates would have been the same as the evolution of private vehicle accidents but for the adoption of the Michuki reforms. A negative and significant value of δ would indicate a reduction in claims associated with the new rules. Specification (1) above can be extended to a multiperiod estimation simply by specifying Post as a set of indicators corresponding to each month of the four-year window around the reforms. Estimating specification (2) below allows one to examine temporal effects of reforms as well as adjust for differential trends. (2)

yijt = u0 + M j + ∑dτM j Dt=τ + ∑lτDt=τ + ´ijt, τ

τ

where we represent each month as an indicator variable Dt equal to 1 in that particular month t, and 0 otherwise. The set of coefficients δt now define the time profile of the impact of the reforms for all time periods after the reform date. This specification allows us to examine the evolution of the reform’s impact over time, including any persistence or waning of its effects. As with specification (1), the implicit counterfactual is that accident rates of matatus have the same trend as the private vehicles (defined by the family of parameters λτ) in the absence of the reform. With multiple periods, we

316

James Habyarimana and William Jack

can relax this assumption by including vehicle category specific trends as in specification (3) below. (3)

yijt = u0 + M j + ∑dτM j Dt=τ + fT + kM jT + ´ijt, τ

where ϕ and ϕ + κ define the vehicle category specific linear time trends.11 8.3.2

Heckle and Chide: Summary Statistics and Empirical Strategy

We recruited a total of 2,276 long-distance matatus in Nairobi and a number of regional centers, adopting field-based randomization to treatment and control status using the last digit of the vehicle’s license plate.12 Claims data were provided by insurance companies that at the time together covered about 90 percent of matatus in the country. Measurement error could arise due to selective reporting of accidents, although we do not believe this would have been associated with treatment status. (See table 8.2.) In assessing the impact of the passenger empowerment intervention, we adopt a similar statistical methodology to that employed in evaluating the Michuki rules, this time comparing matatus assigned to treatment and control groups before and after the assignment of stickers. As discussed in Habyarimana and Jack (2011), compliance with the random assignment was high, but not perfect, so we report intent-to-treat estimates and instrumental variable estimates using assignment status as an exogenous instrument. We also allow for different trends in accident rates for treatment and control vehicles before the intervention. Our first specification is the analog of equation (1) above: (4)

yit = a + bTi + gPostt + dTi × Postt + fX i + hi + ´it ,

where Ti is a dummy equal to 1 if vehicle i was assigned to the treatment group, and Postt = 1 for observations that occur after the intervention.13 Because we collected survey data on individual matatus and their drivers, we include additional controls, Xi, and a matatu credit cooperative fixed effect ηj. To allow for differences in trends between treatment and control groups and potential seasonality, we augment equation (4) to the following form: (5)

yit = a + bTi + gPost t + dTi × Post t + fX i + ∑ τ (c0τ + c1τTi ) × Qt=τ + st + hi + ´it ,

where Qt is an indicator for the quarter corresponding to time t. 11. Nonlinear time trends can be estimated by including higher-order terms of the time variable T. 12. Those with odd last digits were assigned to receive the stickers, and those with even last digits were assigned to the control group. Informed consent was received from all participating vehicles. 13. Thus in this specification we collapse all the data into just two periods.

Table 8.2

Selected vehicle and driver characteristics by random assignment Control

Treatment

356,506 (7,236) [327,266] 14.52 (0.05) 0.45 (0.02) 20.19 (0.36) 420.48 (6.14) [400] 1.00 (0.001) 0.49 (0.02) 0.004 (0.002) 0.061 (.008) 1.02

361,386 (6,350) [343,603] 14.52 (0.05) 0.48 (0.01) 19.60 (0.30) 414.10 (5.33) [400] 1.00 (0.001) 0.51 (0.01) 0.015 (0.004) 0.071 (.007)

Difference P-value

A. Vehicle characteristics Odometer reading

Seating capacity Proportion use tout Number of weekly trips Average daily distance, kilometers

Proportion with an installed speed governor Share owned by large cooperative (> 300 vehicles) Involved in accident in last 12 months, self-reported Insurance claim filed in last 12 months before recruitment, (from administrative data) F-stat and p-value of joint test of significance of all vehicle characteristics Number of observations

1,006

.612 [.288] .995 .087 .211 .433

.373 .419 .008 .355 .415

1,155

B. Driver characteristics Has access to phonea Owns a phonea Percent less than 30 years old Percent 30–40 years old Percent primary schooling Percent secondary schooling Percent married Number of children Proportion drivers assigned to one car only Proportion drivers started after recruitment Median driver tenure, days F-stat and p-value of joint test of significance of all driver characteristics Number of observations

0.96 (0.01) 0.89 (0.01) 18.5 (3.4) 54.8 (4.3) 22.8 (3.5) 13.9 (2.8) 74.8 (3.7) 2.0 (0.1) 0.72 (0.04) 0.37 (0.04) 296 0.39

0.98 (0.00) 0.91 (0.01) 16.2 (3.0) 56.1 (4.1) 26.2 (3.5) 14.7 (2.8) 77.0 (3.5) 2.0 (0.1) 0.70 (0.04) 0.41 (0.04) 305.5

139

145

.052 .135 .612 .831 .494 .842 .665 .918 .649 .515 .89 0.95

Notes: Standard errors are in parentheses; medians are in brackets. The table presents mean/median of vehicle characteristics by treatment assignment. The sample is restricted to matatus for which information on random assignment is available. The 115 matatus that could not be matched to the initial assignment list are dropped. a Statistics reported in these rows are based on the sample of all recruited matatus. The statistics reported in panel B of the table are based on a random sample of 284 matatu drivers who were surveyed six months after recruitment.

318

8.4

James Habyarimana and William Jack

Results

8.4.1

Michuki Rules

The simple difference-in-differences estimate of the impact of the Michuki rules is shown in table 8.3, where the data are aggregated over a period of twenty-four months before the reforms and twenty-four months after. These results suggest that there was no sustained reduction in accidents involving matatus relative to private vehicles over the long term. This analysis shows there was no discernible impact of the reforms over the long term. While the point estimate (–.003) is negative and economically large, it is statistically indistinguishable from the reform having had no effect on accident rates. To assess the temporal impacts of the reforms, we estimate equation (3) and illustrate the results in figure 8.5. The horizontal axis measures months before or after the introduction of the reforms, with t = 0 representing February 2004. The solid line represents the values of the month-level coefficients δt, which can be interpreted as measuring the differential likelihood of an insurance claim by a matatu compared with that of private vehicles, assuming a common trend for t < 0. The dashed lines show the 95 percent confidence interval around the estimated effects. There is a significant negative impact of about 2 percentage points on the likelihood of a matatu claim at t = 0. For t > 0 the estimates are mostly negative, but they are statistically insignificant. Figure 8.6 illustrates the same information, but allows for differential trends in matatu and private vehicle claims rates before the reform. If anything, these results indicate more emphatically that there was no discernible effect of the reforms. Apart from the impact at t < 0, the only significant coefficient occurs more than a year after their implementation, and is positive. We interpret the sharp fall in claims likelihood during the first month of the reforms as deriving from matatus being pulled off the road in order to be fitted with the required equipment. Since we do not have data on miles traveled or days of active operation, we cannot say for sure if this is the case,

Table 8.3

Private Matatus Total

The 2 × 2 difference-in-differences estimate of Michuki reforms Before reform

After reform

0.001 (0.001) 0.016 (0.004) 0.015 (0.004)

0.001 (0.001) 0.013 (0.001) 0.012 (0.001)

After-before 0.000 (0.001) –0.003 (0.004) –0.003 (0.004)

Fig. 8.5 Differential likelihood of claim for matatus vis-á-vis other private vehicles, common time trends Note: The figure plots the monthXminibus_indicator coefficients from a regression of claims rates on a set of month dummies, a vehicle class indicator (the baseline category is private vehicles), and their interaction. The dashed lines show the upper and lower limits of the 95 percent confidence interval.

Fig. 8.6 Differential likelihood of claim for matatus vis-á-vis other private vehicles, class-specific time trends Note: The figure plots the monthXminibus_indicator coefficients from a regression of claims rates on a set of month dummies, a vehicle class indicator (the baseline is private vehicles), and their interaction. The dashed lines show the upper and lower limits of the 95 percent confidence interval.

320

James Habyarimana and William Jack

Fig. 8.7 Differential likelihood of claim involving an injury or death for matatus vis-á-vis other private vehicles, class-specific time trends Note: The figure plots the monthXminibus_indicator coefficients from a regression of injury or death-related claims rates on a set of month dummies, a vehicle class indicator (the baseline is private vehicles), and their interaction. The dashed lines show the upper and lower limits of the 95 percent confidence interval.

but the anecdotal evidence reported above is consistent with a large reduction in the volume of matatu traffic in that period. We also report the differential likelihood of matatus, compared with private vehicles, making claims that involve an injury or death, as shown in figure 8.7 (again, allowing for differential trends). Although there is a negative coefficient at t = 0 once again, and while it is sustained until four months after the reform, there appears to have been an (imprecise) negative differential for some months leading up to February 2004 as well. This suggests that other factors could have been at work to reduce serious matatu accidents leading up to the reforms, although we have no specific evidence of such. From t = 5 onward, the differential likelihood of a claim is either zero or positive. Our results suggest that the Michuki rules had little if any effect on the safety of matatu travel in Kenya. Compliance with the new regulations disrupted transportation services in the early days of implementation considerably, but appears not to have reduced the likelihood that any operational vehicle would have an accident leading to the submission of an insurance claim.

State versus Consumer Regulation

Fig. 8.8

321

Quarterly claims data, treatment and control matatus

Note: The figure presents the number of insurance claims by quarter between January 1, 2006, and May 25, 2009. All insurance claims are used to construct this figure. Solid and dashed lines represent fitted linear trends for the treatment and control group. We fit a linear trend to all claims for the pretreatment period for the treatment group (all claims from 2006–2007). The dotted line traces out counterfactual claims for the treatment group. For the control group, we fit a linear trend to all claims from 2006–2008, excluding claims from quarters 1 and 2 of 2009 due to incompleteness. We make the simplifying assumptions that matatus continue to operate after a claim event and were operating throughout this period.

8.4.2

Heckle and Chide

The effects of the consumer empowerment intervention are both more reliably estimated, due to the randomized assignment of treatment, and demonstrably larger than the impact of the regulatory reform. Figure 8.8 (from Habyarimana and Jack 2011) reports quarterly data by random assignment from the first quarter of 2006 to the second quarter of 2009 (the intervention was implemented in Q1, 2008). Each point in the graph represents the number of claims per 1,000 insured matatus, and the two lines are trend lines (estimated using preintervention data for the treatment group, and data through 2008 for the control). The data were collected in 2009, but the figures reported for that year are incomplete, given the considerable lag between a claims-related event and the digital recording of the associated claim. Prior to the intervention, claims rates exhibited an upward trend, and showed no discernible differences between vehicles assigned to the control

322

James Habyarimana and William Jack

Table 8.4

Regression results All claims (1)

Post: γ Assigned to treatment: β PostXAssigned to treatment Constant: α Percentage effect: Effect of treatment on the treated Percentage effect: Controls for SACCO Observations R-squared Mean post recruitment claims rate for vehicles assigned to control group First stage: F-stat

Driver-at-fault claims (2)

(3)

(4)

(5)

(6)

0.025 (0.011)* 0.011 (0.010) –0.046 (0.014)** 0.052 (0.007)** –52

0.026 (0.011)* 0.011 (0.010) –0.047 (0.014)** 0.039 (0.012)** –62

0.018 (0.009)+ 0.011 (0.008) –0.040 (0.012)** 0.038 (0.006)** –60

0.018 (0.009)* 0.011 (0.008) –0.041 (0.012)** 0.036 (0.010)** –63

B. IV estimates –0.075 –0.068 (0.023)** (0.021)** –93 –77 X

–0.069 (0.021)** –91 X

–0.059 (0.017)** –88

–0.060 (0.017)** –92 X

4,318 0.01

4,322 0.002

4,318 0.01

A Intent-to-treat 0.029 0.030 (0.013)* (0.012)* 0.010 0.009 (0.010) (0.011) –0.050 –0.051 (0.016)** (0.016)** 0.061 0.042 (0.008)** (0.013)** –50 –63 –0.073 (0.023)** –73

Injury/death claims

4,322 0.003

4,318 0.02

0.09 2,421.33

2,364.44

4,322 0.002 0.077

0.055

Notes: Table reports the estimates of ordinary least squares regression in specifications (1–4) and instrumental variables estimates in specifications (5–6). The dependent variable is the annualized rate of a claim-generating accident for each matatu in the sample. We make a simplifying assumption that matatus continue to operate after a claim event and were operating throughout the pre- and postrecruitment period. First-stage F-stat reports the F-stat of the test of the null that random assignment to treatment does not predict actual treatment status at recruitment. The sample excludes 3 percent of recruited vehicles for which treatment assignment information could not be reliably established. Robust standard errors in parentheses. **Significant at the 1 percent level. *Significant at the 5 percent level.

and treatment groups.14 However after the intervention, claims rates were consistently lower for vehicles assigned to the treatment group. Panel A of table 8.4 reports intent-to-treat (ITT) estimates for equation (3), collapsing all the data into two periods—before and after the intervention.15 In the first two columns, the probability of any claim falls by between 50 and 63 percent (from projected annual rates of 10 and 8.1 percent, respectively). Similar reductions are observed for claims in which the matatu driver was at fault (columns [3] and [4]), and for claims involving an injury or death (columns [5] and [6]). 14. A similar upward trend was underway for larger thirty to forty-one-seater buses (see Habyarimana and Jack 2011), suggesting a secular trend. 15. Columns (2), (4), and (6) include SACCO fixed effects. Most long-distance matatus are organized in SACCOs, Savings and Credit Cooperatives, of which there were twenty-one in our study.

State versus Consumer Regulation

323

Panel B reports instrumental variable results, where we use assignment as an instrument for treatment status. The estimates of the impact of the treatment on the treated, or local average treatment effect, are effectively scaled by the inverse of the compliance rate, and are correspondingly larger than the ITT estimates. According to these estimates, among those vehicles that were induced to accept stickers by being assigned to the treatment group, claims were close to eliminated, falling by between 73 and 93 percent. Similar results are obtained using the specification with more temporal structure in equation (5), and can be found in Habyarimana and Jack (2011). 8.5

Evidence on Mechanisms of Change Underlying Heckle and Chide

Although the consumer empowerment intervention appears to have sizable impacts on accidents and insurance claims, exactly what mechanisms underlie these large effects is unclear. One concern is that drivers of treated matatus misunderstood the lottery as being a prize for safe driving, and that they responded to the perceived financial incentive. Since the design did not include a placebo intervention, it is impossible to know whether this is the case. And even if we are confident that the lottery did not drive our results, we do not know if the stickers worked by inducing passenger complaints, or if they affected the driver directly (he was aware of them, even if they were not in view as he drove). In this section we present some suggestive evidence for potential mechanisms that underlie the reduction in accident rates estimated above. Although we do not have the data to definitively discriminate among all plausible mechanisms that could underlie our results, we nonetheless present two pieces of evidence in support of passenger-action mechanisms, and discuss the plausibility of a number of other mechanisms including direct effects on drivers, ex post sorting of drivers, and the effects of the lottery. 8.5.1

Survey Evidence

The obvious mechanism by which the intervention leads to improved safety is that the stickers empower passengers to voice their concerns over bad driving and that the resulting social pressure changes the behavior of the driver. To investigate this we analyze data from a survey fielded in November 2008 of drivers, plus up to three passengers per vehicle. A total of 284 vehicles were sampled for this survey.16 We face two difficulties in detecting evidence for this mechanism. First, even if the stickers are effective in empowering passengers, we might observe little or no difference in heckling if drivers of treatment vehicles quickly learn to adapt their behavior to minimize passenger complaints. On the 16. We interviewed 306 drivers, but twenty-two of them were operating vehicles that had not been recruited earlier.

324 Table 8.5

James Habyarimana and William Jack Sticker retention

Number of stickers in vehicle 0 1 2 3 4 5 Total

Distribution at recruitment (%) (1)

Distribution in November 2008 (%) (2)

46.5 2.1 2.8 4.2 0.3 44.0 100.0

63.0 4.9 4.2 7.4 2.5 18.0 100.0

Note: Table reports the distribution of stickers for the random sample of matatus surveyed eight months after recruitment. Column (1) reports the distribution at recruitment while column (2) reports the distribution eight months after recruitment.

other hand, whether heckling is observed in equilibrium or not, we might expect passengers to report their trips as being safer in treatment matatus. Second, given the rarity of traffic accidents, events that generate heckling will also be rare. Compounding this power problem is the fact that, despite the weekly lottery, after eight months a considerable number of the treatment vehicles had lost some or all of their stickers. Table 8.5 shows that among our sample of 284 matatus the share with all five stickers had fallen from 44 percent at recruitment to 18 percent eight months later, and the share with at least one sticker had fallen from 53 percent to 37 percent. Table 8.6 presents evidence of heckling from the survey of drivers (panel A) and passengers (panel B) and passenger-reported safety ratings (panel C). We present intent-to-treat estimates for all outcome measures. Note that this considerably limits our ability to find any evidence for this mechanism as a result of low sticker retention. The results are suggestive of passenger heckling as one of several potential contributors to the reduction in accident rates. In rows 1 and 2 of panel A, we estimate the effect of assignment on the likelihood that the driver reports passenger heckling in the past week and most recent trip. The point estimate in row 1 has the right sign, but is imprecisely estimated. The sign of the coefficient in row 2 is wrong, but again imprecise. However, in OLS results not reported here, we find substantial and marginally significant effects of having a sticker eight months into the study. In particular, drivers of vehicles with stickers at the time of the survey were about three times more likely to report passenger heckling.17 17. In a simple OLS estimation of the effect of stickers on heckling, nonrandom removal or depreciation of stickers could bias our results. On the one hand, dangerous drivers might have removed them, either in advance or in response to unwelcome heckling as they learned about their effectiveness over time. This would work against finding evidence of passenger action in treated vehicles. On the other hand, if the stickers provided drivers who otherwise lacked self-

State versus Consumer Regulation Table 8.6

Dependent variable

Evidence on passenger action mechanisms Assigned to treatment

Unsafe trip

Assigned to treatments * unsafe Trip

Number of observations

—

259

—

259

–0.022 (0.097) –0.043 (0.078) –0.092 (0.145) 0.031 (0.093)

788

A. Driver reports of heckling (1) Driver reports heckling (past week) 0.027 — (0.034) (2) Driver reports heckling (last trip) –0.027 — (0.027) B. Passenger reports (most recent trip) (1) Any passenger expressed concern –0.005 0.172 (0.088) (0.070)* (2) Respondent expressed concern 0.014 0.084 (0.071) (0.064) (3) At least two respondents expressed concern 0.092 0.300 (0.130) (0.101)** (4) All three respondents expressed concern –0.058 0.081 (0.079) (0.077) (1) Safety rating

325

C. Passenger perceptions of safety (most recent trip) –0.007 — (0.078)

—

788 260 260

788

Notes: Panel A reports the results of a linear probability model on the likelihood of drivers reporting heckling in the past week and on the most recent trip. Panel B reports the results of an OLS regression of the likelihood of passengers reporting expressions of concern to driver/conductor on treatment assignment status, safety rating, and the interaction of the two variables. A sample of up to three passengers exiting each matatu surveyed eight months after recruitment is used to construct these estimates. Passengers from twenty-two matatus that could not be matched to the assignment lists are dropped, leaving a total of 788 passengers (see below on the coding of unsafe). Panel C reports the results of an ordered probit model on passenger perceptions of safety. Passengers were asked to rate the safety of the justcompleted trip on a scale from 1 to 10, where 1 implies no danger, 10 implies high likelihood of serious injury/death, and 55 corresponds to “cannot say.” A trip is considered safe or unsafe if at least one respondent reports a safety rating of 6 or higher. We recode this variable as follows: 1 = Safe ( a rating 1–5), 2 = Cannot Say (55), and 3 = Dangerous (a rating 6–10). Robust standard errors are in parentheses. **Significant at the 1 percent level. *Significant at the 5 percent level.

We next turn to self-reported evidence of passenger action in panel B of table 8.6. Sampled passengers were asked to report if they or any other passengers had said something to the driver/conductor about reckless driving behavior on the just-concluded trip. In order to avoid conflating potentially frivolous actions with legitimate heckling, we control for the reported safety of the trip. In particular, passengers were asked to rank the safety of the control with an effective enforcement technology, removal could be concentrated in the pool of relatively safe drivers, who simply find them distasteful and perhaps bad for business. This would bias our results in favor of finding an effect. Although we cannot distinguish econometrically between these two directions of bias, we believe the former is more plausible and highly likely to dominate the latter.

326

James Habyarimana and William Jack

trip on a scale of 1 to 10, with 1 denoting no danger and 10 denoting lifethreatening danger. While nearly 45 percent of the respondents reported that they “could not say,” we define a trip to be reported as safe if the passenger reports a rating equal to or less than 5. For our current purposes we create an indicator for whether at least one passenger had rated the safety of the trip as dangerous (a rating of 6 or higher). Evidence for the passenger heckling mechanism is then captured by the extent to which there is a greater likelihood of heckling on trips deemed dangerous by at least one passenger. We present ITT estimates for four different outcomes that correspond to the rows in panel B of table 8.6: likelihood of heckling by (a) the respondent, (b) any passenger, (c) at least two respondents, and (d) all respondents. The latter two outcomes represent a crude measure of the extent to which the intervention facilitates collective passenger action and the unit of observation is the vehicle.18 The coefficient of interest is the interaction of the indicator for stickers and whether at least one passenger rated the trip as unsafe. Our estimates for this parameter are of the wrong sign in rows 1–3, but in all cases are very imprecise. In row 4, that estimates the likelihood that all correspondents heckle the driver, we obtain the right sign but once again the coefficient is statistically insignificant. One way in which this mechanism could operate is by making passenger heckling a credible threat to reckless driving. In the absence of more objective measures of driving behavior, we rely on passenger ratings of safety of the just-concluded trip. Our results in panel C report the results of an ordered probit estimation across three safety ratings categories (safe trip, cannot say, unsafe). About two-thirds of all passengers in the control matatus rated the most recent trip as safe according to this definition. The ordered probit estimate in panel C has the right sign, but is very imprecisely estimated. While the evidence above suggests that passenger action may well lie at the heart of the observed effects, we cannot definitively rule out a number of other potential mechanisms. For instance, while passenger ratings of safety do not confirm this (see panel C of table 8.3), it is possible that a driver’s beliefs regarding the preferences of the vehicle’s owner, over either passenger safety or the life of the vehicle, could be affected by this intervention. More direct observations of driver behavior might shed more light on the plausibility of this mechanism. 8.5.2

Driver Sorting

Alternatively, although the ex ante assignment of stickers to drivers was random, the ex post assignment may have exhibited sorting. That is, it is possible that rather than stickers having altered the behind-the-wheel behavior of drivers, either directly or via passenger action, they induced sorting of 18. For two or more respondent reports of heckling we are unable to condition on the same dangerous event.

State versus Consumer Regulation

327

drivers across treatment and control matatus. For example, suppose reckless drivers in treated vehicles tended to switch to control vehicles, or to exit this labor market entirely, while safe drivers in control vehicles on average moved to treated matatus. Such sorting behavior could have led to the observed changes in claims rates, but would not have been associated with any change in driving practices per se. We present three pieces of evidence suggesting that this kind of ex post sorting does not constitute a likely explanation of the results. First, the share of treatment vehicles within each matatu cooperative (SACCO) is about half, so sorting within SACCOs is definitely feasible. However, the authority to hire and fire drivers rests not with the SACCO, but with the owners of the vehicles. But since matatu ownership is very diffuse, sorting within an individual owner’s fleet (which can be as small as one or two vehicles) is unlikely to generate our measured effects. And given the costs of sorting out-of-treatment vehicles, it would be much easier for the drivers to remove the stickers than to find an eligible and willing partner with whom to switch. Second, it is possible that this sorting operates more on the participation margin, if reckless drivers tend to quit the treatment group. Data on driver tenure suggests that the median tenure is about ten months and that while overall turnover since recruitment has been high (an average of 39 percent), there is no statistically significant difference in turnover rates across treatment and control vehicles (41 vs. 37 percent). This holds true among the drivers assigned to a single vehicle. And third, selective sorting could take place within just those SACCOs that have a policy of regularly rotating drivers across vehicles, as long as such rotation was nonrandom. However, our results could be driven by selective sorting among the relatively small group of drivers in such SACCOs only if there was a high concentration of claims among “reckless” drivers. The insurance claims data from the period before our intervention do not support this pattern. Although the identity of the driver is not recorded in the data, we do know that before our intervention, fewer than 8 percent of all claims were associated with multiple-claim vehicles (and possibly drivers). Overall, these three pieces of evidence suggest that while we cannot rule out driver sorting as a response to the intervention, the scale at which such sorting could be occurring cannot explain the results obtained above. 8.5.3

Direct Effects of the Lottery

Finally, we discuss the possibility that the presence of the lottery, designed to improve sticker retention, could itself lead to our empirical results. Recall that drivers who accepted all five stickers at recruitment were divided into five groups of roughly 200 vehicles, and that each week on a five-week rotating basis, members of one of the groups were eligible to win one of three prizes if, when randomly drawn, upon inspection they were found to have retained all five stickers. The total prize money each week of 10,000 shillings

328

James Habyarimana and William Jack

(about two weeks’ wages) was awarded in three amounts (5,000, 3,000, and 2,000) to three different winners. The lottery itself could have changed the beliefs of drivers of treatment vehicles about the likelihood and consequences of an accident. Alternatively, while the rules of the lottery were very explicit, and drivers were told that eligibility was based on sticker retention and not an accident-free record, it is still conceivable that drivers with stickers might have misconstrued the lottery as a reward for safe driving. The policy implications of such findings would, of course, be radically different to those that would otherwise be drawn. On the first point, knowledge of the lottery and its association with the road safety project were not confined to treatment vehicles alone or lottery nominees. Inspection of stickers was done at parking lots where control and treatment drivers interacted quite frequently, and where awareness of the role of the sticker inspector was clear to both. As a result, we believe that any small differences in road safety salience attributable to the lottery across the two groups is unlikely to explain the large effect measured above. On the second point, which is potentially of greater concern, the payment is likely to have been too small to alter driving behavior. Expected winnings were very low (equal to wages equivalent to about twenty minutes work), and even if drivers had unreasonable priors of winning, the first prize was considerably less than what a driver could make by squeezing in one extra trip (unreported to owner) per month. Nonetheless, to address this second issue more quantitatively, we investigate the beliefs that drivers would have had to maintain in order that the observed reduction in claims rates could be rationalized in terms of a response to the misguided belief that safe driving would increase the chance of winning the lottery. This kind of exercise is, of course, laden with assumptions and can only inform the analysis if the results suggest wildly counterfactual driver beliefs. In fact, we find that such extreme beliefs, plus an impossibly high response of accidents to speed reductions, are indeed necessary to support the claim that the lottery was the driving force behind the impact we observe. The key parameter in this exercise is the elasticity of accidents with respect to speed, estimates of which are not available in Kenya or other developing countries to our knowledge. Ashenfelter and Greenstone (2004) report data for the United States suggesting an elasticity of fatalities of about four, which provides a benchmark against which to compare our data.19 As we illustrate below,20 even if a driver (a) thought he would win the 19. The approximately equal estimated proportional impacts of the intervention on all claims, claims in which the driver was at fault, and claims involving an injury or death suggest that this fatality elasticity is a good proxy for the elasticity of all accidents. 20. Each week three prizes totaling 10,000 shillings were awarded. We assume driver risk neutrality and denote the size of the average weekly prize by x = 10,000/3. The probability the driver assesses to winning a prize, conditional on not having had an accident, is denoted p, and

State versus Consumer Regulation

329

lottery with certainty (instead of with average weekly probability 0.003), (b) was sure of reducing his chance of an accident to zero, and (c) thought that there was a single prize of 10,000 shillings every week, the elasticity of accidents with respect to speed would still need to be more than thirty times larger that the US estimate for the expected financial benefit of slowing down to outweigh the expected costs. In light of the evidence, recently reviewed by Delavande, Gine, and McKenzie (2009), that people in developing countries generally understand the concept of probability, we believe this calculation, while clearly subject to wide margins o]f error, nonetheless strongly suggests the lottery itself did not affect driver behavior enough to account for any meaningful share of the estimated effects of the intervention. In ongoing work we attempt to explicitly address the concerns about mechanisms voiced above. In particular, in a new study of more than 10,000 matatus we include a placebo arm in which vehicles are assigned stickers that say simply “Travel well,” while remaining eligible for the lottery. In addition, we send enumerators on up to 7,000 trips, during which they monitor driving behavior and passenger responses directly. 8.5

Conclusions

We present evidence that tough government regulations were unsuccessful in inducing sustained changes in accident rates of minibuses in Kenya, the expected winning each week are px. Let w be the driver’s weekly wage, and denote z = x/w as the ratio of the average prize to the wage. In order to reduce the chance of being involved in an accident, thereby increasing his chance of winning, the (misinformed) driver slows down. We want to compare the expected increase in winnings to the cost this would impose on him. Let π0 be the weekly probability of having an accident under the assumption of no behavior change. (The projected counterfactual annual accident rate among treated matatus during the year following the intervention was approximately 10 percent, so π0 = 0.1/52.) Drivers in the treatment group experienced a claims rate about half the projected rate. Assuming for the sake or simplicity a constant proportional reduction over the year, their actual weekly probability of having an accident was π1 = π0 /2, which is also the change in the probability, ∆π. Engaging in this behavior change increases expected weekly winnings by B = ∆πpx = ∆πpzw. The expected cost per week of slowing down is the wage times the extra time taken, w∆t, which is approximately equal to C = w(∆s/s), where s is the average speed of the vehicle. Define the elasticity of accidents, a, with respect to speed, s, by ε = [(∆a/a)]/[(∆s/s)]. Although the relationship between speed and accident rates in Kenya is not known, Ashenfelter and Greenstone (2004) present fatality and speed data from the United States that suggest an elasticity of fatalities with respect to speed of about 4. (In their data, a 4.55 percent reduction in speed is associated with a 15.46 percent reduction in fatalities.) Thus the cost incurred by the driver in reducing accidents by this much is approximately C = w(αε) where α = ∆a/a ≈ 1/2.. This cost is less than the expected benefit 1 < αε∆πpz. Using our data, the right-hand side of this expression is approximately 2 × 4 × (((0.05)/(52))) × ((3/(1,000))) × ((2/3)) = 1/65,000. That is, for a driver to respond only to the incentive of a lottery whose eligibility criteria he misinterpreted, and not to the stickers or the response they evoked on the part of passengers, he would need to overestimate the right-hand side of the inequality condition above by a factor of 65,000. Even if he thought he would win the lottery with certainty (p = 1), was sure of reducing his chance of an accident to zero (∆p = 0.1/52), and thought that there was a single prize of 10,000 shillings (z = 2), the elasticity of accidents with respect to speed would still need to be 32.5 times larger that the US estimate for the condition above to be satisfied.

330

James Habyarimana and William Jack

despite strong political leadership and dedicated resources. On the other hand, an intervention that relied on no third-party enforcement, and whose implementation was in fact unknown to and unsupported by the government or the police, appears to have been remarkably successful in bringing down accidents rates, by at least one-half. We do not have enough information to explain why the Michuki reforms had little effect. Although there was an initial dip in the number of insurance claims involving vehicles subject to the regulations, we argue that most of this was due to the need to comply with new hardware requirements, which took a large number of minibuses off the road. Once the necessary vehicle modifications had been made, it appears that claims rates returned quickly to their prereform levels. It would be unhelpful to claim that all regulation is doomed to failure. Instead, our analysis suggests that in institutionally weak environments, innovative consumer-driven solutions might provide an alternative solution to low-quality service provision.

References Ashenfelter, Orley, and Micahel Greenstone. 2004. “Using Mandated Speed Limits to Measure the Value of a Statistical Life.” Journal of Political Economy 112 (2, part2): S226–67. Chitere, Preston, and Thomas Kibua. 2004. “Efforts to Improve Road Safety in Kenya: Achievements and Limitations of Reforms in the Matatu Industry.” Institute of Policy Analysis and Research (IPAR), Kenya. http://www.ssatp.org/sites/ ssatp/files/publications/CountryDocuments/Road-Safety-Kenya-IPAR.pdf. Delavande, Adeline, Xavier Gine, and David McKenzie. 2009. ‘‘Measuring Subjective Expectations in Developing Countries: A Critical Review and New Evidence.’’ BREAD Working Paper no. 203, Bureau for Research and Economic Analysis of Development. http://ibread.org/bread/search/wpaper/delavande. Habyarimana, J., and W. Jack. 2011. “Heckle and Chide: Results of a Randomized Road Safety Intervention in Kenya.” Journal of Public Economics 95 (11–12): 1438–46. Lopez, A., Mathers, C., Ezzati, M., Jamison, D., and Murray, C. 2006. “Global and Regional Burden of Disease and Risk Factors, 2001: Systematic Analysis of Population Health Data.” Lancet 367:1747–57. Mathers, C., and D. Loncar. 2006. “Projections of Global Mortality and Burden of Disease from 2002 to 2030.” PloS Medicine 3 (11): 2011–30. Mutugi, Marion, and Samuel Maingi. 2011. “Disaster in Kenya: A Major Public Health Concern.” Journal of Public Health and Epidemiology 3 (1): 38–42. Odero W., Khayesi, M. and Heda, P. M. 2003. “Road Traffic Injuries in Kenya: Magnitude, Causes, and Status of Intervention.” Injury Control and Safety Promotion 10:53–61.

9

Fifteen Years On Household Incomes in South Africa Murray Leibbrandt and James Levinsohn

9.1

Introduction

Measuring South African economic growth since the fall of apartheid is a tricky business. One can simply measure gross domestic product (GDP) per capita and there the picture is a bright one. Real GDP per capita since the democratic elections in 1994 has risen an average of close to 1.5 percent per year. Individuals, though, can’t really spend GDP when they go to the store. Rather, they spend their incomes. One can instead measure individual incomes, but this measure too is problematic. Examining the distribution of individual incomes will typically not speak to the welfare of the roughly 40 percent of South Africans age eighteen and younger. In this chapter, we measure incomes at the household level (adjusting for household size). This measure encompasses all household members, even those not participating in the labor market, while still capturing a measure of economic welfare at the level of individuals. Our reasoning is that, to the extent that real household per capita incomes increase, households are generally economically better off in a narrow but well-defined sense. With real household per capita income as our metric, we measure economic growth in South Africa from 1993 to 2008. Our approach is a very microeconomic one. We rely on two nationally Murray Leibbrandt holds the NRF/DST Research Chair in Poverty and Inequality Research in the Southern Africa Labour and Development Research Unit at the University of Cape Town. James Levinsohn is the Charles Goodyear Professor of Global Affairs and professor of economics and management at Yale University and a research associate of the National Bureau of Economic Research. We thank the NBER Africa Project for funding. We are also grateful to Matthew Welch for outstanding assistance with the data. For acknowledgments, sources of research support, and disclosure of the authors’ material financial relationships, if any, please see http://www.nber .org/chapters/c13394.ack.

333

334

Murray Leibbrandt and James Levinsohn

representative surveys of individuals. Our data, though, do not make up a panel, as such longitudinal data simply do not exist over the time period under consideration. Rather, we have used nationally representative household surveys from 1993 and 2008 and meticulously matched definitions of incomes so that we are confident that the temporal comparisons are valid. Because we rely on microdata, we are able to both measure the changes in incomes and investigate what explains these changes. Additionally, we are able to examine changes throughout the entire distribution rather than focusing simply on a mean or median. We do so using relatively new nonparametric techniques augmented by more traditional parametric estimates. Whether the news is good or bad surely depends on one’s prior and the previous evidence is sufficiently diverse that it is hard to know just what constitutes a happy story. As noted above, the national income accounts tell a story of success. While the macroeconomy has shown robust growth over most of the past fifteen years, it has been a period of relatively little job growth, and unemployment has increased dramatically. Depending on the measure used, unemployment has increased from around 15 percent to well over 30 percent. Exactly how the growth in GDP together with the rise in unemployment has impacted households is something of an open question. In an earlier paper, we documented that the first five years after the new government (from 1995 to 2000) saw real individual incomes decline almost 40 percent (see Leibbrandt, Levinsohn, and McCrary 2010). Hoogeveen and Ozler (2005) also found that the first five years after transition were especially tough on the poor as poverty increased and household expenditures at the lower end of the distribution fell in real terms. With the recent release of a new nationally representative income survey, the dismal but provisional picture painted by these earlier studies merits revisiting.1 In the next section, we describe our data. Section 9.3 describes the changes in real incomes from 1993 to 2008. Section 9.4 investigates what underlies these changes, while section 9.5 concludes. 9.2 9.2.1

The Data The 1993 Data

We benchmark incomes at transition using the Living Standards Measurement Study (LSMS) household survey conducted by the World Bank in 1993. This survey is well vetted and has been used by many researchers, including Case and Deaton (1998), Duflo (2003), and Thomas (1996). The survey was nationally representative and included about 44,000 individuals 1. Using much of the same data that we employ in this chapter, Leibbrandt, Woolard et al. (2010) examined the changes in inequality and poverty from 1993 to 2008. They found that inequality had increased while aggregate poverty had declined slightly. The authors did find some positive trends in indicators of nonmonetary well-being (e.g., access to piped water, electricity, and formal housing).

Fifteen Years On: Household Incomes in South Africa

335

comprising just over 8,800 households. One reason for this survey’s widespread use is that it serves as a benchmark for what South Africa looked like on the eve of transition. Also, this survey has not been subject to some of the criticisms leveled at a plausible substitute survey, the 1995 Income and Expenditure Survey. We have elected to simply bypass that issue by using the 1993 LSMS survey. Rather than using the widely available and easily downloaded merged version of the 1993 data, we have gone back to the original source data. We have done so because we want to be confident that our comparisons to 2008 are valid. This means making sure that every component of income is comparably defined in each of the two surveys—something we have taken great care to do. For this reason, we do not include imputed housing in our measurement of imputed income. Especially for poorer households, the value of housing can represent a substantial fraction of real income. Most households do not report the value of the flow of housing they receive from their residence when they own it. It is of course possible to impute the value of housing, and indeed one of us was responsible for this task for the current National Income Dynamics Survey. If we could be confident that the housing imputation used in 2008 (which we designed) could be applied to the 1993 data to construct a housing value that would then be comparable to that used in 2008, we would do so. Because we are not able to do this, we strip housing out of our income measures for 2003 and 2008. We note, though, that if we include housing using the probably noncomparable definitions from 1993 and 2008, we find larger increases in household per capita income. 9.2.2

The 2008 Data

The most recent nationally representative income data for South Africa come from the first wave of the National Income Dynamics Survey (NIDS.) This survey, like the 1993 survey, is publicly available, free, and readily downloadable.2 We use Wave 1 of the NIDS. These data were collected in 2008 and comprise the initial wave of what will be a national panel study. As was the case with the 1993 data, we use the original source data and then construct aggregates so as to ensure comparability with the 1993 data. The 2008 NIDS includes data on 28,225 individuals comprising 7,305 households. As noted above, we exclude the value of housing from our definition of income. The data include detailed expenditure data as well as income data. We focus in this chapter on the latter. 9.2.3

Why Log Household Per Capita Income?

Throughout this chapter, our analysis is focused on what happened to incomes at the household level. We have made this decision for a couple of 2. See http://www.datafirst.uct.ac.za/home/index.php?/Metadata-and-Data-Downloads to obtain the source data.

336 Table 9.1

Murray Leibbrandt and James Levinsohn Households in South Africa Race African Coloured Asian White Missing race Total

1993

2008

6,057,916 658,717 228,238 1,551,149 0 8,496,020

10,436,201 1,146,969 334,613 1,717,498 87,637 13,722,918

Note: Unit of observation is the self-reported household. Rates are calculated using sample weights.

reasons. First, we are trying to capture what happened to economic welfare at a national level using microdata. The obvious alternative to a householdlevel analysis is an individual-level analysis. An individual-level analysis has some advantages. It allows the researcher to investigate issues of income recipiency and to better understand how the labor market works (or not). The drawbacks to the individual-level analysis are that it excludes children (who comprise almost half the population) and, if one elects to work with log incomes as is often done, the analysis excludes all those adults who did not receive income. The recipiency issue can be addressed with careful econometric analysis, but the exclusion of children is part and parcel of any analysis of individual incomes. Because we want to better understand economic welfare at the national level, we elect the household-level approach and hence include children. The household-level approach has the advantage of making moot most issues around recipiency. Almost all households report at least some income, be it from remittances, grants, labor market earnings, or informal activities. We have elected to work with per capita household incomes so as to adjust for household size. This has the obvious advantage in that it corrects for household size, but it is a somewhat blunt way of dealing with changes in household composition over the course of the fifteen years between the surveys. We have not employed equivalence scales and, in the analysis below, simply treat all household members equally. Table 9.1 reports the number of households (after applying frequency weights) in South Africa in 1993 and 2008. From 1993 to 2008, there was an increase of about 5.2 million households in South Africa with about 4.4 million of those self-identifying as “African.”3 Over this same period, household composition changed. Table 9.2 gives average household size by year and by population group. For all groups 3. We use the population group names, African, Coloured, Asian, and White to maintain consistency with the existing literature despite the fact that all South Africans are, in another sense, African.

Fifteen Years On: Household Incomes in South Africa Table 9.2

337

Mean Household Size Race

1993

2008

African

5.30 (3.56) 4.90 (2.29) 4.50 (1.80) 3.25 (1.57) 4.87 (3.25)

3.68 (2.69) 3.76 (2.16) 3.72 (2.22) 2.62 (1.23) 3.55 (2.52)

Coloured Indian White Total Note: Standard errors are in parentheses.

the mean household size declined, with the most marked decline being for African households. In only fifteen years, mean household size declined from 5.30 to 3.68. This compositional shift underscores the importance of normalizing household incomes by some measure of household size when undertaking any intertemporal comparisons. Having decided to work with household per capita incomes, we elect to conduct most of our analysis looking at log incomes. This has the advantage of decreasing the influence of outliers (and there are a handful, especially in 2008) and of making our graphical analyses more practical. Both surveys provide sampling weights and all of our analysis employs those weights. 9.2.4

Household Incomes from 1993 to 2008

Table 9.3 reports descriptive statistics on the distribution of per capita household real incomes (hereafter “incomes” for the sake of expositional ease) in South Africa in 1993 and 2008. Focusing first on the top line of the table, the mean per capita income in 1993 was ZAR 10,741.4 In 2008, the comparable figure was ZAR 24,409. On the surface, this appears to be an impressive increase in real incomes at the household level. Not surprisingly, given the inequality documented by other researchers, these mean figures hide huge heterogeneity in household welfare—both within and across population groups. The average African income increased from ZAR 6,018 in 1993 to ZAR 9,718 in 2008, and for Coloured households the increase was from ZAR 7,498 to ZAR 25,269—an almost fourfold increase. For Whites, the increase was similarly dramatic, from ZAR 29,372 to ZAR 110,195. The standard deviation of incomes is reported in line 2 of the table and, for the population overall, this increased about tenfold. The within-race inequality documented in Leibbrandt, Woolard (2010) is evident in our data 4. In 2000, the exchange rate fluctuated mostly in the range of 6.5 to 7.5 rand per US dollar.

6,018 7,903

0 328 621 1,288 3,098 7,651 15,799 21,155 34,241 6,057,916

0 230 644 1,581 4,444 12,558 27,145 40,773 86,521 8,496,020

African

10,741 17,994

All

0 0 966 2,439 5,172 9,254 16,374 24,633 41,538 658,717

7,498 8,599

Coloured

1993

0 0 1,191 4,680 8,607 17,161 26,713 39,686 62,210 228,238

12,555 12,423

Asian

Household incomes in South Africa: 1993 and 2008

Note: Unit of observation is per capita household total income.

Mean Std. dev. Percentiles 1st 5th 10th 25th 50th 75th 90th 95th 99th n

Table 9.3

0 0 1,159 9,606 21,316 38,844 62,468 88,582 154,007 1,551,149

29,372 31,483

White

62 510 834 1,911 5,096 14,982 35,672 59,496 189,826 13,635,281

24,409 172,733

All

87 510 764 1,663 3,757 9,690 23,279 35,222 76,440 10,436,201

9,718 23,869

African

62 764 1,276 2,354 5,809 11,466 23,063 40,768 485,035 1,146,969

25,269 145,855

Coloured

2008

80 80 1,558 4,179 11,833 26,754 57,330 164,346 207,025 334,613

27,423 43,051

Asian

306 4,000 6,803 14,619 28,801 55,495 101,920 190,490 3,967,236 1,717,498

110,195 453,800

White

Fifteen Years On: Household Incomes in South Africa

339

as well. For all population groups, the ratio of the mean income to its standard deviation increased from 1995 to 2008. The bottom panel of table 9.3 reports percentiles of the distribution of income both overall and by population group. The median incomes (50th percentile) show increases that are substantially more modest than those of mean incomes. While mean income for all South Africans rose about 130 percent from 1993 to 2008, the median income rose just 15 percent over the same period—from ZAR 4,444 to ZAR 5,096. Especially for Whites, the increases are being driven by a small number of very large incomes. Another “small numbers” issue with the data concerns zero incomes that are reported in 1993. It is somewhat hard to believe that these incomes are truly zero, especially for the White households, which are more likely to report missing values. The 1993 data do, in principle, though, account for the difference between zero incomes and missing or nonreported incomes. We elect to treat the data as it stands. In table 9.3, we have not deleted the huge incomes reported nor have we set zero incomes to missing. In most of the analysis that follows, though, we work with log incomes, and this addresses each of these issues in different ways. The zero incomes are dropped and, especially given the large fraction of them that belong to White households, this strikes us as reasonable. More importantly, the huge outliers have diminished influence on means when working with log incomes. Hence, by working with log incomes, we report statistics that are both more interpretable in percentage terms and more robust to the handful of outliers. Table 9.4 reports the means and distributions of log per capita household real incomes (hereafter “log incomes”). In 1993, the mean log income was 8.44 and it had grown to 8.58 by 2008—a 14 percent increase in real incomes. This figure is quite close to the 15 percent increase in median incomes found in table 9.3. It is smaller than the 25 percent increase implied by the 1.5 percent average growth rate compounded over fifteen years as indicated by the macrodata. In terms of orders of magnitude, though, the micro- and macrodata convey very similar messages. Again, the growth in mean log incomes was not equal across the population groups. African households experiences a 26 percent point increase, while for Coloured households the figure was 8 percentage points. Asian households saw a 5 percent increase and White households a 28 percent increase. For all groups, the standard deviation of log incomes increased over this period. The bottom panel of table 9.4 reports percentiles. All population groups experienced increases in the median log income. Examination of the entire distribution for the overall population shows increases at each reported percentile except the first. For African households, the first percentile is the only one to report a decline in log real income—all other reported percentiles increased. For Coloured households, the gains were less pervasive. Only the top half of the reported percentiles saw increases in real incomes. The

8.03 1.26

4.75 5.96 6.52 7.20 8.06 8.95 9.68 9.97 10.44 5,228,275

4.82 6.18 6.71 7.49 8.46 9.47 10.23 10.63 11.39 7,323,317

African

8.44 1.41

All

5.80 6.90 7.27 7.97 8.60 9.17 9.73 10.15 10.70 592,242

8.54 1.02

Coloured

1993

5.24 7.57 8.11 8.57 9.14 9.83 10.25 10.59 11.04 193,320

9.13 1.00

Asian

6.18 8.08 8.65 9.44 10.09 10.63 11.08 11.43 11.95 1,209,480

9.94 1.09

White

Household log incomes in South Africa: 1993 and 2008

Note: Unit of observation is per capita household total income.

Mean Std. dev. Percentiles 1st 5th 10th 25th 50th 75th 90th 95th 99th n

Table 9.4

4.47 6.28 6.76 7.57 8.55 9.62 10.49 10.99 12.15 12,468,985

8.58 1.54

All

5.04 6.23 6.64 7.42 8.24 9.18 10.06 10.47 11.24 9,406,025

8.29 1.35

African

4.13 6.64 7.17 7.77 8.67 9.35 10.05 10.62 13.09 1,071,028

8.62 1.39

Coloured

2008

4.39 4.39 7.35 8.34 9.38 10.19 10.96 12.01 12.24 318,660

9.18 1.81

Asian

5.72 8.29 8.83 9.59 10.27 10.92 11.53 12.16 15.19 1,606,743

10.22 1.48

White

Fifteen Years On: Household Incomes in South Africa

341

same was true for Asian households, although this group is much smaller. Like African households, White households saw increases throughout the distribution except for the bottom percentile. The overall picture painted by table 9.4 is one of modest but pervasive increases in real incomes over the fifteen years since the fall of apartheid. An important exception to this is the bottom half of the distribution of Coloured households. Anecdotes that the Coloured population has been left behind relative to the larger African population group are supported by the nationally representative data in table 9.4. On the whole, though, log incomes have increased. As is to be expected given the inequality in South Africa, the increase in log incomes is but a fraction of the increase in (level) incomes. Figures 9.1 and 9.2 display the cumulative density functions of log incomes for all South Africans and for African households, respectively. As indicated by table 9.4, figure 9.1 shows more modest gains, but it is still the case that in most (but not every) parts of the distribution, log real incomes were higher in 2008. Figure 9.2, for African households only, shows a more distinct pattern of increased log incomes throughout the distribution. Figures 9.3 and 9.4 give the kernel density estimates of the income distributions for all South Africans and for Africans only, respectively. These are presented for two reasons. First, they better highlight relative gains of different segments of the income distribution. We illustrate this point immediately below. Second, the probability density functions (as opposed to the

Fig. 9.1

Household income cumulative density functions

Fig. 9.2

Household income cumulative density functions, African only

Fig. 9.3

Log per capita household incomes

Fifteen Years On: Household Incomes in South Africa

Fig. 9.4

343

Log per capita household incomes, African only

cumulative density functions) will serve as the basis for our investigation of what might explain the differences between the 1993 and 2008 distributions. We rely on methods developed in DiNardo, Fortin, and Lemieux (1996) and Leibbrandt, Levinsohn, and McCrary (2010), and those methods are based on probability density functions. We discuss the relationship between the density functions using the African-only examples given in figures 9.2 and 9.4. This is because the cumulative density function in figure 9.2 is easier to read than that in figure 9.1. The logic is the same for the density functions for the entire population (figures 9.1 and 9.3.) Figure 9.2 showed that the cumulative density function for 2008 lay to the right of that for 1993, indicating gains in real income throughout the distribution. Figure 9.4 highlights the fact that those gains were greater for the bottom and top third of the distribution than they were for the middle third. There is a section of the 2008 distribution, from log incomes of about 7 to log incomes of about 10 for which the 2008 distribution lies mostly to the left of that for 1993 in figure 9.4. Put another way, while real incomes were higher for African households throughout almost the entire distribution of income, the larger gains went to the bottom and top third of the distribution. Having documented the changes in incomes from 1993 to 2008, we now turn to an analysis of what explains these changes.

344

9.4

Murray Leibbrandt and James Levinsohn

What Drives the Changes in Household Incomes?

We investigate three possible explanations for what might account for the shift in the density functions given in figures 9.2 and 9.4. The first candidate is that endowments have changed, the second that returns to those endowments changed, and the third that the Child Support Grant explains at least the shift for the bottom half of the income distribution. Each are discussed in turn. 9.4.1

Does a Change in Endowments Explain the Shift in the Distribution of Incomes?

To investigate the role that changes in endowments might have played in shifting the distribution of log real incomes, we apply the approach of DiNardo, Fortin, and Lemieux (1996) (hereafter simply DFL.) This is a nonparametric approach and as such has both advantages and disadvantages. A key advantage is the ability to examine how a counterfactual impacts the entire distribution of income and to do so in a way that does not impose strong parametric assumptions (as, for example, is the case in Blinder [1973], and Oaxaca [1973]). A disadvantage is that the standard sort of hypothesis tests typically applied in parametric settings are not applicable to the nonparametric approach.5 We begin by setting notation.6 The density functions for household income in periods t and t′ may be written as (1)

f (y|T = t) = ∫ g(y|x,T = t)h(x|T = t)dx

and (2)

f (y|T = t ′ ) = ∫ g(y|x,T = t ′ )h(x|T = t ′ )dx

respectively, where T is a random variable describing the year from which a given household in the pooled data set of observations from both survey years is drawn, g(y|x,T = t) is the density of household income evaluated at y, given that the observable attributes of the household, X, are equal to x and that the survey year is t, and h(x|T = t) is the density of attributes evaluated at x, given that the survey year is t. It is perhaps helpful to think of g(y|x,T = t) as the function that “translates” observable attributes into income. Were this a traditional parametric regression of household income on household endowments for a given year t, the density of household income, f(y|T = t), would be analogous to the dependent variable, income; 5. It is possible, though, to investigate the impact of a change in only one endowment as opposed to all of them. 6. The description of how the endowments’ counterfactual distribution is estimated draws from Leibbrandt, Levinsohn, and McCrary (2010).

Fifteen Years On: Household Incomes in South Africa

345

h(x|T = t) would be analogous to the endowments data; and g(y|x,T = t) would be analogous to the returns to those endowments. We are interested in how the density of household (log) income changes if attributes and/or returns to those attributes changed. In this case, we are interested in how the distribution of income in period t would differ, were the endowments as they were in period t′. That is, what if households’ endowments were those that were obtained in 2008 (t′) instead of the actual 1993 (t) endowments? We denote this counterfactual by fht→t′; it may be written symbolically as (3)

fht→t ′ ( y ) ≡ ∫ g(y|x,T = t)h(x|T = t ′ )dx.

Notationally, the subscript h indicates that it is the density of attributes, or h(x|T = t), that is being changed from an actual to a counterfactual density. The superscript, t → t′ indicates that in this counterfactual, we are going to start with data from period t and use statistical techniques, in particular a reweighting scheme, to transform the actual density of attributes from the h(x|T = t) that reigned in period t to the counterfactual density h(x|T = t′) that reigned in period t′. The key insight from DFL is that the counterfactual in equation (3) is easy to implement by simply reweighting the data. The reweighting idea of DFL is based on the simple recognition that Bayes’ Axiom implies (4)

h(x|T = t′) P(T = t′|X = x) P (T = t′) t′ (x). = / ≡ tt→ h h(x|T = t) 1 − P(T = t′|X = x) 1 − P (T = t′)

t′ In words, τt→ (x) is just the ratio of the conditional odds to the uncondih tional odds. This is the weighting function needed to conduct the endowments counterfactual of equation (3). To see this, rewrite the object of interest fht→t′ ( y) as

(5)

fht→t ′ ( y ) = ∫ g ( y|x,T = t ) h ( x|T = t ′ ) dx = ∫ g ( y|x,T = t ) h ( x|T = t )

h ( x|T = t ′ ) dx h ( x|T = t )

t′ = ∫ g ( y|x,T = t ) h ( x|T = t ) tt→ ( x ) dx. h t′ which differs from equation (1) only by the weight τt→ (x). Consequently, h t→t′ we estimate the weighting function τh (x) and then compute the counterfactual (3) using a reweighted density estimate of incomes. A recipe-style description of exactly how this is done is given in Leibbrandt, Levinsohn, and McCrary (2010). In order to estimate the counterfactual density, it is necessary to estimate the numerator of equation (4) using a simple logit regression. This is a regression in which the dependent variable is an indicator for whether the

346 Table 9.5

Murray Leibbrandt and James Levinsohn Logit regression for reweighting

Variable

Coefficient

Standard error

z

Household size African Coloured Asian Number w/formal jobs Number of adults Highest education in HH Gender of HH head SOAP eligible Number of children under 14 Urban Metro Constant

–.211995 1.473985 1.155165 .9407829 –.6269353 –.0474666 .1236585 –.3668922 –.1605647 .1361046 .858294 1.586205 –.5383919

.039686 .060685 .083275 .1223874 .0277821 .0446675 .0062593 .042225 .0366606 .0438766 .0509862 .0511592 .1522273

–5.34 24.29 13.87 7.69 –22.57 –1.06 19.76 –8.69 –4.38 3.10 16.83 31.01 –3.54

Notes: Dependent variable is a 1 if year is 2008, 0 if 1993. Whites are the excluded population group. Highest education is given in years. Rural is the excluded region-type category.

year is 1993 or 2008 and the dependent variables are the household endowments. The results of this regression are given in table 9.5. Although the sole purpose of this regression is to estimate the conditional probabilities that enter the numerator of the DFL weight, the results are interesting in their own right. The dependent variable is coded so that it is 1 if the year is 2008 and 0 if the year is 1993. The results show that conditional on other regressors, household size shrank while the fraction of households that were African increased the most followed by Coloured, followed by Asian, with White households as the excluded group. All of these are consistent with the simple correlations in the data. Other results (again conditional on other regressors) indicate that the number of adults with formal jobs declined, the number of adults in the household declined, the highest education level of the household rose, the likelihood that a household member was eligible for a state old age pension fell, the number of children eligible for a Child Support Grant rose, and the fraction of households that were metro or urban rose (relative to those that were rural.) Except for the number of adults in the household, all of these variables are quite statistically significant.7 The estimated counterfactual is given in figure 9.5. This figure is for all households. The results for only African households are quite similar. It is 7. There is a pretty good argument that the number of formally employed adults should not be included as a regressor and we have replicated all results without this regressor. Results are essentially identical.

Fifteen Years On: Household Incomes in South Africa

Fig. 9.5

347

The “change in endowments” explanation

clear from the figure that the endowments counterfactual does not change the upper tail of the 1993 distribution at all. Thus, the actual improvement in incomes in the top tail by 2008 is not driven by changes in endowments. However, the counterfactual simulation changes the shape of the rest of the 1993 distribution fairly dramatically. The bottom two-thirds of 1993 distribution shifts to the right. This implies that for all but the top end of the 1993 distribution, incomes would have been greatly improved with 2008 endowments. This strong positive result is interesting because, superficially at least, the logit results shown in table 9.5 show a mixed bag of positive and negative (conditional) endowment changes. Higher levels of urbanization, higher levels of education, and smaller household sizes are potential positives. However, the declining population share of White South Africans, the lower numbers of employed members per household, and the lower number of members eligible for the old age pension are negative endowment changes. The fact that the counterfactual distribution shifts well past the actual 2008 distribution implies that, in reality, some other factors offest the impact of these improved 2008 endowments. Actual income changes in the bottom tail were much smaller than simulated and improvements in the middle of the distribution did not happen at all. The change in the returns to these endowments is one such factor that could either accentuate or counterbalance the endowments’ effect and we now turn to this issue.

348

Murray Leibbrandt and James Levinsohn

9.4.2

Does a Change in Returns Explain the Shift in the Distribution of Incomes?

An alternative explanation is that the returns to a household’s endowments have changed from 1993 to 2008. Just as it was possible to simulate what the entire distribution of incomes would have been if returns were constant but endowments changed, one can simulate what the distribution of household incomes would be if endowments were constant but returns were those obtained in 2008. We do just this using the methodology developed in Leibbrandt, Levinsohn, and McCrary (2010). We label this counterfactual by fgt→t′ and note that it may be written symbolically as (6)

f gt→t ′ ( y ) ≡ ∫ g ( y|x,T = t ′ ) h ( x|T = t ) dx.

We again use Bayes’ Axiom to derive an appropriate weight (7)

g(y|x,T = t′) P(T = t′|X = x,Y = y) P(T = t′|X = x) t′ = / ≡ τt→ g (x, y) g(y|x,T =t) 1− P(T = t′|X = x,Y = y) 1− P(T = t′|X = x)

and note that the counterfactual distribution may be rewritten as: (8)

f gt→t ′ ( y ) = ∫ g ( y|x,T = t ′ ) h ( x|T = t ) dx = ∫ g ( y|x,T = t ) h ( x|T = t )

g ( y|x,T = t ′ ) dx g ( y|x,T = t )

t′ = ∫ g ( y|x,T = t ) h ( x|T = t ) τt→ ( x, y ) dx. g

In practice, estimation of the weight given in equation (7) requires estimating the same logit as used in the endowments counterfactual and an additional logit regression in which household income is included both as a regressor itself and also interacted with all the included household attributes. In previous work (Leibbrandt, Levinsohn, and McCrary 2010), the returns counterfactual showed that returns to endowments played a major role in explaining the change in the distribution of individual incomes between 1995 and 2000. However, the counterfactual distribution at the household level that is shown in figure 9.6 makes it clear that simulating a change in returns for the 1993 distribution had very little impact on the distribution. As with the endowment simulation, there is no change to the upper tail of the 1993 distribution. Thus the actual improvement in incomes at the top end in 2008 is explained by neither endowments nor returns to endowments.8 The counterfactual shifts the lower tail of the 1993 distribution to the right, but 8. This finding is probably due to a violation of the common support assumption underlying the nonparametric approaches.

Fifteen Years On: Household Incomes in South Africa

Fig. 9.6

349

The “change in returns” explanation

not as significantly as the actual rightward shift in the density between 1993 and 2008. Nonetheless, this lowering of mass in the density at the bottom is accommodated by some improvements in the middle of the distribution. 9.4.3

Does the Child Support Grant Explain the Shift in the Distribution of Incomes?

Over the entire postapartheid period the state old age pension has formed the central plank of an extensive social security system. See Case and Deaton (1998) for an early analysis of this. Over 80 percent of the elderly receive this pension. However, as this pension has been in place over the entire period at roughly constant real values, it is unlikely to have been responsible for major changes in the distribution of income. A new grant, the Child Support Grant (CSG), was introduced in April 1998. It initially provided ZAR 100 for every child in the household younger than seven years of age. Over time, it became both more generous (the grant is rose to ZAR 240 per child) and more pervasive as the means test was relaxed and the age below which a child qualified was raised to fifteen in January 2008. By April 2009, 9.1 million children were benefiting from Child Support Grants.9 In short, in the period between 1993 and 2008, the Child Support Grant became a significant income source for poorer households. In this section, we investigate the role the CSG might have played in explaining the difference in the distribution of real household incomes. 9. This figure is from the National Treasury (2010).

350

Fig. 9.7

Murray Leibbrandt and James Levinsohn

The “Child Support Grant” explanation

The returns explanation begins to address this issue because we have included the number of children who would qualify for the CSG as a household “endowment” or attribute. The CSG acts to increase the return to this household attribute. It is not possible, though, to estimate the counterfactual density that would obtain if the return to only one attribute changed.10 In order to investigate the impact of the CSG alone, we have simply computed what household incomes would have been but for the CSG by subtracting this source of income from 2008 household incomes. The results are reported in figure 9.7. This figure gives the level of income for incomes below the median 2008 household income (including the CSG). Figure 9.7 shows that the CSG has played an important role in increasing incomes for poorer households. By comparing the actual 2008 density from that which would obtain but for the CSG, it is clear that while the CSG has benefited all income levels below the median, the benefit is larger the poorer the household. This is evidenced by the fact that the gap between the actual and but-for-the-CSG incomes is larger the poorer the household. Indeed, without the CSG, there would have been about three times as many households reporting zero incomes.11 For most income levels in figure 9.7, the but-for-the-CSG density lies below the 1993 density. This suggests that 10. The reason for this is explained in detail in Leibbrandt, Levinsohn, and McCrary (2010). 11. Figure 9.7 is presented in levels rather than in logs so as to make this point. The issue of zero incomes is brushed aside when working with log incomes.

Fifteen Years On: Household Incomes in South Africa

351

the CSG more than explains the income gains by households below the median income level. We conclude that the CSG has played an important role in explaining why incomes increased for the bottom half of households. 9.5

Conclusions: Elements of Success, but is it Sustainable Going Forward?

This chapter is based on national household surveys conducted in 1993 and 2008. These years mark the start and the end of the first fifteen years of postapartheid South Africa. The data are constructed so as to ensure that the two years are comparable. What does this comparison show? The data show an increase in average per capita real incomes. For the most part, this increase is evident across the distribution. This means that growth has been shared, albeit unequally, across almost the entire spectrum of incomes. This is especially true for the African group that makes close to 80 percent of the population. We cite evidence from other researchers that this income improvement was accompanied by strong improvements in access to important services such as water, housing, and electricity. However, as the kernel density estimations that we present make clear, these real income changes are not dramatic. The increases are modest and the densities hint at the fact that inequality has increased. Our research and that of others confirm that the very high levels of inequality that apartheid bequeathed the incoming government in 1994 have increased even further. Also, rising unemployment makes it clear that the labor market has been a problem rather than part of the solution over the last fifteen years. We conduct a series of semiparametric decompositions in order to see if we can better understand the source of the shifts in the distribution of incomes. These decompositions look at the role of changes in endowments and changes in the returns to these endowments in driving the observed changes in the income distribution between 1993 and 2008. This analysis proves to be very useful in highlighting the positive role played by changes in endowments over the period. Indeed the resulting endowments counterfactual indicates that, if these endowment changes were all that changed in South Africa over the postapartheid period, we would have seen a pervasive rightward shift of the distribution of per capita real incomes. This contrasts sharply with the actual shifts in the densities, which show clear improvements only at the bottom and the top of the densities. This is an important finding as it highlights the fact that the strong spending by the state on education and services led to measurable improvements in levels of education and access to essential services, but these improved endowments did not translate into generalized increases in real incomes. Therefore, something dampened the translation between improved endowments and improved real incomes. Our semiparametric analysis of returns indicates that, at the household level, this dampening was not due to a pernicious change in returns to endow-

352

Murray Leibbrandt and James Levinsohn

ments. Ceteris paribus, the change in returns makes a small positive contribution to the bottom and middle sections of the distribution. Unfortunately, the semiparametric analysis is not able to assess the impact of changes in returns to each separate endowment. This is a pity as the evidence coming from the analysis of individual earnings in the labor market (e.g., Banerjee et al. 2008) is that there has been a skills twist in the returns to education in South Africa that has lowered the returns to education for all but the highest levels of schooling. This includes the incomplete secondary school years where the greatest gains have been made in postapartheid South Africa. From the advent of the postapartheid period, South Africa has always had an extensive social welfare system based on a large state old age pension. This pension persisted through the postapartheid years but has not been extended significantly. There has been one major extension to the welfare system; from 1998 onward, a Child Support Grant was implemented with very high take-up in the middle of the first decade of the twenty-first century. In our semiparametric framework, this would change the returns to the endowment of the number of young children in the household. While we cannot isolate the impact of this change within the semiparametric framework, we run a simple with CSG/without CSG simulation that shows just how important CSG income is to the lower part of the distribution of per capita real incomes. This is suggestive of the fact that it is the system of social grants in general, and the new support coming from the Child Support Grant in particular, that counterbalances a strongly negative set of changes coming from the labor market. The strong social spending on social services, education, and health have a potentially positive role to play. However, our evidence suggests that they are yet to generate broad-based income returns. The net effect of all of these changes is a positive increase in real incomes over the postapartheid period. Figure 9.8 and table 9.5 taken together reiterate the point that this increase in real incomes is the net outcome of some strongly positive and some strongly negative forces. Figure 9.8 presents social expenditures over the postapartheid period and extrapolates these expenditures into the next few years. It retells the remarkable story of the expansion of the social grants and also the large (by international standards) expenditures on education and health. As shown by the debt service figures, one of the accomplishments of South African government policy over the period has been that these expenditure expansions were accomplished while bringing down the daunting public debt that the apartheid state handed over to the new South African democracy. It is exactly this combination of cash transfers and the expansion of education that is credited with the reduction of inequality in Brazil and Mexico since 2000. However, as we have shown, inequality has risen, not fallen in South Africa. The key difference between the Latin American and the South

Fifteen Years On: Household Incomes in South Africa

Fig. 9.8 Table 9.6

353

Social expenditures as a percentage of GDP Unemployment rates by per capita income deciles Decile

1993 (%)

2008 (%)

1 2 3 4 5 6 7 8 9 10 Overall

49.1 33.6 26.8 22.0 23.4 18.7 14.5 9.4 4.3 1.5 13.7

69.4 46.0 46.7 36.9 30.3 26.1 20.1 16.4 9.0 4.5 24.4

Source: Leibbrandt et al. (2010).

African experiences seems to be that social grants and improved levels of education accompanied and contributed toward strong employment creation in Latin America, whereas this employment creation has not happened in South Africa. Table 9.6 shows this quite vividly. It can be seen that even in 1993 high unemployment rates were the marker of those in the lowest deciles. By 2008 unemployment rates rise across all deciles and they rise particularly sharply in the bottom half of the distribution. Taken in isola-

354

Murray Leibbrandt and James Levinsohn

tion this table does not accord with a society generating positive, inclusive economic growth and social stability. It is this balance that makes it hard to be unequivocally positive. The postapartheid state has clearly been proactive. However, other than through the generation of rising tax revenues, this appears to have failed to generate virtuous interactions with the real economy. Indeed, the global financial crisis has sharpened these dilemmas. It can be seen in figure 9.8 that the debt service is starting to rise again. This is a reflection of the fact that the growing budget deficits are being generated in order to finance the states’ expenditure programs. While real spending on social grants has been protected, it has not continued to grow. To some extent this is due to the tighter financial conditions. However, this is also due to a growing recognition that these grants cannot be expanded indefinitely. Woolard and Leibbrandt (2010) review a large corpus showing that these unconditional transfers result in many virtuous behavioral effects. These studies can be cited in support of the state’s program over the last fifteen years to expand these grants to where they are now, one of the largest programs in the world. However, with the state old age pension being larger than the median per capita income and with this pension and the Child Support Grant making up the dominant share of income for those in the lowest deciles, the case for future expansion of these grants is not clear. For one thing, the grants are specifically targeted at the elderly, the disabled, and children and rely on a set of indirect behavioral responses to connect to the labor market. Policies that directly address the labor market also warrant consideration.

References Banerjee, Abhijit, Sebastian Galiani, Jim Levinsohn, Zoe McLaren, and Ingrid Woolard. 2008. “Why Has Unemployment Risen in the New South Africa?” Economics of Transition 16 (4): 715–40. Blinder, Alan. 1973. “Wage Discrimination: Reduced Form and Structural Estimates.” Journal of Human Resources 8 (4): 436–55. Case, Anne, and Angus Deaton. 1998. “Large Cash Transfers to the Elderly in South Africa.” Economic Journal 180:1330–61. DiNardo, John, Nicole Fortin, and Thomas Lemieux. 1996. “Labor Market Institutions and the Distribution of Wages, 1973–1992: A Semi-Parametric Approach.” Econometrica 64 (5): 1001–44. Duflo, Esther. 2003. “Grandmothers and Granddaughters: Old Age Pensions and Intrahousehold Allocation in South Africa.” World Bank Economic Review 17 (1): 1–25. Hoogeveen, Johannes, and Berk Ozler. 2005. “Not Separate, Not Equal: Poverty and Inequality in Post-Apartheid South Africa.” William Davidson Institute Working Paper no. 739.

Fifteen Years On: Household Incomes in South Africa

355

Leibbrandt, Murray, James Levinsohn, and Justin McCrary. 2010. “Incomes in South Africa after the Fall of Apartheid.” Journal of Globalization and Development 1:Article 2. Leibbrandt, Murray, Ingrid Woolard, Arden Finn, and Jonathan Argent. 2010. “Trends in South African Income Distribution and Poverty since the Fall of Apartheid.” Social, Employment and Migration Working Papers no. 101, Organisation for Economic Co-operation and Development. Leibbrandt, Murray, Ingrid Woolard, Hayley McEwen, and Charlotte Koep. 2010. “Better Employment to Reduce Inequality Further in South Africa.” In Tackling Inequalities in Brazil, India, China and South Africa: The Role of Labour Market and Social Policies. Paris: OECD. National Treasury. 2010. “2010 Budget Review.” Pretoria, South Africa: Government Printer. Oaxaca, R. 1973. “Male-Female Wage Differentials in Urban Labor Markets.” International Economic Review 14:693–709. Thomas, Duncan. 1996. “Education across the Generations in South Africa.” American Economic Review 86 (2): 330–34. Woolard, Ingrid, and Murray Leibbrandt. 2010. “The Evolution and Impact of Unconditional Cash Transfers in South Africa.” Southern Africa Labour and Development Research Unit Working Paper no. 51, University of Cape Town.

10

Is Tanzania a Success Story? A Long-Term Analysis Sebastian Edwards

10.1

Introduction

In 1991, three decades after obtaining independence from Great Britain, Tanzania was the second poorest country in the world. According to the World Bank’s World Development Report, its gross national product (GNP) Sebastian Edwards is the Henry Ford II Professor of International Economics at the University of California, Los Angeles, and a research associate of the National Bureau of Economic Research. This work is part of the National Bureau of Economic Research Africa Project. Many people helped me with this work. In Dar es Salaam I was fortunate to discuss a number of issues pertaining to the Tanzanian economy with Professor Samuel Wangwe, Professor Haidari Amani, Dr. Kipokola, Dr. Hans Hoogeveen, Mr. Rugumyamheto, Professor Joseph Semboja, Dr. Idris Rashid, Professor Mukandala, and Dr. Brian Cooksey. I am grateful to Professor Benno Ndulu for his hospitality and many good discussions. I thank David N. Weil for his useful and very detailed comments on an earlier (and much longer) version of the chapter. Gerry Helleiner was kind enough to share a chapter of his memoirs with me. I thank Jim McIntire and Paolo Zacchia from the World Bank, and Roger Nord and Chris Papageorgiou from the International Monetary Fund for sharing their views with me. I thank Mike Lofchie for many illuminating conversations, throughout the years, on the evolution of Tanzania’s political and economic systems. I am grateful to Steve O’Connell for discussing his work on Tanzania with me, and to Anders Aslund for helping me understand the Nordic countries’ position on development assistance in Africa. Comments by the participants at the National Bureau of Economic Research Africa conference, held in Zanzibar in August 2011, were particularly helpful. I am grateful to Kathie Krumm for introducing me, many years ago, to the development challenges faced by the East African countries, and for persuading me to spend some time working in Tanzania in 1992. Juan Marcos Wlasiuk, a devoted Africanist and a great friend, provided wonderful research assistance in Los Angeles. I also thank my students at UCLA’s Anderson Graduate School of Management for providing helpful insights on Tanzania’s political, business, and economic environment. Finally, I thank Elisa Pepe from the National Bureau of Economic Research for her amazing support throughout this project. The kind financial support of the NBER Africa Project is gratefully acknowledged. For acknowledgments, sources of research support, and disclosure of the author’s material financial relationships, if any, please see http:// www.nber.org/chapters/c13448.ack.

357

358

Sebastian Edwards

per capita was barely USD 100; the only country with a lower income per person was Mozambique, with USD 80 per year. The fact that Tanzania was very poor was not in itself surprising. What was startling, however, was that in 1976, merely fifteen years earlier, twentyfour countries were poorer than Tanzania. According to the World Bank, between 1976 and 1991 Tanzania’s nominal GNP per capita declined by 45 percent—from USD 180 to USD 100. When data adjusted for purchasing power parity (PPP) are used, the results are less dramatic, but still show a very impressive reduction in the standard of living: between 1976 and 1991 real income per capita fell by 15 percent, or almost 1 percent per year. The collapse of the Tanzanian economy between the mid-1970s and the early 1990s represents one of the most spectacular economic disintegrations ever experienced in a country not affected by a major war or natural disaster.1 Since 1991, however, Tanzania has gone through a major comeback. According to data from the Penn World Tables, between 1990 and 2008 real (PPP-adjusted) gross domestic product (GDP) per capita increased by 64 percent, or at a rapid 2.8 percent per year. In the year 2000, the country had reached its previous peak GDP per capita (achieved in 1976).2 During the last two decades there has also been a marked improvement from a comparative point of view: although Tanzania continues to be very poor, it is not any longer at the very bottom of the income per capita tables. According to the World Bank, in 2009 nineteen nations had a GDP per capita lower than Tanzania—sixteen of which are in Africa. Moreover, the country has been able to weather the recent global financial crisis without suffering major setbacks. According to the International Monetary Fund (IMF), GDP growth slowed down to 5 percent in 2008; it then increased to 5.5 percent in 2009 and to 6.2 percent in 2010; it is projected to be 6.7 percent in 2011. There has also been important progress in terms of social indicators. According to the United Nations’ Human Development Index (HDI), Tanzania has made significant improvements when compared to both sub-Saharan Africa and to the rest of the world. In 1995 Tanzania’s composite HDI was barely 90 percent that of the rest of the sub-Saharan nations; by 2010, Tanzania 1. In October 1978, Tanzania was invaded by Idi Amin’s Ugandan forces. This was a short war that ended with Tanzania’s victory and with the fall of Amin in April 1979. Although this was not a protracted or major conflict, it did have significant negative effects on the Tanzanian economy, as did other external events, such as the drought of 1974–1975 and the oil price shocks of 1973 and 1979. As I will argue in sections 10.3 and 10.4, the fact that these shocks were amplified significantly and contributed to topple the economy was, to a large extent, a reflection of how fragile the economy had become after almost twenty years of socialist experiments (see the discussion below). For an early comprehensive study of the political economy of Tanzania, see Coulson (1982). For more recent analyses see, for example, Mwase and Ndulu (2008). See, also, the bibliography to section 10.2 of this chapter. 2. An important question is when the growth “breakpoint” actually took place. In a recent innovative paper, Robinson, Gaertner, and Papageorgiou (2011) use time series techniques to investigate whether there has indeed been a structural break in Tanzania’s growth process. Their analysis indicates that such a breakpoint took place in 1996.

Is Tanzania a Success Story?

359

had surpassed the African region, and its HDI was almost 3 percent above that of sub-Saharan Africa as a group. What makes Tanzania’s story fascinating is that foreign assistance has been at the center of the country’s economic failures and successes. After independence in 1961, the country became one of the “darlings” of the international aid community; between 1962 and 1983 Tanzania was one of the highest recipients of foreign aid in the world.3 A high proportion of this aid was bilateral and came from the European nations—especially from the Nordic countries. The multilateral institutions, and in particular the World Bank, also contributed significant volumes of funds to the country’s early development effort. Massive foreign aid was largely used to finance President Julius Nyerere’s African Socialism vision. In particular, international donors helped fund the initiatives contained in the Arusha Declaration, a broad political manifesto presented by Nyerere to the official party (the Tanganyika African National Union [TANU]) in February 1967.4 By 1973, net official development assistance (ODA) per capita to Tanzania was already 18 percent higher than net foreign aid received, on average, by sub-Saharan African (SSA) countries. By 1975, net per capita ODA to Tanzania had surpassed the SSA average by 75 percent, and by 1981 it was almost twice as much as the average for SSA.5 Nyerere’s socialist policies, however, did not work. The collectivization of agriculture backfired, the villagization process that forced peasants to move to villages designed by planners was strongly resisted by the population, the parastatal sector became a huge financial burden and a source of corruption, and grandiose industrial projects became mired in inefficiencies. In the mid-1970s, significant shortages of all sorts of goods developed, and black market activities became rampant.6 The collapse of the Tanzanian economy in the late 1970s and early 1980s happened in spite of the involvement of donor countries—in fact, it is possible to argue that this disintegration happened because aid agencies were heavily involved in supporting (and even helping design) Nyerere’s ujamaa socialism economic policies. In many ways, the Tanzanian experience between 1968 and the mid-1980s provides a stark example of the excesses of foreign assistance. During that period the international aid community supported policies—including the taxation of peasants and agriculture—that weakened the economy, encouraged corrup3. Bigsten et al. (1999, 2001). See also the data and discussion below. 4. The TANU became the only legal political party in Tanganyika in January 1963. Zanzibar, which formed a union with Tanganyika in April 1964, had its own official party, the Afro-Shirazi Party (ASP). In 1977 the TANU and the ASP merged to form the Chama Cha Mapinduzi (CCM), the official and only party for the whole country. Tanzania adopted a multiparty political system in July 1992. 5. See the data and references in Edwards (2011). 6. See Mtei (2009) for an insider’s personal account of this period. Edwin Mtei was the first governor of the Bank of Tanzania, and an actor in many of the country’s early economic dilemmas.

360

Sebastian Edwards

tion, and generated economic dependency. Indeed, it may be argued that Tanzania provides the typical case of “deadly aid” described by critics such as W. Easterly (Easterly 2007) and D. Moyo (Moyo 2009). However, the story of the relationship between Tanzania and the aid community is much more complex than what it appears at first. Indeed, there were many excesses until the mid-1980s, and many wasteful projects were financed with aid monies. A questionable development strategy was encouraged, and policies that reduced the well-being of millions of people were supported. However, in the early 1980s the same donor community that had financed Nyerere’s experiments demanded significant policy rectification. Already in 1979, immediately after the war with Uganda and the collapse of the East African Community, the IMF requested major changes in economic policy. These included significant fiscal adjustment, a reduction in parastatals’ deficits, and a major devaluation of the Tanzanian shilling. The government, however, resisted these recommendations, arguing that they would work against the nation’s overall socialist strategy. In the years that followed, the crisis deepened and social conditions worsened significantly. Starting in 1982, and as a way to induce adjustment and policy changes, the international community began to curtail foreign aid flows. In the four years between 1981 and 1985, net official assistance, in per capita terms, declined by a remarkable 40 percent. In 1986, and after President Nyerere decided not to run for reelection, the government of Tanzania and the IMF reached an agreement, and a Standby program was put into place. The shilling was devalued by 57 percent, and fiscal adjustment policies were undertaken.7 In the years that followed, and with the assistance of the World Bank and other donor agencies, a process of reform aimed at dismantling controls, reducing inflation, and (eventually) eliminating the black market for foreign exchange was launched. Starting in 1986, and as a way to reward the change in policies, net aid once again increased. In 1988, and in per capita terms, it was 92 percent higher than in 1985. The reforms were gradual, and at times they stalled; some of them even backtracked. Slowly, and after significant strife within the government, the reform agenda gained some momentum and the economy began to recover: fiscal imbalances were reduced, the external sector was liberalized, foreign direct investment was welcomed, the exchange rate was unified, the black market for foreign currency was reduced in size and eventually eliminated, and a series of social programs aimed at assisting the poor and reducing poverty were put in place. Throughout this period the international community continued to use development assistance as a tool to induce change and guide policy. When 7. If the devaluation is measured as the percentage change in the official price of one USD in terms of shillings, its magnitude would be 135 percent.

Is Tanzania a Success Story?

361

the reforms stalled, the donors would withhold aid flows. A particularly serious impasse between the government and the aid community erupted in 1993–94, and was only solved in 1995 when a high-level committee chaired by Gerry Helleiner—a University of Toronto professor, and a Tanzania old hand—mediated between the parties and devised a new approach to coordinate aid. At the time of this writing (late 2011) the relationship between the international community and the government is largely based on credibility and trust. So much so, that an increasing fraction of assistance is provided as general budget support, as opposed to project financing. During the last few years Tanzania has become an often-discussed example of “African successes”.8 Officials in the multilateral organizations— the World Bank and the International Monetary Fund (IMF)—and in bilateral aid agencies repeatedly refer to Tanzania’s performance as a sign that, if properly provided, foreign assistance can be extremely useful and can help a country grow rapidly while reducing poverty. In advertising Tanzania’s “success” they mention its rapid rate of growth, the improvement in social conditions as reflected in the Human Development Index, low inflation, and macroeconomic stability. There are, however, a number of skeptics, mostly from nongovernmental organizations (NGOs). They argue that growth is overestimated, that corruption is generalized, that the government is authoritarian, and that poverty has not declined sufficiently. A serious shortcoming of much of the current discussion on Tanzania’s reforms and economic performance is that it lacks a historical perspective. Most recent studies by economists do not provide an appropriate background for understanding economic policy, the relation between the government and the aid community (both multilateral and bilateral), and the prospects of growth. Indeed, the years of Nyerere and African Socialism are mentioned on passing, but are not analyzed in details; there is no inkling on how devastating that period was for the population, or on the ferocious political battles that were waged in the years leading to the reforms. The purpose of this chapter is to provide a historical perspective on the reform process initiated in 1986 and deepened in 1996. In order to do this I concentrate mostly on the period spanning from 1967, when the Arusha Declaration was adopted by the official political party the TANU, and 1996, when a new approach toward foreign aid was implemented. I am particularly interested in investigating how external aid affected Tanzania during the early years, and how it contributed to the demise of the economy in the 1970s and 1980s. In doing this I emphasize both technical as well as political economy issues related to imbalances, disequilibria, devaluation, black markets, adjustment, and reform. Although I touch on many issues, I do not attempt to provide an exhaustive account of every aspect of the country’s 8. Nord et al. (2009); Robinson, Gaertner, and Papageorgiou (2011). For a more nuanced view see Lofchie (2011).

362

Sebastian Edwards

economic and political developments. Such an effort is beyond the scope of a single chapter, even if it is a long one. In discussing these issues I focus on macroeconomic policies and overall economic performance. Because of its emphasis on foreign aid and macroeconomics, the chapter pays special attention to three important episodes in Tanzania’s economic history: (a) the exchange rate crisis of the late 1970s and early 1980s; (b) the IMF Standby program and the maxi-devaluation of 1986; and (c) the serious impasse between donors and the Tanzanian authorities in the mid-1990s. My analysis of the peculiarities of the reform process after 1996 is brief and somewhat sketchy. This is deliberate, since there are a number of recent works that cover this period. However, I do provide an evaluation—from a historical perspective—of these reforms. Before proceeding, a word on methodology: In order to deal with the issues at hand from a historical perspective I follow the methodology of analytical narratives, an approach that I have used in my previous work and that I believe is best suited for addressing the intricacies of a complex and long saga that has gone from hope and enthusiasm in the 1960s, to collapse in the late 1970s and 1980s, and back to hope starting in the late 1990s.9 Although this is not a chapter on the politics of Tanzania—there are many books on the subject, and some of them are very good—I do emphasize, time and again, political economy angles. The rest of the chapter is organized as follows: In section 10.2 I deal with the current (that is, 2011–2012) “official” narrative regarding Tanzania’s reforms and economic performance. This narrative talks about a major success, and has its origins in the multilaterals institutions; not surprisingly, it has been enthusiastically embraced by the Tanzanian authorities. In section 10.3 I put things in historical context by discussing the main views on economic development that dominated thinking in the 1950s and 1960s. I then provide an analysis on the evolution of foreign aid since independence. Section 10.4 covers the period 1961 (when the country became independent) through 1980. The section opens with the expulsion of the IMF mission from Tanzania in November 1979. In section 10.5 I analyze the events that eventually led to the acceptance of an IMF program, and to the maxi-devaluation of 1986. Section 10.6 concentrates on the “war of ideas” and the role of academic and technocrats in forging the reforms. In section 10.7 I focus on the first years after the maxi-devaluation of 1986, and on the first round of reforms. Here I document the extent of policy changes and I investigate the causes behind the slow progress in the reform process. I argue that this had largely to do with the fact that—as has historically been the case in many reform episodes—reform opponents were able to regroup and regain a foothold in the power structure. In section 10.8 I (very briefly) provide some information on the second round of reforms. This is a 9. On “analytical narratives,” see the introductory chapter of Bates et al. (1998).

Is Tanzania a Success Story?

363

deliberately short section and is included here for the sake of completeness; as noted, the main objective of the chapter is to provide a detailed analysis of the historical background to these reforms. In the final section (10.9) I briefly deal with the reforms since 1996, and I ask whether recent economic performance has indeed been as good as the multilateral institutions, and the official data for that matter, have suggested. 10.2

Tanzania’s Market-Oriented Reforms and Economic Performance: The “Official Story”

There are a number of works on the evolution of the Tanzanian economy during the last fifteen to twenty years. The most comprehensive of these are Mans (1994), Mutalemwa and Ndulu (2002), Utz (2008), Mwase and Ndulu (2008), Nord et al. (2009), and Robinson, Gaertner, and Papageorgiou (2011). Many (but not all) of these wide-ranging analyses have been undertaken by economists at the International Financial Institutions (IFIs). All of these studies tell, with some variations, a story that emphasizes the role played by the market-oriented reforms in the country’s economic takeoff the mid-1990s. In many ways these analyses have contributed to the creation of a generally accepted or “official story” about Tanzania’s recent economic accomplishments. (In table 10.1 I provide data on the most important economic and social indicators for 1996–2010.)10 The most salient components of this consensus or “official” view may be described as follows:11

• Since the mid-1990s, Tanzania has experienced high economic growth. Some authors—most notably Robinson, Gaertner, and Papageorgiou (2011)—date the country’s takeoff around 1996. • From a comparative perspective, and since 1996, Tanzania has done much better, in terms of economic growth and macroeconomic stability, than the average sub-Saharan country. • Growth has been accompanied by increased macroeconomic stability, including a major reduction of inflation, which in 1984 was a high 36.1 percent, and averaged only 8 percent between 2005 and 2010. (See table 10.1.) • This acceleration of growth has been attributed to the market-oriented economic reforms undertaken since the mid-1980s. According to Robinson, Gaertner, and Papageorgiou (2011, 22): “[The] key factors behind the takeoff in growth include the significant structural changes that occurred as the basic institutions of a market economy were introduced.”

10. See Edwards (2014) for details. 11. The data presented in the bullets that follow come from the papers cited above and/or from the Bank of Tanzania, the IMF, and the World Bank.

13.2 17.5 –4.3 6.8 –6.4 13.4 38.7 21.0 8.4 580 122.2 19.9 31.9 117.1 86.0 95.3

(5) (6) (7) (8) (9) (10) (11) (12) (27) (28) (14) (15) (16) (17) (18) (19) (21) (22) (23) (24) (25) (26)

Social indicators Population (millions) Population growth (%) Urban population (% of total) School enrollment, primary (% gross) School enrollment, secondary (% gross) Life expectancy at birth 31.6 2.7 21.2 67.7 5.6 49.9

44.8 16.1 12.9 612 136.0 16.2 25.7 75.7 57.7 98.7

–5.7 12.3

13.9 11.0 2.9 8.1

3.5 0.9 14.7 78.1

1997

50.1

32.5 2.6 21.6 65.8

46.1 12.8 10.8 664 144.0 12.4 25.0 72.2 76.3 98.4

–7.5 10.7

13.3 13.1 0.1 6.4

3.7 1.2 19.6 67.5

1998

33.3 2.5 21.9 66.6 6.1 50.3

45.0 7.9 18.6 744 139.8 12.5 22.9 75.8 93.7 99.1

–8.0 10.2

13.4 12.7 0.7 8.0

4.8 2.3 17.1 65.5

1999

34.1 2.6 22.3 68.1 6.3 50.7

46.9 5.9 14.8 800 141.1 13.4 20.1 100.0 100.0 100.0

–4.2 10.4

14.6 16.1 –1.5 9.6

4.9 2.4 16.4 58.6

2000

35.0 2.6 22.7 74.2 6.8 51.1

56.6 5.1 34.2 876 143.4 17.0 21.3 125.9 116.2 95.2

–3.8 12.3

14.7 15.9 –1.2 11.1

6.0 3.4 17.0 53.2

2001

36.0 2.7 23.1 88.9 7.4 51.7

56.0 5.3 25.6 966 133.3 17.6 19.8 139.0 111.3 98.3

–0.4 11.7

14.9 15.3 –0.4 14.1

7.2 4.5 18.7 55.4

2002

36.9 2.7 23.4 95.5 7.6 52.3

62.2 5.3 18.0 1038 113.1 18.6 22.8 154.7 131.6 101.7

–1.5 14.8

15.9 17.3 –1.4 17.5

6.9 4.2 20.0 51.4

2003

37.9 2.8 23.8 100.7 9.3 52.9

56.5 4.7 13.5 1089 102.7 19.7 26.1 166.7 152.9 103.0

–3.8 13.8

17.4 20.4 –3.0 17.9

7.8 5.1 21.2 54.7

2004

39.0 2.8 24.2 104.7 11.0 53.6

45.3 5.0 34.8 1128 100.0 20.8 29.7 174.9 161.1 98.0

–7.9 10.6

19.3 23.1 –3.8 14.5

7.4 4.6 22.5 49.5

2005

40.1 2.8 24.6 107.7 13.6 54.3

52.5 7.3 21.5 1251 93.9 22.6 35.7 166.7 193.5 109.0

–7.7 12.8

22.9 28.2 –5.3 15.8

6.7 3.9 23.8 21.2

2006

41.3 2.9 25.1 109.6 20.0 55.0

72.9 7.0 20.5 1245 93.3 24.3 37.1 173.1 220.6 110.0

–10.8 16.8

22.8 22.8 0.0 17.2

7.1 4.3 25.0 23.3

2007

42.5 2.9 25.5 110.5 25.2 55.6

54.9 10.3 19.8 1196 97.5 22.6 39.1 205.9 249.2 107.9

–13.0 11.3

21.3 26.1 –4.8 13.8

7.4 4.5 26.3 22.2

2008

43.7 3.0 26.0 104.9 27.4 56.3

69.7 12.1 17.7 1320 99.4 23.2 35.2 182.9 256.3 121.1

–8.5 13.7

20.4 27.4 –7.0 16.2

6.0 3.1 29.3 24.6

2009

1409 90.6

5.4

27.1(*)

7.0(*) 4.1(*)

2010

Sources: (1), (2), (3), (4), (8), (9), (10), (11), (12), (15), (16), (21), (22), (23), (24), (25), (26), (27), (28): WDI; (5), (6), (7): IFS and IMF reports; (14): Author’s calculation. Data source: UNCTADstat and WDI; (17), (18), (19): UNCTADstat (only merchandise). *Estimates.

30.8 2.8 20.9 67.4 5.2 49.9

4.5 1.7 16.5 94.7

1996

(1) (2) (3) (4)

Source

Tanzania: Economic and social indicators, 1996–2010

Economic indicators Real GDP growth (%) Per capita GDP growth (%) Gross fixed capital formation (% of GDP) External debt (total, long-term, % of GDP) Revenue, cash (budgetary central gov’t., after grants) (% of GDP) Expense, cash (budgetary central gov’t.) (% of GDP) Cash surplus/deficit, cash (budgetary cen. gov’t.) (% of GDP) Total international reserves (excluding gold, % of GDP) Current account balance (excluding exceptional financing, % of GDP) Net official development assistance received (% of GDP) Per capita net official development assistance received (constant 2008 US$) Inflation, consumer prices (annual %) Money and quasi-money (M2) growth (%) Official exchange rate (TSh/US$, period average) Real effective exchange rate (CPI based, 2005 = 100) Exports (% of GDP) Imports (% of GDP) Volume indices of exports (2000 = 100) Volume indices of imports (2000 = 100) Terms of trade (merchandise, 2000 = 100)

Table 10.1

Is Tanzania a Success Story?

365

• The reforms came in two waves (see Nord et al. 2009; Robinson, Gaertner, and Papageorgiou 2011):

• Phase 1: Between 1986 and 1995 the country went through a pro-

cess of partial liberalization and reforms. Major distortions were addressed, an effort was made to reduce the black market for foreign exchange, and imports of some goods were allowed as long as buyers used their own funds. This initial phase of the reform effort stalled around 1992–93, when a major spat developed between the aid agencies and the government. • Phase 2: Since 1996 (and until the present time) deeper reforms have been put in place, and a serious effort at stabilizing the economy has been made. The reform effort took off after an agreement was reached between the aid community and the government—the seeds for the agreement were detailed in the so-called Helleiner Report. Starting in 1996 the economy was opened further, the civil service was reformed, rules on foreign direct investment (FDI) were relaxed and streamlined, privatization was implemented, banking reform was put in place, and massive programs aimed at improving education and health services were implemented (Nord et al. 2009; Robinson, Gaertner, and Papageorgiou 2011; Edwards 2011).

• Although the reforms have been gradual, in most areas they have been

•

•

•

•

deeper than in the rest of the sub-Saharan countries. This has been documented by Robinson, Gaertner, and Papageorgiou (2011), and is captured by “policy indexes,” such as those put together by the Fraser Institute and other think tanks. In Tanzania the reforms impacted economic performance with a considerable lag; while the reforms were initiated in 1986—albeit mildly and in a go-stop-backtrack-go fashion—the inflection point in economic performance did not occur until a decade later, in 1996. Macroeconomic stability, including the reduction of very large fiscal imbalances, has played an important role in the positive performance of the Tanzanian economy. By avoiding outbursts of inflation, the private sector has been able to concentrate on expanding output and improving efficiency. A very high fraction of government expenditures has been financed by foreign aid. Official assistance was 12.1 percent of GDP in fiscal year 2008/09 (total foreign aid—including private aid—in that year reached 17 percent of GDP). In recent years an increasing fraction of foreign aid has taken the form of “government budget support,” as opposed to “program support.” Between 1996/97 and 2008/09 government budget support more than doubled as a fraction of GDP, from 2.5 percent to 6 percent. Although financial reforms have been deep, there are still a number of distortions that constrain the economy. In particular, the banking sector

366

•

• •

•

• •

Sebastian Edwards

continues to be dominated by a small number of banks, and credit to the agricultural sector (the most important sector in the economy) continues to be relatively scarce. Until recently, exports have played a (very) minor role in growth. In the last few years, however, investment in the mineral sector—most notably in gold mining—has increased significantly. Much of this investment has been from multinational companies. The boom in commodity (and especially metal) prices has also contributed to this surge in investment. There has also been increased foreign investment in the tourism sector. Foreign direct investment has reached 3 percent of GDP in recent years. Recent legal reforms—and, in particular, the Mining Act of 2010—are expected to further bolster investment in the sector. From a sectoral point of view, the agricultural sector has lagged behind throughout the high growth period. This is important in a country where more than 74 percent of the population lives in the rural area. In terms of sources of growth, most authors have estimated that since 1996 total factor productivity (TFP) growth has been the most important driver of economic expansion. According to Nord et al. (2009), during 1996–2000 TFP growth contributed 2.3 percentage points to growth; according to these authors, during 2001–2008 TFP’s contribution to growth had climbed to 3.5 percentage points. Mwase and Ndulu (2008) used the Collins-Bosworth model to estimate TFP growth for Tanzania and sub-Saharan Africa for the period 1960–1997. Their results largely correspond to those in Nord et al. (2009) and Robinson, Gaertner, and Papageorgiou (2011).12 During the last few years there has been a marked improvement in tax collection. As a percentage of GDP, tax revenues have increased from less than 10 percent early in the twenty-first century to 16 percent in 2009/10. In spite of this improvement, there is still a large gap between government revenues and expenditures. As noted earlier, this gap is financed through foreign aid. The improved fiscal situation has allowed the government of Tanzania—as well as those of other SSA countries—to run countercyclical fiscal policy. What I have called the “official story” emphasized the improvements in social conditions. For example, in 1995 life expectancy at birth was slightly lower in Tanzania than in SSA; by 2010, however, life expectancy at birth was 2.5 years higher in Tanzania. The results are similar when other HDI indicators are considered. The government has emphasized the provision of social services through the MKUKUTA programs (I and II), which have obtained approximately 70 percent of budget allocations.

12. See also Treichel (2005) and World Bank (2007).

Is Tanzania a Success Story?

367

• In spite of the government’s efforts, in some areas progress has been slow—this has been particularly the case in rural education. It is estimated that by 2015 one-half of the Millennium Goals will be achieved (see table 10.2). In many ways this situation may be characterized as the glass being half full.

To summarize, the official story of Tanzania’s recent developments is one of success. To be sure, every one of the authors cited above recognizes that the country continues to be very poor and that it faces major challenges if it is to maintain the growth momentum. These challenges are, largely, of the “garden variety,” and are faced by every SSA country—or by every emerging and poor country in the world, for that matter. They include issues related to: (a) capital (physical and human) accumulation, (b) deepening the extent of competition, (c) improving infrastructure and the quality of education, (d) streamlining regulation, (e) encouraging entrepreneurship, (f) strengthening institutions, and (g) improving governance. In terms of sectors, Tanzania’s main economic challenges are related to improving productivity in the agriculture, improving the provision of public services, and avoiding the overvaluation of the currency. The fact that there is some type of agreement on how to interpret the country’s recent history—what I have called the “official story”—is not, on its own, a bad thing. Quite the contrary, it shows that there is clarity in the facts and their interpretation. The problem, as noted, is that many of these ac-counts—and, in particular the most recent ones—provide a very limited historical background. Indeed, after going through these works some readers may conclude that before the reforms launched in the mid-1980s Tanzania was just another third world country with poor to mediocre performance. This, of course, is not the case. As pointed out in the introduction—and, to be fair, in many earlier works on Tanzania’s development—the decade that preceded the reforms was, to put it mildly, a disaster. Worse yet, this disaster was the consequence of misguided policies that were often encouraged by foreign donors. In that regard, it is not possible to understand Tanzania’s recent growth takeoff without some detailed reference to the policies that followed the Arusha Declaration of 1967. That is, it is important to put statements such as “Tanzania has emerged as one of the most rapidly growing economies in south-Saharan Africa,” in context, and to explicitly address the calamitous collapse of the Tanzanian economy during Nyerere’s long rule. A second limitation of the official narrative is that very few studies penned at the international financial institutions address the issue of the quality of data. In fact, most of the accounts discussed above proceed as if the official figures are fully reliable and should not be questioned or contested. The reality, however, is different, and as I have argued in Edwards (2011), there are a number of indicators that suggest that official figures should be interpreted with care.

39.0 28.8 46.6 54.2 191 115 529 43.9 6.0 51 68

Proportion of population below basic needs poverty line Under 5 underweight (%) Under 5 stunted (%) Primary school net enrollment rate Under 5 mortality rate (per 1,000 live births) Infant mortality rate (per 1,000 live births) Maternal mortality rate (per 100,000 live births) Births attended by skilled health personnel (%) HIV prevalence, 15–24 years (%) Access to potable water (% of rural population) Access to potable water (% of urban population)

Source: United Republic of Tanzania (2009).

1990 36.0 29.5 44.4 58.7 153 99 — 35.8 — 42 85

2000 33.6 22.0 38.0 97.2 112 68 578 63 2.5 57.1 83

Actual

Tanzania (mainland): Millennium Development Goals, midway evaluation

Millennium Development Goal

Table 10.2 2008

25.0 18.4 29.8 87.2 99.6 59.6 244 77.1