Purchasing Behavior and Personal Attributes 9781512817850

Table of contents :
PREFACE
TABLE OF CONTENTS
TABLES
I INTRODUCTION AND PLAN OF THE STUDY
II THE DIMENSIONS OF PURCHASING BEHAVIOR
III FURTHER ANALYSIS OF PURCHASING BEHAVIOR DIMENSIONS
IV THE DESCRIPTION OF PERSONALITY AND SOCIO-ECONOMIC STATUS
V EXPERIMENTS ON THE ANALYSIS SAMPLE
VI RELATIONSHIPS BETWEEN PERSONALITY AND PURCHASING BEHAVIOR
APPENDICES
APPENDICES A TABLES
APPENDIX B THE EDWARDS PERSONAL PREFERENCE SCHEDULE
CONCLUSIONS
INDEX

Citation preview

PURCHASING BEHAVIOR AND PERSONAL ATTRIBUTES

PURCHASING BEHAVIOR AND PERSONAL ATTRIBUTES by

William F. M a s s y Stanford University

Ronald E . F r a n k University of Pennsylvania

T h o m a s Lodahl Cornell University

Philadelphia • University Of Pennsylvania Press

Copyright © 1968 by the Trustees of the University of Pennsylvania Library of Congress Catalog Card Number 67-26219

Second Printing, 1971

ISBN: 0-8122-7568-1 Printed in the United States of America ISBN: 0-8122-7568-1

To our charming normal deviates Bill Lauren

Beth Clair

Linda Andrea

who for a time lost their fathers to a study of personality and buying behavior.

PREFACE T h i s study is primarily concerned with assessing the relations between household socio-economic and personality characteristics and purchasing behavior for frequently purchased branded grocery items. T h e nature of the analytical problem posed by the study required a diverse set of intellectual and methodological resources: 1. O u r a c a d e m i c specialization spanned four areas, namely, econometrics, marketing, psychology and statistics. 2. A variety of multivariate statistical techniques, including factor, regression and discriminant analysis as well as simulation where used at different times during the course of the investigation. Because of the nature of the problem as well as the diverse methodology used in the analysis, the study may be useful to individuals possessing a wide range of interests, such as: 1. Researchers, regardless of their institutional affiliation or area of substantive interests, who are interested in the application of multivariate statistical techniques to the study of human behavior. 2 . M a r k e t i n g p r a c t i t i o n e r s w h o m a y b e p r i m a r i l y i n t e r e s t e d in t h e nature of our findings and their implications for formulating promotional programs. 3. Behavioral scientists who are interested in how personality affects behavior. T h e study was supported by the Graduate S c h o o l of Business, S t a n ford University, under a grant from the F o r d Foundation. T h e calculations were made at the S t a n f o r d University C o m p u t a t i o n C e n t e r . M a c h i n e t i m e w a s s u b s i d i z e d by t h e C e n t e r u n d e r N a t i o n a l S c i e n c e Foundation Grant N o . N S F - 9 P 9 4 8 . W e wish to thank the Advertising Research Foundation for their willing cooperation in making the data available to us. A debt of gratitude is also owed the J . W a l t e r Thompson C o . from whose panel these data originally c a m e . Without their willingness to m a k e the data available to the Advertising Research Foundation this study would never have been possible. A special word of thanks is due both Ingrid Kildegaard and Charles R a m o n d of the Advertising Research Foundation for their counsel and encouragement. A word of thanks is also due to Professors Philip Kotler and William Wells for their excellent reviews of the manuscript. Last, but not least, we owe a debt of gratitude to our wives, J u n e , Iris and J a n i c e , who had to put up with the three of us for a summer during which time we virtually lived at the S t a n f o r d Computation Center.

TABLE OF CONTENTS

CHAPTER I II

I N T R O D U C T I O N AND P L A N OF THE S T U D Y T H E D I M E N S I O N S OF P U R C H A S I N G BEHAVIOR

III

F U R T H E R A N A L Y S I S OF P U R C H A S I N G

V VI

11

BEHAVIOR

DIMENSIONS IV

I

45

T H E D E S C R I P T I O N OF PERSONALITY AND SOCIO-ECONOMIC STATUS

73

E X P E R I M E N T S O N THE A N A L Y S I S S A M P L E

89

R E L A T I O N S H I P S B E T W E E N PERSONALITY AND P U R C H A S I N G BEHAVIOR

107

A

TABLES

127

B

T H E EDWARDS

APPENDIX

SCHEDULE

PERSONAL

PREFERENCE 145

REFERENCES

163

INDEX

169

TABLES 2.1

2.2 2.3

2.4

2.5

2.6

2.7

2.8 2.9

2.10

2.11

2.12

3.1

3.2

Names of the Raw Variables Used to Describe Purchasing Behavior

18

Means, Standard Deviations and Ranges for 29 Purchasing Variables: Coffee, Analysis Sample

22

Means, Standard Deviations and Ranges for 29 Purchasing Variables: Tea, Analysis Sample

23

Means, Standard Deviations and Ranges for 29 Purchasing Variables: Beer, Analysis Sample

24

Simple Correlations Among 29 Purchasing Variables: Coffee, Analysis Sample

26

First Principal Component and Four Varimax Rotated Factor Loading: Coffee, Analysis Sample

29

First Principal Component and Four Varimax Rotated Factor Loading: Tea, Analysis Sample

30

First Principal Component and Four Varimax Rotated Factor Loading: Beer, Analysis Sample

31

Factor Labels for the First Principal Component and Varimax Rotated Factors: Coffee, Tea, Beer

33

Estimated Communalities, Reliabilities, and Specificities for 29 Purchasing Behavior Variables

36

Simple Correlations Among Four Varimax Rotated Factors: Coffee, Analysis Sample

40

Simple Correlations Among Scores for the First Principal Components and Four Varimax Factors, and Two Raw Variables: Coffee, Analysis Sample

41

Comparison of Differences in the Factor Loadings Among Coffee, Tea, and Beer

47

Means and Standard Deviations for Actual and Simulated Purchasing Statistics: Beer

57

T A B L E S (Continued) 3.3

Factor Loadings for Simulated Beer Purchasing Statistics Based on Analysis Sample: Actual Number of Trips

59

Factor Loadings for Simulated Beer Purchasing Statistics Based on Analysis Sample: Actual Number of Trips Times Three

60

Factor Loadings for Simulated Coffee Purchasing Statistics Based on Analysis Sample: Actual Number of Trips

62

Correlations Between Purchasing Variables for Coffee and Tea, and Coffee and Beer: Analysis Sample

68

Standardized Weights for the First Canonical Variate: Raw Purchasing Variables for Coffee and Tea: Analysis Sample

69

4.1

Definitions of Scales Used in the Edwards Test

76

4.2

Orthogonal and Varimax Factor Analysis

3.4

3.5

3.6

3.7

oftheEPPS

79

4.3

Results of the Stability Test on Five Canonical Variâtes

82

4.4

Weights for the First Two Canonical Variates: Analysis Sample

83

4.5

Means, Standard Deviations and Ranges for the Socioeconomic Variables: Analysis Sample

85

Factor Loadings and Communalities for Eleven Socioeconomic Variables: Analysis Sample

86

Explanatory Variable Set Used in Analysis Sample Regressions

88

The Proportion of Cases Included in the Regression, and Cases with Significant F-Ratios, by Type of Personality Characterization for First Stepwise Experiment

93

4.6

4.7

5.1

T A B L E S (Continued) 5.2

5.3

5.4

5.5

6.1

6.2

6.3

The Proportion of Cases Included, and Cases with a Significant F-Ratio, by Type of Personality Characterization for the Second Stepwise Experiment

95

The Proportion of Cases Included, and Cases with Significant F-Ratios for the First Experiment Standard Scores and the Remaining Personality Characterizations in the Third Experiment

97

Coefficients of Determination, F-Ratio and Number of Variables for Each Stepwise Regression Experiment

98

F-Ratio for Socioeconomic Status Variables Included in the First and Third Experiments

99

Predictive Power of Personality and Socioeconomic Variables Against Purchasing Behavior, Validation Sample Only

110

Number and Per cent of Significant (.05 Level) Predictions, by Type of Purchasing Behavior

111

N u m b e r and Per cent of Significant (.05 Level) Predictions, by Type of Product

111

6.4

Selected Results on Brand Behavior

115

6.5

Selected Results on Store Behavior

119

6.6

Selected Results on Activity Variables

122

A.I.I

Coffee Validation Sample Regression Coefficients and t-Ratios for Activity Variables

128

A. 1.2

Tea Validation Sample Regression Coefficients and t-Ratios for Activity Variables

130

Beer Validation Sample Regression Coefficients and t-Ratios for Activity Variables

132

A. 1.3

T A B L E S (Continued) A.2.1

A.2.2

A.2.3

A.3.1

A.3.2

A.3.3

Coffee Validation Sample Regression Coefficients and t-Ratios for Brand Behavior

134

Tea Validation Sample Regression Coefficients and t- Ratios for Brand Behavior

136

Beer Validation Sample Regression Coefficients and t-Ratios for Brand Behavior

138

Coffee Validation Sample Regression Coefficients and t- Ratios For Store Behavior

140

Tea Validation Sample Regression Coefficients and t-Ratios for Store Behavior

142

Beer Validation Sample Regression Coefficients and t-Ratios for Store Behavior Variables

144

A4

Variable Dictionary for Validation Sample Regressions

175

B. 1

De finitions of the 15 Scales of the Edwards Persona] Preference Profile

148

B.2

Reliability Coefficients for the 15 EPPS Variables

153

FIGURES 3.1

Flow Diagram of Simulation System

53

5.1

S-Transformation of Personality Variables

96

5.2

Quadratic Transformation of Socio-economic Variates

100

5.3

Square Root Transformation of Family Size Variable

100

I INTRODUCTION A N D PLAN OF T H E STUDY Consumer purchasing characteristics are often used as a partial basis for market segmentation. The purpose of this report is to examine some of the empirical underpinnings of such usage, in the context of data on three frequently purchased products: coffee, tea, and beer. We shall consider the relationship of purchasing behavior, as it is recorded in the diaries of a consumer panel, to the socio-economic and personality characteristics of households. Although the results cannot be formally generalized beyond the three products chosen for analysis, we believe that they have implications for a wide variety of frequently purchased food and household items.

THE PROBLEM OF MARKET SEGMENTATION Wendell Smith (1956) has defined the concept of market segmentation as follows: Segmentation is based on developments on the d e m a n d side of the market and represents a rational and more precise adjustment of product and marketing effort to consumer or user requirements.

In the language of

the economist, segmentation is disaggregative in its effects and tends to bring about recognition of several demand schedules where only one was recognized before.

The key concept in the quotation is that there are different and presumably identifiable groups in the population, each of which tends to

2

P U R C H A S I N G BEHAVIOR A N D P E R S O N A L A T T R I B U T E S

exhibit different behavior with respect to use of the product, attitudes toward brands, and response to promotional efforts. We have found it useful to think of these differences in terms of answers to the following questions: W h a t is the degree of v a r i a t i o n in c u s t o m e r sensitivity t o c h a n g e s in t h e f i r m ' s p r o m o t i o n a l policies a m o n g s e g m e n t s ? T h a t is, d o c u s t o m e r s in o n e s e g m e n t tend to r e s p o n d t o a g r e a t e r or lesser d e g r e e t h a n t h o s e in a n o t h e r to c h a n g e s in such p r o m o t i o n a l inputs as t h e r a t e of a d v e r t i s i n g a n d dealing, or retail price?

The two questions relate to differences among the average and marginal properties of the demand curves for different market segments, respectively. Given knowledge of these properties, it is possible to apply elementary price theory to the problem of allocating marketing effort among the various segments. Unfortunately, two problems stand in the way of such a simple solution to the problem of market segmentation. First, consumer groups do not come as tidy packages. There are not just a few " g r o u p s " of consumers, each exhibiting its own particular demand curve, active in the market. A model which implies that each individual consumer has his own unique set of requirements, and hence his own demand curve, would seem to come closer to the truth than one that postulates a relatively small number of homogeneous groups of consumers with all the members of a group having the same demand curve. Thus there is a problem of classifying consumers into groups or segments on the basis of their purchasing characteristics. This must be solved before the notion of "segments" can be fully defined. One approach would be to try to find groupings where the within-groups variance of demand properties is small relative to the between-groups variance. The resulting groups could then be called " m a r k e t segments." Problems of identifying particular segments are beyond the scope of this monograph. They are discussed by Duhamel (1966), who has applied the theory of Markov processes and statistical factor analysis in an attempt to provide the necessary methodology. The second problem facing those who would apply the segmentation concept is that of finding marketing strategies that can in fact be concentrated upon members of the target consumer groups. Suppose that a firm has decided to aim a promotional campaign at families who are heavy users of toothpaste—believing (rightly or wrongly) that concentration upon this group will provide greater returns than it would for average or light users in the population. Unfortunately, the firm will find that few if any promotional media cater to heavy users of toothpaste, as such. Rather, particular media attract families of various socio-economic groups at different stages of their lives and with different i n t e r e s t s and styles of life. T h e s p e c i f i c b e h a v i o r a l d i m e n s i o n

INTRODUCTION A N D PLAN OF THE STUDY

3

"heavy t o o t h p a s t e u s e " m u s t be t r a n s l a t e d into one of these m o r e general dimensions before a m e d i a s e g m e n t a t i o n strategy can be f o r m u l a ted. Of course the same problem must be faced for other types of promotional vehicles. T h e following list presents the m a j o r steps t h a t a r e necessarv t o put a segmentation s t r a t e g y into o p e r a t i o n . 1. Relevant behavioral d i m e n s i o n s have to be identified. M a n a g e ment must decide w h e t h e r it is interested in usage rates, patterns of preference for b r a n d s or varieties, responses t o particular types of p r o m o t i o n , and so on. T h e i m p o r t a n t idea is that t h e dimensions identified at this stage m u s t be a p p r o p r i a t e to the product in question and its use e n v i r o n m e n t and must be relevant to the m a r k e t i n g decision p r o b l e m s faced by the f i r m . 2. Given a n u m b e r of relevant behavioral dimensions, it is desirable t o identify o p e r a t i o n a l m e t h o d s for m e a s u r i n g t h e extent to which individual c o n s u m e r s exhibit the indicated characteristics. E x a m ples of such m e a s u r e s include total a m o u n t s p u r c h a s e d , shares of purchases by b r a n d or variety, and p r o p o r t i o n of purchases m a d e on a deal. While d i m e n s i o n s f o r s e g m e n t a t i o n t h a t cannot be operationalized and m e a s u r e d c a n be useful in subjective types of analysis, application of any of t h e powerful m e t h o d s of quantitative m a r k e t research requires t h a t m e a n i n g f u l m e a s u r e m e n t s be obtained. 3. T h e m e a s u r e m e n t tools a r e applied to a representative g r o u p of c o n s u m e r s and the results noted. For e x a m p l e , indices of usage, deal-proneness, etc., might be calculated for m e m b e r s of a cons u m e r panel or s u r v e y s a m p l e . T h i s p r o c e d u r e a l l o w s the researcher to d e t e r m i n e the relative n u m b e r s of people having each of the indicated characteristics and assess their potential i m p o r t a n c e for the f i r m ' s o p e r a t i o n s . It also provides the raw d a t a for step 4. 4. In many cases it is imperative t o find new variables t h a t are correlated with the relevant p u r c h a s i n g behavior indicators for the sample under study. P e r h a p s the new variables c a n be m o r e easily observed or m e a s u r e d when w o r k i n g with f u t u r e samples. M o r e c o m m o n l y they provide a link between the behavioral dimensions and published statistics on t h e p e n e t r a t i o n p a t t e r n s of m a r k e t i n g c o m m u n i c a t i o n channels. T h e new variables serve as surrogates for the original ones where t h e latter are difficult to m e a s u r e or where published sources d o not provide direct coverage of the behavioral variables. 5. Once the identification and analysis of segments is complete, the firm is in a position t o design a m a r k e t i n g strategy that is directed t o w a r d or will appeal t o a p p r o p r i a t e c o n s u m e r types.

4

P U R C H A S I N G BEHAVIOR A N D P E R S O N A L A T T R I B U T E S

This m o n o g r a p h is primarily c o n c e r n e d with point four in t h e above outline. W e seek t o d e t e r m i n e whether the items in a fairly b r o a d battery of socio-economic (we consider this t e r m t o include d e m o g r a p h i c variables) and personality variables are correlated with s o m e generally applicable m e a s u r e s of p u r c h a s i n g behavior. T h e correlates of p u r c h a s i n g behavior a r e i m p o r t a n t because they m a y in many cases be used as s u r r o g a t e s for t h e original behavioral variables. Consider again the case of the t o o t h p a s t e advertiser. If he desires to reach heavy t o o t h p a s t e users a n d k n o w s or c a n d e t e r m i n e that family size and stage in life cycle are i m p o r t a n t correlates of t o o t h p a s t e usage rates, he c a n m a k e use of published m e d i a statistics giving reach and frequency figures for these socio-economic classes. T h u s a media strategy can be devised without recourse to statistics relating media coverage directly t o t o o t h p a s t e usage. T h e correlates of p u r c h a s i n g behavior m a y be useful in a n o t h e r way as well. Situations arise where it is not possible to physically direct p r o m o t i o n a l e f f o r t s to m e m b e r s of the target m a r k e t segments. In these cases it may be desirable t o b r o a d c a s t the p r o m o t i o n t o a wider audience than would be suggested by the s e g m e n t a t i o n analysis and then design the message so t h a t the p r o m o t i o n will be self-selected by the relevant groups. Knowledge of the socio-economic correlates of purchasing behavior m a y be invaluable in designing p r o m o t i o n a l messages. M o r e o v e r , it is here t h a t personality correlates, if any should exist, c o m e into their own. If it were found t h a t — o t h e r things being e q u a l — heavy t o o t h p a s t e users exhibited identifiable personality traits it might be possible to tailor the p r o m o t i o n a l message so t h a t it would be m o r e visible or salient to people with these traits t h a n to the public at large. Finally, we should note t h a t k n o w l e d g e of t h e personality and socioe c o n o m i c correlates of c o n s u m e r buying behavior m a y be useful in a purely scientific sense, in addition t o whatever i m m e d i a t e practical use such knowledge can be put in the process of devising m a r k e t s e g m e n t a tion strategies. W e need t o k n o w m o r e a b o u t the underlying s t r u c t u r e of the purchasing process. W e m a y also be interested in e x a m i n i n g the behavioral implications of differences in socio-economic characteristics or personality p a t t e r n s — a s e c o n o m i s t s or psychologists r a t h e r t h a n in our role as m a r k e t i n g scholars. T h e correlates of purchasing behavior c a n shed light on these questions, t h o u g h of c o u r s e statistical analyses can never provide the last word on behavioral cause-effect relations. T h e needs of the scientist differ s o m e w h a t f r o m those of the m a r k e t ing practitioner—even in t e r m s of a purely statistical study like the one r e p o r t e d in t h i s m o n o g r a p h . T h e s c i e n t i s t i m p o s e s m o r e s t r i n g e n t d e m a n d s on the kind of theoretical s t r u c t u r e t h a t is built to provide rationales for the observed results t h a n does the practitioner who is primarily interested in using the correlates as surrogate variables for the actual behavioral dimensions. (In t h e latter case a t t e n t i o n r e m a i n s

I N T R O D U C T I O N A N D PLAN O F T H E S T U D Y

5

centered on the behavioral d i m e n s i o n s , whereas in the f o r m e r the socioe c o n o m i c and personality variables a r e b r o u g h t into the analysis as c o n s t r u c t s with intrinsic scientific interest.) T h e scientist is o f t e n m o r e concerned with the precise n a t u r e and validity of the statistical m e t h o d s utilized in the analysis, t h o u g h t h e r e a p p e a r s to be no reason why the practitioner should be any less c o n c e r n e d a b o u t this point. Finally, the scientist is usually satisfied if he can show t h a t there is a real relation between the two sets of variables (i.e., it is very unlikely t h a t the statistical results are d u e to chance). It o f t e n d o e s not m a t t e r if the observed relationship is r a t h e r weak, in t h e sense t h a t the predictor variables provide little reduction in the v a r i a n c e of the d e p e n d e n t variable, as long as the results are clearly " s i g n i f i c a n t . " T o the practitioner who w a n t s to use the p r e d i c t o r v a r i b l e s as s u r r o g a t e s for t h e a c t u a l behavioral dimensions, on the o t h e r h a n d the size of the correlation is usually very i m p o r t a n t indeed. A s u r r o g a t e t h a t a c c o u n t s for only a few percent of the variance of the target variable is hardly a surrogate at all. In this study we have not hesitated to shift our basis of interpretation f r o m that of the scientist to t h a t of the practitioner and back again, according to what a p p e a r s to be the most a p p r o p r i a t e in the particular context. F o r e x a m p l e , we d e v o t e time to discussing "signific a n t " relationships even where it is fairly obvious that the variables under study are not related strongly e n o u g h to serve the practical needs of m a r k e t s e g m e n t a t i o n .

PSYCHOLOGICAL VERSUS SOCIO-ECONOMIC SEGMENT IDENTIFIERS

T h e need for a c o n t r a s t of t h e relative efficacy of socio-economic and psychological characteristics (of which personality variables are one m a j o r class) is especially i m p o r t a n t in light of the growing dissatisfaction with socio-economic c h a r a c t e r i s t i c s as a basis for predicting the m e m b e r s h i p of c u s t o m e r s in p a r t i c u l a r m a r k e t segment g r o u p s . Several practitioners have cited the need for extending m a r k e t analyses to include various psychological characteristics as a basis for identifying c u s t o m e r buying behavior s e g m e n t s . N e w m a n (1957) presents a study c o n d u c t e d by M c C a n n E r i c k s o n in which heavy and light sewers were identified, in part, on the basis of their personality characteristics. T h e results of the investigation were used to develop an advertising c a m paign for one of the a g e n c y ' s clients. T h e agency has c o n d u c t e d a n u m b e r of studies which show t h a t certain types of p r o d u c t s a t t r a c t c u s t o m e r s with different personalities. T h e product categories studied include cigarettes, gasoline, a u t o m o b i l e s , sanitary napkins, dentifrices and hair tonics.

6

PURCHASING BEHAVIOR A N D PERSONAL ATTRIBUTES

The case favoring the idea that psychological characteristics should play a more prominent role in segmentation analyses is summed up in Grey Matter {1965): To most marketers "market segmentation" means cutting markets into slices—demographically, g e o g r a p h i c a l l y , a c c o r d i n g t o e c o n o m i c status. race, national origin, e d u c a t i o n , sex and other established criteria. But the idea of relating m a r k e t i n g strategy t o psychological differences a m o n g c u s t o m e r s has been s l o w t o g e r m i n a t e . W e call it " P s y c h o g r a p h i c Market S e g m e n t a t i o n . " . . . T h e profit potential in p s y c h o g r a p h i c segmentation of m a r k e t s is greater than is generally realized and those advertisers w h o see these o p p o r t u n i t i e s clearly, and exploit them skillfully, are scoring and will score triumphs, while those w h o c o n t i n u e t o dissipate c o m p e t i t i v e energy only on established notions of market s e g m e n t a tion may find t h e m s e l v e s o n a " m e t o o " m e r r y - g o - r o u n d .

In spite of the relatively strong positions that have been taken in favor of segmentation based on psychological characteristics of customers: (1) there has been relatively little research published that provides an empirical basis for these conjectures; and (2) that which has been published supports the notion that where effects of psychological characteristics do exist they are relatively small. Gottlieb (1958) found only slight differences in personality characteristics between heavy and light users of a proprietary medication. With respect to compulsiveness (one dimension of personality studied) he concluded that, " I t is probably not possible to address oneself to a clearly delineated compulsive group. What one should do is to address himself to the compulsiveness in all oj us." Evans (1959) and Westfall (1962) report an attempt to predict customer automobile ownership based on knowledge of socio-economic and personality characteristics. Like Gottlieb, they found only a modest association between personality and customer buying behavior. Findings of a similar nature are reported by Koponen (1960) and the Advertising Research Foundation (1964). Both studies examine the relationship of personality characteristics to customer buying behavior for various frequently purchased food products. In one analysis Koponen found that only 13 percent of the variance in total household consumption for a particular product could be explained on the basis of h o u s e h o l d s o c i o - e c o n o m i c and p e r s o n a l i t y c h a r a c t e r i s t i c s . In still another product class he could only explain 6 percent of the variance in total household purchases. The A R F study was based on household purchases of one-ply and two-ply toilet tissue. Their results were similar to Koponen's in that they were only able to explain 12 and 6 percent of the variance in household consumption of the two types of tissue. Both the Koponen and A R F studies used data from the J. Walter Thompson consumer panel. Though the products differ, this was the

INTRODUCTION AND PLAN OF T H E STUDY

7

same panel that served as the source for the data upon which our investigation was based. A description of the data is presented in the following section of this chapter. Other researchers report similar findings for other product categories such as baking products ( R u c h , 1965) and furniture, carpets, fabrics and liquors (Pessemier and T i g e r t , 1966). A series of reports of the s o c i o - e c o n o m i c and demographic correlates of different measures of buying behavior have also been published. T h e s e include studies of private-brand-proneness ( F r a n k and Boyd, 1965), total consumption ( F r a n k , M a s s y and Boyd, 1967), and brand loyalty ( F r a n k , Douglas and Polli, 1967). Each of these studies reports separate analyses of the s o c i o - e c o n o m i c and demographic correlates of buying behavior for each of 4 4 products. T h e multiple correlation coefficients regardless of product and/or study are seldom over .20. Though most recent work has attempted to develop a theoretical basis for predicting patterns of buying behavior (especially with respect to brand loyalty (Sheth, 1966 and T u c k e r , 1964) little progress had thus far been made. M e a s u r i n g the n a t u r e a n d e x t e n t o f t h e r e l a t i o n s h i p b e t w e e n household buying behavior and household socio-economic and/or personality characteristics (such as has been done in the above studies) presupposes that one has developed satisfactory ways of specifying each of the three sets of variables for the purpose of analysis. T o date there has been no systematic attempt t o analyze alternative variable specifications as they relate to the problem of identifying different patterns of customer buying behavior. W e first present a detailed analysis of the measurement of each set of variables. After appropriate measures have been worked out attention is turned to the problem of measuring their interrelationships. T h e remainder of this chapter discusses the data used as a basis for this investigation, the design of the study, and the organization of the main body of the monograph.

P R E L I M I N A R Y D E S C R I P T I O N O F DATA T h e J . Walter T h o m p s o n c o n s u m e r panel served as the source of the data used in this investigation. Household purchase records from July 1956 through J u n e 1957 for three frequently purchased grocery products (beer, coffee, and tea) were included in t h e study. F o r each purchase of a given product m a d e by a household we had a record of the month and year during which the purchase was made, the brand, the quantity, the package size, the cost of the transaction, and whether or not a deal was involved. A " d e a l " is defined as any special offer made to a household, such as a cents-off label or a coupon offering a discount or premium. T h e r e were a total of 2 3 0 , 0 0 0 purchase transactions

8

P U R C H A S I N G BEHAVIOR A N D P E R S O N A L A T T R I B U T E S

for all three products combined, which in turn were generated from monthly diaries of approximately 5,000 households. Socio-economic and personality information were also available for each household in the sample. The socio-economic variables included: (1) the sex, age, income, and occupation of the head of the household; (2) the level of education for both the husband and wife; (3) the ownership of home, TV, and automobiles; and (4) the geographic region and market size in which the household was located. The personality characteristics were based on the results of the Edwards Personal Preference Schedule, which was separately submitted to both the husband and wife in each household participating in the panel. Details of the method by which the test was administered and a discussion of the general personality statistics exhibited by the total sample may be found in Koponen (1957). The Edwards test provides measures of the following personality needs: achievement, compliance, order, exhibition, autonomy, association, analysis, dependence, dominance, self-depreciation, assistance, change, endurance, heterosexuality, and aggression. A more detailed description of both the socio-economic and personality variables can be found in Chapter 4. The analyses which are reported in this monograph are based on about 3,500 of the originai 5,000 households included in the J. Walter Thompsons sample. The reduction was due to the fact that they control deck which reported the frequency with which the monthly purchase diaries (used for recording purposes) were turned in by each household was lost before we acquired the data. This was important because for an analysis like ours one should include only those households which turned in all of the monthly diaries for the period under study. The inclusion of other households could result in confounding the effect of personality and socio-economic characteristcs on buying behavior with their effect on panel reporting. This problem is especially acute when it comes to analyzing the association between household socio-economic and personality characteristics and total consumption for a given product. If households with incomplete reporting histories are included in the analysis they would be erroneously treated as having lower purchase rates than is actually the case. This might have the tendency to bias the resulting measures of association. Given the loss of the relevant reporting control decks, we were forced to sacrifice some data in order to assure ourselves that our analysis would be relatively free of reporting bias. To do this we created a chronological record of purchases for each household based on its combined purchasing behavior for all three products. If one or more purchases were reported in a given month by a household, we could be quite sure that it turned in a diary for that month. We eliminated all of the households whose combined purchase history contained more than

I N T R O D U C T I O N A N D PLAN O F T H E S T U D Y

9

one month without any purchases of coffee, tea, or beer. This resulted in the elimination of about 1,500 households. The analyses presented for each product category are based on fewer than the 3,500 households included in the study because: (1) none of the products were purchased by 100 percent of the sample; (2) some of the analyses required the deletion of households with low purchase rates (even after the adjustment for incomplete reporting); and (3) we split the sample into two parts at the initiation of the study for reasons which will be discussed in the next section of this chapter. The actual number of households included in each category, at each stage of the analysis, is reported at appropriate points in the presentation of results. There are a number of other sources of bias that are generally associated with consumer panels. They are discussed by Boyd (1960) and in a report by the Market Research Corporation of America (1952) and hence are not included as part of this monograph. T H E DESIGN O F T H E I N V E S T I G A T I O N At the start of our investigation there was no body of theory sufficiently well developed to provide a guide for specifying the nature and extent of relationships between various aspects of household buying behavior (e.g., total consumption of a product or brand loyalty) and household socio-economic and personality characteristics. In addition, we were not sure what set of measures would provide a complete but parsimonious description of each of the three sets of variables. We did, however, have some thoughts about where to start, though we were not sure how many experiments would have to be tried before we would arrive at what, in our judgment, would be a satisfactory basis for specifying the set of interrelationships we had chosen to study. We recognized that the process of intensive experimentation with a given set of data can easily lead to erroneous interpretations. Even if a set of data is in fact uncorrelated sequences of random numbers, for example, the performance of a sufficient number of experiments using different combinations of explanatory variables will eventually lead to a statistically significant association. In order to protect ourselves against this possibility we split our data base into two parts: (1) an "analysis sample" consisting of approximately one-third of the households in the screened sample; and (2) a "validation sample" consisting of the remaining two-thirds of the sample respondents. Households were assigned at random to each of the two groups. (A given family is in the same sample for runs on all three product classes.) Our experiments (reported in Chapters 2 through 5) were performed on the analysis sample. Once we had developed a reasonable model we provided a check as to the stability of our results by re-estimating the relationships involved using the validation sample.

10

P U R C H A S I N G BEHAVIOR A N D P E R S O N A L A T T R I B U T E S

T h i s p r o c e d u r e p r o v i d e s an i m p o r t a n t f o r m of i n s u r a n c e a g a i n s t arriving at spurious conclusions, even where the data in the analysis sample receive considerable manipulation. Chapters 2 and 3 report the results of an intensive analysis of household buying behavior by itself, for each of the three products. This part of the study was aimed at developing a set of measures of buying behavior that would be both parsimonious and efficient. Chapter 4 presents analyses aimed at specifying household personality and socioeconomic variables. The next chapter (5) presents a series of experiments, based on the analysis sample, which are used as the basis for developing a final model of the relationship between the various measures of customer buying behavior for each product, developed in Chapters 2 and 3, and the socio-economic and personality variables as developed in Chapter 4. The series of experiments in Chapter 5 resulted in some further changes in the specification of the variables in our model. Chapter 6 reports and discusses the results from the final model as estimated separately for the validation sample.

II THE DIMENSIONS OF PURCHASING BEHAVIOR What is meant by "purchasing behavior?" The two words that make up the term suggest that we should expect to deal with matters that are: first, related to buying; and second, observable, to the extent the term may be applied to self-generated reports of overt behavior. In the present research we naturally limit ourselves to the study of overt behavior based on the acquisition of frequently purchased food products—in particular, coffee, tea, and beer. This information is contained in the panel data which form the basis for our study. Even with these constraints, however, the notion of "purchasing behavior" is not defined well enough to permit systematic analysis. A great many summary statistics may be culled from the basic panel data, and each of them has a claim to represent some aspect of the phenomenon we wish to study. The list of raw statistics used in this study, and a review of previous work involving most of them, will be presented in a later section of this chapter. The raw statistics which are used to summarize a given family's purchasing behavior over a period of time may be classified into a number of different categories. The percentage of a family's total consumption which is devoted to its favorite brand of a product may be considered as a measure of " b r a n d loyalty" or the total number of units of the product purchased described as "activity," for example. These summary categories represent alternative dimensions, or ways of looking at the overall notion of purchasing behavior. Brand loyalty, store loyalty, and total consumption have tended to be the major dimensions considered in the literature.

II

12

P U R C H A S I N G BEHAVIOR A N D P E R S O N A L A T T R I B U T E S

Dimensions t h a t a r e defined in t e r m s of a specific s u m m a r y variable, like total c o n s u m p t i o n , m a y be perfectly acceptable for m a n y purposes. The a p p r o a c h t a k e n in this study, however, is to consider the m a t t e r of defining d i m e n s i o n s f r o m the beginning in order to check the efficacy of the established m e a s u r e s . Several f a c t o r s favored this course of action. First, little is k n o w n a b o u t t h e properties of the univariate dimensions m e n t i o n e d above. T h u s we felt t h a t the intercorrelations between t h e m a n d their stability over t i m e and across p r o d u c t categories ought t o be investigated. S e c o n d , t h e established m e a s u r e s d o not include variables that have been f o u n d useful in recent research: this applies p a r t i c u l a r l y to statistics based on the n u m b e r a n d length of runs for the s a m e b r a n d or store. W e feel t h a t these statistics m e a s u r e something i m p o r t a n t a n d t h a t they m i g h t be related to personal attributes. Third, s d m e of the established m e a s u r e s do not a p p e a r t o exhaust the i n f o r m a t i o n implied by their own " d e f i n i t i o n s . " If b r a n d loyalty is based only on t h e share of p u r c h a s e s a family devotes t o its favorite b r a n d , f o r e x a m p l e , i n f o r m a t i o n o n t h e d i s t r i b u t i o n of p u r c h a s e s a m o n g its second, third, and s u b s e q u e n t b r a n d s is ignored. Yet this aspect of behavior m a y be a relevant dimension of what we m e a n by brand loyalty. T h e p u r p o s e of this c h a p t e r is to e x a m i n e the different d i m e n s i o n s that are c o n t a i n e d in a fairly large set of variates defining different aspects of p u r c h a s i n g behavior. M u l t i p l e factor analysis is the principal technique to be utilized in the e x a m i n a t i o n . T h e results a r e used to define the p a r t i c u l a r d i m e n s i o n s of purchasing behavior t h a t will be predicted using personality a n d socio-economic i n f o r m a t i o n . In addition, we hope t h a t they will shed light on what have been poorly illuminated aspects of p u r c h a s i n g behavior m e a s u r e m e n t .

P R E V I O U S W O R K ON T H E D E V E L O P M E N T O F H O U S E H O L D PURCHASING STATISTICS T h e following review r e p o r t s t h e work of a u t h o r s who a t t e m p t e d to define a n d m e a s u r e the d i m e n s i o n s of brand a n d s t o r e loyalty on a household-by-household basis. In effect, their results set the stage for our own a t t e m p t s to define r e a s o n a b l y c o m p l e t e but e c o n o m i c a l purchasing behavior variables, as r e p o r t e d in this and the following c h a p ter. The review s u m m a r i z e s references t h a t were published prior t o the s u m m e r of 1964—the ones that were available before the beginning of this study. T h e review is not exhaustive; in particular, space d o e s not permit us to deal with studies t h a t have focused primarily on probability models for b r a n d or store choice r a t h e r t h a n on the d e v e l o p m e n t of household-specific p u r c h a s i n g descriptor variables. The earliest investigation concerned with the description and analysis of raw p u r c h a s e variables like those used in t h e present study was

THE DIMENSIONS OF P U R C H A S I N G

BEHAVIOR

13

conducted by G e o r g e A . Brown (1952-53). Brown analyzed purchase records f r o m the Chicago Tribune's c o n s u m e r panel for one hundred households during 1951, for each of the following nine product categories: margarine, toothpaste, coffee, all-purpose flour, s h a m p o o s , readyto-eat cereals, h e a d a c h e remedies, soaps and sudsers, and concentrated orange juice. Brown's principal interest was in determining the extent to which households were loyal to brands within the context of a particular product class. Although he tried a number of ways to measure the household purchasing dimensions for particular products (number of brands purchased and proportion of purchases accounted for by the brand purchased most frequently), the following definition presents the general principles of the approach upon which he put primary emphasis: "Any family making five or more purchases during the year was placed in one of four basic categories depending upon the purchase pattern shown . . . : 1. Family showing undivided loyalty bought brand A in the following sequence: A A A A A A 2. Family showing divided loyalty bought brand A and B in the following sequence: A B A B A B 3. Family showing unstable loyalty bought brand A and B in the following sequence: A A A B B B 4. Family showing no loyalty bought b r a n d s A, B, C, E, and F in the following sequence: A B C D E F " T h o u g h Brown's definition permits one to roughly classify groups of households, it does not provide a m e a s u r e of how m u c h m o r e or less loyal a given household is relative to a n o t h e r . An additional problem arises when one tries to use Brown's m e a s u r e to c o m p a r e the degree of brand loyalty exhibited in different product classes. For example, the n u m b e r of consecutive purchases which he required to classify a household as loyal to a particular brand is either two or three for s h a m p o o , two for flour, three for headache remedies, and four for concentrated orange juice. C h a n g e s such as this obviously will effect any interproduct comparisons. Lastly, it seems likely that Brown's system of classification is apt to suffer from low reliability. T w o researchers, both a r m e d with his definition, are apt to array households in s o m e w h a t different ways. Ross M. C u n n i n g h a m (1955, 1956) developed and illustrated the use of a somewhat different set of " b r a n d loyalty" measures. The m a j o r portion of his work concentrates on the analysis of d a t a for a 66-family g r o u p from the Tribune panel who purchased a certain threshold quantity of each of the following product categories during the period 195153: toilet soap, scouring cleanser, regular coffee, canned peas, m a r g a rine, frozen orange juice, and headache tablets. His operational defini-

14

P U R C H A S I N G BEHAVIOR A N D P E R S O N A L A T T R I B U T E S

tions of "brand loyalty" were based on household brand share (i.e., the proportion of purchases that a household devoted to a particular brand during a given period of time). The measures were defined as follows: 1. Single brand loyalty: The proportion of total household purchases represented by the leading single brand used by the household. 2. Dual brand loyalty: The proportion of total household purchases represented by the two leading single brands used by a given household. 3. Triple brand loyalty: The proportion of total purchases represented by the three leading brands purchased by a given household. In addition, he computed brand share measures which are identical to the first and second above except for the fact that purchases involving special price inducements or deals were excluded from the computation. Cunningham's definitions went a long way toward mitigating the principal limitations associated with Brown's approach to the problem. They provide a somewhat better means of making household and product comparisons. In addition they bypass the problem of subjective ratings. In the work that has been thus far discussed the principal purpose of the research was to provide a set of descriptive statistics with respect to selected aspects of buying behavior for frequently purchased food and household items. The work of Brown and Cunningham served to provide a map of this previously uncharted territory and has helped to stimulate interest in this area of investigation. Given this map, Alfred A. Kuehn (1958, 1962) and Ronald E. Frank (1960, 1962) were ready to take the next step. They concerned themselves with explaining variations in buying behavior from household to household by means of aggregative analyses as well as with the develo p m e n t of m e a n i n g f u l h o u s e h o l d - s p e c i f i c p u r c h a s i n g s t a t i s t i c s . 2 Kuehn's study was based on the purchases of 650 Tribune panel households for frozen orange juice during the period 1951-53. He found that a model equivalent to the generalized form of the Bush-Mosteller (1955) stochastic learning model appears to describe consumer brand shifting on an aggregative basis. Frank's work was partially motivated by the implications of the Kuehn learning hypothesis. Dealing with Tribune data on the regular coffee purchasing behavior of 536 families for the period 1956-58, he examines the possibility that the "learning" which seems to be apparent in aggregative measures could be caused by heterogeneity of the purchase probabilities for families in the panel, given that each family really chooses brands according to a simple Bernoulli process with constant probabilities. The aspect of his work that is relevant for our purposes here involves the use of household-specific statistics that are

T H E D I M E N S I O N S O F P U R C H A S I N G BEHAVIOR

15

related to the order and stationarity of individuals' purchasing processes. They include the statistics for run length, number of runs, and the standard normal deviate for assessing the departure of a brand switching process from a zero order stationary base that will be considered later in this chapter. Frank points out that two households purchasing the same number of brands (one of the measures used by Brown) and concentrating the same share of their total purchases on the brand purchased most often (one of the measures advanced by Cunningham) could still have fundamentally different brand purchasing patterns. Consider the following purchase sequences for two different households: 1: A A A A A B B B B B 2: A A A B B A B B B A Both households made ten purchases. Both split their purchases evenly between the same two brands. However one of them made an abrupt and sudden switch while the other seems to vacillate back and forth between the two brands. If one measures the probability that a household will buy a certain brand by the relative frequency with which the household bought the brand during the period under investigation both of these households have an apparent purchase probability for brand A of 0.5. Household No. 1 appears to have a much less stable probability than No. 2. Based on the observed pattern it could well have gone from 1.0 to 0.0 half way through the purchase sequence, whereas the probability of the second household buying A may well have remained constant during the entire period. Given a household's rate of purchasing and the share of purchases it concentrates on a particular brand, one can use the average brand run length, where a run is defined as a set of consecutive purchases of the same brand, as a measure of the extent to which a household's probability of purchasing the brand in question has remained stable. A long average run length (given a household's total consumption rate and brand purchase probability) suggests that the underlying probability is relatively unstable. Frank used this statistic (average run length) as one basis for characterizing the brand purchasing patterns for individual households. In addition, he did some exploratory work with respect to the distribution of the underlying purchase probabilities across households and their stability from year to year. He also used statistical run theory as a basis for contrasting the actual number of runs generated by a household with the expected number given the household's consumption rate and relative brand purchase frequencies. The purpose of the measure is essentially the same as that for the average run length measure described above. Cunningham (1961) broadened the spectrum of analysis by focusing upon store as opposed to brand loyalty. Using the same measures he had developed earlier for brands, Cunningham performed a detailed

16

PURCHASING BEHAVIOR AND PERSONAL ATTRIBUTES

analysis of the store loyalties of fifty families, based on the purchases they made for seven product categories during 1956. The products were: canned corn, canned fruit cocktail, vegetable shortening, canned p e a c h e s , r e g u l a r c o f f e e , white b r e a d , a n d m a r g a r i n e . C u n n i n g h a m found that within the period of a year a household's store shopping habits remained relatively stable. He also found that there appeared to be no relationship between the degree of store or brand loyalty exhibited by a household and its volume of consumption. By 1962 the number of purchasing measures which purportedly describe different dimensions of buying behavior for frequently purchased food and household items had grown rather large. Brown, Cunningham, and Frank each had developed different measures of buying behavior which could be applied to brand or store purchasing behavior. Then William T. Rice (1962) conducted an investigation of the extent to which the different statistics actually appear to measure the same underlying phenomena. Rice g e n e r a t e d t w e n t y - t w o m e a s u r e s of b u y i n g b e h a v i o r , which covered those used by previous researchers as well as several variants of his own. His results were based on a sample of one hundred households from the Tribune panel in each of five product categories: margarine, toilet soap, canned corn, fruit cocktail and tuna fish, for the year 1958. He subjected the data to a factor analysis and found that the twentytwo raw variables collapsed into seven underlying dimensions. Two sets of three factors each could be used to describe brand and store loyalty. The three factors were the same in both cases, namely household share of purchase loyalty, run length loyalty, and run distribution loyalty. Rice's run distribution factor was closely associated with the proportion of runs that are longer than one purchase in length. His seventh factor was a measure of activity which included such raw variables as the number of shopping trips and the number of units purchased. The factor analyses reported in this chapter represent an extension of Rice's fundamental idea that sets of correlated purchasing statistics can be combined into meaningful summary variates. T H E RAW P U R C H A S I N G V A R I A B L E S Before embarking on our analysis of purchasing behavior, it is necessary to define the raw variables that are used to summarize the i n f o r m a t i o n c o n t a i n e d in the f a m i l y - b y - f a m i l y p u r c h a s e r e c o r d s reported by the panel. The statistics are based on the following data that are included in each purchase record: 1. The serial number of the family reporting the purchase. 2. The month and year in which the purchase was made. 3. The day of the month on which the purchase was made. (Day of month was punched only on the cards containing the beer pur-

T H E D I M E N S I O N S OF P U R C H A S I N G

BEHAVIOR

17

c h a s e r e c o r d s . T h e r e is a p r e s u m p t i o n t h a t the c a r d s were arranged in chronological order within months for the other two products as well, and the data decks were handled so as to maintain this integrity. The hypothesis that the days have been inadvertently randomized will be considered and rejected in Chapter 3.) 4. The code number for the brand purchased. 5. The code number of the store at which the purchase was made. 6. The code number of any special deal which may have been reported in connection with the purchase. 7. The size of the purchase, in some convenient units of product (e.g., pounds or ounces). 8. The total cost of the purchase. In addition, the purchase records show the particular subtype of product that was purchased on the indicated shopping trip. The present analysis is based on regular coffee (not instant), bag and leaf tea (not instant), and beer (not ale). Measures based on this raw purchase information fall into three classes. There are statistics showing how much of a certain type of activity the family engaged in during the sample period. Second, the proportion of one kind of activity to another may be considered. Third, the order in which various events occur may be important. The group of twenty-nine such statistics discussed in the following paragraphs was obtained by using the Household Purchasing Characteristics Generation System computer programs reported by Frank and Massy (1965). DEFINITIONS.

Table 2-1 gives the names of the variables used in the study, classified by type of measure. Each has been assigned a six-digit neumonic identification code to aid in the interpretation of results; these codes are given in the first column of the table. Part A of the table needs little explanation. The units in which the products were measured were defined on the basis of convenience and computer format considerations. A trip is defined as an occasion when at least one unit of the product in question is purchased. Every card in the panel purchase data file represents one, and only one, trip. Brands were separately coded in almost all cases. Stores are coded separately as to major and some minor chains, but not in terms of which store in a given chain was patronized. That is, the record may show that the purchase occurred in an A & P store, but it does not distinguish between the separate outlets for the chain. The statistics based on ratios are also straightforward. The average number of units purchased per trip is a measure of either "lumpiness" of purchases, total consumption, or a combination of both. The propor-

PURCHASING BEHAVIOR AND PERSONAL ATTRIBUTI ES

18

T a b l e 2-1 NAMES OF THE RAW VARIABLES USED TO PURCHASING BEHAVIOR

A.

S t a t i s t i c s b a s e d on s i m p l e a g g r e g a t i o n UNITS

(3)*

UNIDE

(4)

TRIPS

(5)

B R A N D (1) S T O R E (6)

B.

DESCRIBE

N u m b e r of u n i t s of t h e p r o d u c t purchased.** N u m b e r of units of t h e p r o d u c t p u r c h a s e d o n s o m e k i n d of d e a l . N u m b e r of s h o p p i n g t r i p s on which the product was purchased. N u m b e r of d i f f e r e n t b r a n d s p u r c h a s e d . N u m b e r of d i f f e r e n t s t o r e s at which the p r o d u c i was p u r c h a s e d .

S t a t i s t i c s based on r a t i o s of d i f f e r e n t k i n d s of activity UNPTP

(2)

SH1 LB

(21)

SH2LB

(24)

SH3LB

(27)

SH ILS

(16)

SH2LS

(19)

SH3LS

(20)

A v e r a g e number of units of the product p u r c h a s e d on o n e s h o p p i n g t r i p . P r o p o r t i o n of the u n i t s of the p r o d u c t d e v o t e d to the f a v o r i t e b r a n d . P r o p o r t i o n of t h e u n i t s of t h e p r o d u c t for the second brand. P r o p o r t i o n of the u n i t s of the p r o d u c t for t h e t h i r d b r a n d . P r o p o r t i o n of t h e u n i t s of the p r o d u c t for the favorite store. P r o p o r t i o n of t h e u n i t s of the p r o d u c t for the second store. P r o p o r t i o n of the u n i t s of the p r o d u c t for the third store.

C.

S t a t i s t i c s b a s e d on t h e o r d e r of p u r c h a s e s NOBRR NBRXD

(7) (8)

NBRG1

(9)

N O S T R (10) N S R G 1 (11)

Number Number purchase Number purchase Number Number purchase

of b r a n d r u n s . of b r a n d r u n s g r e a t e r t h a n o n e long. of b r a n d s r u n s g r e a t e r t h a n o n e long. of s t o r e r u n s . of s t o r e r u n s g r e a t e r t h a n o n e long.

THE D I M E N S I O N S O F P U R C H A S I N G B E H A V I O R

19

Table 2-1 (Continued) A R L B U (12) A R L X D (13)

A R L B T (14) A R L S T (15) A LUIS

(17)

ALT1S

(18)

ALU IB (22)

A L T IB

(23)

A L U 2 B (25) ALT2B

(26)

A L U 3 B (28) ALT3B

(29)

SND1S

***

SND1B

***

Average length of brand runs, measured in units of product. Average length of brand runs that were terminated by a deal, measured in units. Average length of brand runs, measured in trips. Average length of store runs, measured in trips. Average length of run for the first loyal store, as taken against all others, measured in units. Average length of run for the first loyal store, measured in trips. Average length of run for the first loyal brand, measured against all others, in units. Average length of run for the first loyal brand, in trips. Average length of run for the second loyal brand, in units. Average length of run for the second loyal brand, in trips. Average length of run for the third loyal brand, in units. Average length of run for the third loyal brand, in trips. A standard normal deviate which compares the actual and expected number of runs, for the family's favorite store. (See text for definition.) A standard normal deviate which compares the actual and expected number of runs, for the family's favorite brand. (See text for definition.)

'Sequence numbers used in tables in Chapters 2 and 3. **The following units were used: coffee, pounds; tea, ounces; beer, pints. ** T h e s e variables were added after the main part of the study had been completed and so do not appear in any of the tables in Chapters 2 or 3. They are discussed in Chapter 6.

20

P U R C H A S I N G BEHAVIOR A N D PERSONAL ATTRIBUTES

t i o n of u n i t s of t h e p r o d u c t d e v o t e d t o the first s t o r e a n d b r a n d , e t c . , a r e t h e s t a t i s t i c s c o m m o n l y l a b e l e d as m e a s u r e s of p r i m a r y , s e c o n d a r y , and tertiary b r a n d and store loyalty. T h e s t a t i s t i c s b a s e d o n r u n s r e q u i r e s o m e i n t e r p r e t a t i o n . A r u n is d e f i n e d as a n y c o n s e c u t i v e s e q u e n c e of p u r c h a s e s of t h e s a m e b r a n d at o n e s t o r e . T h a t is, a r u n is t e r m i n a t e d a n d a new o n e b e g u n w h e n e v e r t h e f a m i l y c h a n g e s its s t o r e o r b r a n d . T h e m i n i m u m l e n g t h of a r u n is o n e u n d e r t h i s d e f i n i t i o n , a n d t h e m a x i m u m is l i m i t e d o n l y by t h e t o t a l n u m b e r of p u r c h a s e s m a d e by t h e f a m i l y . T h e s t a t i s t i c s f o r t o t a l n u m ber of r u n s is d e t e r m i n e d by c o u n t i n g t h e n u m b e r of s w i t c h e s of t h e relevant t y p e m a d e by t h e f a m i l y . T h e n u m b e r of r u n s t e r m i n a t e d by a deal is d e t e r m i n e d by c o u n t i n g t h e n u m b e r of r u n s f o r w h i c h t h e p u r c h a s e t h a t b r o k e t h e c o n s e c u t i v e s t r i n g was m a d e u n d e r s o m e kind of d e a l . T h e d e f i n i t i o n f o r t h e n u m b e r of runs g r e a t e r t h a n o n e p u r c h a s e l o n g is a n o b v i o u s e x t e n s i o n of t h e a b o v e . F i g u r e s f o r a v e r a g e l e n g t h of r u n w e r e d e t e r m i n e d b y c a l c u l a t i n g the l e n g t h of e a c h r u n , m e a s u r e d in e i t h e r units o r t r i p s as d e s i r e d , a n d a v e r a g i n g o v e r t h e n u m b e r of r u n s f o u n d for t h e f a m i l y . A v e r a g e r u n l e n g t h s f o r p a r t i c u l a r b r a n d s o r s t o r e s (e.g., f o r t h e first loyal s t o r e ) , w e r e o b t a i n e d by t a k i n g t h e a v e r a g e o v e r only t h o s e r u n s m a d e u p of p u r c h a s e s of t h e d e s i r e d t y p e . In g e n e r a l , the run l e n g t h i n f o r m a t i o n is a m e a s u r e of t h e s t a b i l i t y of p u r c h a s i n g b e h a v i o r t h r o u g h t i m e . It is a d d r e s s e d t o t h e q u e s t i o n of w h e t h e r f a m i l i e s switch a m o n g b r a n d s or s t o r e s r e g u l a r l y , o r w h e t h e r t h e y t e n d t o stay with e a c h s e l e c t i o n f o r a p e r i o d of t i m e b e f o r e g o i n g o n t o t h e next. A t h e o r e t i c a l a n a l y s i s of t h e e x p e c t e d b e h a v i o r of t h e r u n s t a t i s t i c s u n d e r d i f f e r e n t a s s u m p t i o n s a b o u t t h e u n d e r l y i n g b r a n d s w i t c h i n g p r o c e s s is p r o v i d e d in C h a p t e r 3. The two standard normal deviates ( S N D ) provide additional summ a r y i n f o r m a t i o n about families' b r a n d and store switching processes. T h e S N D c o m p a r e s t h e a c t u a l n u m b e r of r u n s (r) t o t h e e x p e c t e d n u m b e r of r u n s ( M ) f o r t h e b r a n d o r s t o r e in q u e s t i o n . It is d e f i n e d as follows:

SND =

r + .5 - M

2njn2 M

+ 1

2 n 1 n 2 ( 2 n 1 n 2 - n) cr = r

n

(n - 1)

T H E D I M E N S I O N S O F P U R C H A S I N G BEHAVIOR

21

where rt\ is the number of purchases of the given brand (store), n 2 is the number of purchases of other brands (stores), and n = ni + m. If the underlying choice process is stationary and of zero order (Bernoulli) then the S N D is normally distributed with zero mean and unit variance. The two S N D variables were added after the m a j o r part of this study had been concluded, for reasons to be discussed in Chapter 6. Therefore, they do not appear as part of any of the analyses in this chapter or the next, or in Chapter 5. MARGINAL DISTRIBUTIONS O F T H E VARIABLES.

Tables 2-2, 2-3, and 2-4 present the means, standard deviations, and ranges of the 29 variables, for each product class. All the findings are based on only those families in the analysis sample that made a minimum of five or more purchases for the product category during the year covered by the data. Thus the families for whom next to no data were available, and for which the resulting summary statistics would be very unstable, were excluded from the analysis. The final sample sizes are somewhat smaller than we anticipated at the outset of the study, but they were deemed large enough to produce stable correlation statistics. Comparison of the tabulated means show that the average family in the three analysis samples purchased 33 pounds of coffee, 57 ounces of tea, and 189 pints of beer. These figures exclude the families that have been deleted from the sample because they failed to meet the requirement of at least five shopping trips for the relevant product. The relations between the standard deviations and the means for this variable suggest that the volume of purchases is more stable for coffee than for either tea or beer; this finding is reasonable, given the consumption patterns for the three products. This conclusion is also borne out by the ranges for total consumption of the three products: the maximum purchase for coffee is about four times the mean, compared to values of 4.8 and 7.4 for tea and beer, respectively. A direct comparison of the amounts of activity for the three products that were generated by the sample families is provided by the statistic for average number of trips (variable 5). The average is 21.1 for coffee, 22.8 for beer, and only 11.5 for tea. Thus the tea sample is based on a substantially smaller number of purchases for each family than is the case for either of the other two products. This fact may be expected to have a bearing on the stability of some of the other purchasing behavior variables. The mean numbers of brands and stores utilized in the three samples are not very different, although there is some tendency for coffee purchases to be spread out a m o n g more brands than is the case for tea or beer. The tendency receives additional support from the fact that the average primary brand share for coffee is the smallest among the three

22

PURCHASING BEHAVIOR A N D PERSONAL ATTRIBUTES

Table 2-2 MEANS, S T A N D A R D DEVIATIONS, A N D R A N G E S FOR 29 P U R C H A S I N G VARIABLES: COFFEE, A N A L Y S I S S A M P LE

Variable No. Name

Average

Standard Deviation

Max

Range Min

1 2 3 4 5

BRAND UNPTP UNITS UNIDE TRIPS

3.3 1.7 33.0 7.8 21.1

1.9 1.0 19.9 12.1 12.8

12.0 7.6 123.0 113.0 78.0

1.0 0.7 4.0 0.0 5.0

6 7 8 9 10

STORE NOBRR NBRXD NBRG1 NOSTR

1.7 7.5 5.3 3.5 3.8

0.8 6.4 4.5 2.8 4.2

5.0 37.0 27.0 20.0 27.0

1.0 1.0 1.0 1.0 1.0

11 12 13 14 15

NSRG1 ARLBU ARLXD ARLBT ARLST

2.4 10.8 9.9 6.0 11.7

2.3 15.3 13.3 8.1 11.5

20.0 123.0 87.0 50.0 62.0

1.0 1.0 0.5 1.0 1.0

16 17 18 19 20

SHI LS ALUIS ALT1S SH2LS SH3LS

0.885 21.3 12.9 0.094 0.012

0.2 18.7 11.4 0.1 0.0

1.0 123.0 62.0 0.5 0.3

0.3 1.2 1.0 0.0 0.0

21 22 23 24 25

SH1LB ALU1B ALT1B SH2LB ALU2B

0.727 13.1 7.3 0.145 2.6

0.2 15.6 8.3 0.1 2.3

1.0 123.0 50.0 0.5 33.0

0.2 1.0 1.0 0.0 0.5

26 27 28 29

ALT2B SH3LB ALU3B ALT3B

1.7 0.048 1.8 1.4

1.2 0.1 1.0 0.6

17.0 0.3 16.0 9.0

1.0 0.0 0.5 1.0

Sample size - 670.

T H E D I M E N S I O N S O F P U R C H A S I N G BEHAVIOR

Table 2-3 MEANS, STANDARD DEVIATIONS, AND RANGES FOR 29 PURCHASING VARIABLES: TEA, ANALYSIS SAMPLE

Variable Name No.

Average

Standard Deviation

Max

Range Min

1 2 3 4 5

BRAND UNPTP UNITS UNIDE TRIPS

2.6 4.2 57.0 10.1 11.5

1.3 2.1 49.5 16.4 5.6

8.0 9.5 424.0 92.0 45.0

1.0 0.1 5.0 0.0 6.0

6 7 8 9 10

STORE NOBRR NBRXD NBRG1 NOSTR

1.6 4.2 3.5 2.3 2.6

0.7 2.8 2.3 1.2 2.2

5.0 14.0 12.0 7.0 13.0

1.0 1.0 1.0 1.0 1.0

11 12 13 14 15

NSRG1 ARLBU ARLXD ARLBT ARLST

1.7 22.7 18.5 4.6 7.1

1.0 30.1 19.9 4.7 5.2

6.0 352.0 99.8 36.0 34.0

1.0 1.5 1.1 1.0 1.0

16 17 18 19 20

SH1LS ALU1S ALT1S SH2LS SH3LS

0.874 39.8 7.8 0.117 0.012

0.2 41.6 5.1 0.1 0.0

1.0 384.0 34.0 0.5 0.3

0.3 1.7 1.0 0.0 0.0

21 22 23 24 25

SH1LB ALU1B ALT IB SH2LB ALU2B

0.740 27.9 5.4 0.162 8.2

0.2 31.9 4.8 0.1 6.6

1.0 352.0 36.0 0.5 48.0

0.2 1.8 1.0 0.0 0.6

26 27 28 29

ALT2B SH3LB ALU3B ALT3B

1.7 0.47 5.1 1.2

1.0 0.1 2.6 0.3

10.0 0.3 32.0 3.0

1.0 0.0 1.0 1.0

Sample size = 404.

24

PURCHASING BEHAVIOR AND PERSONAL ATTRIBUTES

Table 2-4 MEANS, S T A N D A R D DEVIATIONS, A N D R A N G E S FOR 29 P U R C H A S I N G VARIABLES: BEER, A N A L Y S I S S A M P L E Variable No. Name

Average

Standard Deviation

Max

Range Min

1 2 3 4 5

BRAND UNPTP UNITS UNIDE TRIPS

3.0 5.3 189.0 n.a. 22.8

1.9 2.3 192.3 n.a. 25.1

11.0 9.9 908.0 n.a. 178.0

1.0 0.1 9.0 n.a. 5.0

6 7 8 9 10

STORE NOB RR NBRXD NBRG1 NOSTR

2.1 6.3 n.a. 2.9 4.6

1.0 7.9 n.a. 2.5 5.2

6.0 66.0 n.a. 16.0 33.0

1.0 1.0 n.a. 1.0 1.0

11 12 13 14 15

NSRG1 ARLBU ARLXD ARLBT ARLST

2.8 80.0 n.a. 9.3 9.2

2.5 138.6 n.a. 19.4 13.2

17.0 828.0 n.a. 178.0 80.0

1.0 2.6 n.a. 1.0 1.0

16 17 18 19 20

SHI LS ALUIS ALT1S SH2LS SH3LS

0.810 105.0 10.6 0.141 0.024

0.2 157.4 13.3 0.1 0.1

1.0 908.0 80.0 0.5 0.3

0.3 4.5 1.0 0.0 0.0

21 22 23 24 25

SH1LB ALU IB ALT IB SH2LB ALU2B

0.762 93.5 9.9 0.155 19.2

0.2 139.1 14.2 0.1 24.2

1.0 828.0 80.0 0.5 306.0

0.2 3.1 1.0 0.0 2.0

26 27 28 29

ALT2B SH3LB ALU3B ALT3B

2.1 0.040 13.1 1.6

2.2 0.1 10.0 1.2

17.0 0.3 88.0 18.0

1.0 0.0 1.1 1.0

n.a. — not applicable Sample size = 247.

THE DIMENSIONS OF PURCHASING

BEHAVIOR

25

product classes. Similarly, the number of stores is larger for beer than for the others, and beer has the smallest average primary store share. These relations seem to m a k e sense in the context of known market facts about the three commodities. Coffee is subject to heavy dealing and other kinds of promotion aimed expressly at the encouragement of brand switching behavior. Beer may be purchased in a variety of retail outlets, with different outlets being frequented by different members of the family. Examination of the averages for number of runs and average run length shows that there is no systematic tendency for these figures to fluctuate with total activity. (The apparent exception to this finding is average run length in units, but we must recall that the units in which consumption of the three products is measured are different, so direct comparisons based on units are not valid.) N o r do the patterns appear to be different for coffee and tea, as opposed to beer. While evidence based on marginal statistics is weak here, it does not suggest that the ordering of the purchases for the first two products was spurious. Finally, we may consider the average number of units purchased under some kind of deal as a percentage of the overall average number of units purchased for each product class. This statistic is largest for coffee (24 percent, with tea second 17 percent). No deal purchases were recorded for the beer sample. Once again, these results are compatible with known facts about the type of promotion utilized for the three commodities. CORRELATIONS AMONG T H E VARIABLES.

The simple correlation coefficients for pairs of the twenty-nine coffee purchasing variables are presented in Table 2-5. Only cases where the correlation was greater than 0.3 have been included; the other figures are left out in order to improve the readability of the table. The correlations for tea and beer are not presented, but they follow the same general pattern. The differences among the three correlation matrices will be discussed later in this chapter. Reading down the first column of the table we find that the number of brands bought by a family is highly correlated with the number of brand runs, the average run length, and the primary, secondary, and tertiary share statistics. Lesser correlations may be observed between the number of brands and the number of stores. The relation between the number of brands and total activity, measured in terms of pounds or trips, is too small to be included in the table. Since the statistics for number of brands, number of brand runs, average run length, and brand share are correlated among themselves, there is a strong presumption that they all contribute to a general dimension for purchasing behavior, which we may call brand loyalty.

C/3

UJ -J m


/•>- »^ »^f*^

tu

o¿ 0¿ O U

30 1

C9 i« w

8rsi

13

I—I

jö TT

—

-V

O

8

O © sO OO -C «

O

85

i i . n ï ï t Î B i i M a J j j j d ^ h ^ - D h ^ r S D h ^ z z z a f í o t t o o i / i f l í o í c í í i j j i i i j j i j j i j j aiD3DHZZZZZ>. •ß T3 U

E CS X> .2 * i— CS > « 00 CS

O t/1 -CS c ®£ 2 o ¡2 H í 3 00 -o aCS ^ e uU t/3 CS t/î 'u -O 'u — = - £-

u c t/ï > CS

ï E o ë "

O J C •S o ° S ° i §• « •> c u 55 « u .2 o TJ

Q. E es o C/5 C o LZ u

NS u '3> o ^ C u u* u u o..* u "C M «I ™ M u S

DESCRIPTION OF P E R S O N A L I T Y AND SOCIO-ECONOMIC

STATUS

87

5. Home Ownership. F a m i l i e s w h o own t h e i r own h o m e , h a v e slightly higher than average income, and are slightly older than average are negatively associated with this factor. 6. Car Price Class. Families who own expensive cars and those with low income and no car are at the poles of this factorial dimension. T h e results of the factor analysis are reasonable in terms of both c o m m o n sense and generally accepted theories of social class. T h e communalities show that the f a c t o r analysis has been able to account for most of the variance of the two age variables, and a substantial amount of it for income. On the other hand, only about half the variation in education, occupation, and market size has been captured by the analysis, and the proportion is substantially less than half for the other variables. T h e r e f o r e , we would not feel safe in using the f a c t o r s c o r e s as o u r o n l y s o c i o - e c o n o m i c p r e d i c t o r s of p u r c h a s i n g behavior. Since it is likely that both the specific and c o m m o n parts of the variables are important, it was necessary to reach a c o m p r o m i s e with respect to the choice of s o c i o - e c o n o m i c variables. T h e r e f o r e , five of the factor scores and nine of the raw variables were included in the initial stepwise regression runs for searching out relations between socio-economic and personality variates and purchasing behavior (see C h a p t e r 5). T h e two age variables were left out of this set because of their high correlations with the life cycle factor, and the car price class factor was deleted because it added little to the interpretation of the raw variables.

SUMMARY OF PREDICTOR

SPECIFICATIONS

T h e results reported in this chapter led us to specify the following set of forty-five personality and fourteen socio-economic and demographic variables for use as predictors of purchasing behavior. Various subsets and t r a n s f o r m a t i o n s of these fifty-nine potential explanatory variables are tested against purchase behavior data for coffee, tea, and beer in the next chapter. W o r k on the analysis sample data yielded a final regression model which includes twenty-nine predictors from the above list. This model was applied to fresh data from the validation sample with the results reported in Chapter 6.

88

P U R C H A S I N G BEHAVIOR A N D P E R S O N A L A T T R I B U T E S T a b l e 4-7 E X P L A N A T O R Y V A R I A B L E SET U S E D IN A N A L Y S I S SAMPLE

REGRESSIONS

Personality 1. T h e fifteen s t a n d a r d i z e d E P P S scores for h u s b a n d and wife. ( S e e T a b l e 4-1 for definitions.) S t a n d a r d i z a t i o n is based on the m e a n s and s t a n d a r d d e v i a t i o n s for each sex separately.

30 v a r i a b l e s

2. T h e following principal c o m p o n e n t a n d v a r i m a x r o t a t e d f a c t o r scores, for both h u s b a n d and wife.

10 variables

a. T e n d e r n e s s - t o u g h n e s s (principal c o m p o n e n t factor). b. T e n d e r n e s s alone (first r o t a t e d f a c t o r ) . c. Orderliness-disorderliness (second r o t a t e d factor). d. T o u g h n e s s - w e a k n e s s (third r o t a t e d f a c t o r ) . e. C h a n g e alone ( f o u r t h r o t a t e d f a c t o r ) . 3. T h e following c a n o n i c a l variate scores for b o t h h u s b a n d

4 variables

and wife. a. H e t e r o s e x u a l i t y . b. S u c c o r a n c e , aggression, and c h a n g e . 4. The similarity Z - s c o r e for c o m p a r i n g h u s b a n d s ' and wives personality profiles. T o t a l for personality

I variable

45 variables

S o c i o - e c o n o m i c and d e m o g r a p h i c status ( S E S ) 5. The nine raw socio-economic variates plus the family size m e a s u r e p r o v i d e d by t h e p a n e l . ( A l l t h e v a r i a b l e s in T a b l e s 4-5 and 4-6 with the exception of h u s b a n d ' s and wife's age.)

9 variables

6. T h e following five v a r i m a x r o t a t e d f a c t o r scores.

5 variables

a. b. c. d. e.

Life cycle. Education. M a r k e t size (size of city). Income-occupation. H o m e ownership. Total for S E S

14 variables

G r a n d t o t a l for predictor variables.

59 v a r i a b l e s

y EXPERIMENTS ON T H E ANALYSIS S A M P L E Our research strategy for investigating the relations between purchasing behavior, personality, and socio-economic status was to divide the data base into analysis and validation samples so that alternative models could be tried out without the risk of contaminating our final statistical tests. M o r e detailed rationales for this approach are discussed in Chapter I. T h i s chapter reports the main experiments performed on the analysis sample. T h e outcomes of these experiments permitted us to formulate a model which could be applied to the validation sample. Final results for the validation sample are reported in Chapter 6. O u r efforts at variable definition, reported in the last three chapters, yielded a list of some forty-five personality variables, fourteen socioeconomic variables, and fourteen purchasing behavior variables. ( F o u r additional dependent variables were added to the analysis at the time of the validation sample runs.) Not all of the explanatory variables could be used at once because of problems of collinearity or compatibility of interpretations. On the other hand, a wide variety of transformations of the variables in any regression equation was available (e.g., log transforms, inclusions of interaction terms, etc.). Y e t there was little available in the way of either theory or previous empirical evidence that could help in making the necessary choices. Therefore we were thrust back upon our own data: the whole strategy of using an analysis sam-

89

90

PURCHASING

BEHAVIOR AND PERSONAL

ATTRIBUTES

pie was designed to allow i n d i s c r i m i n a t e searching for the variable c o m b i n a t i o n s a n d t r a n s f o r m a t i o n s t h a t provided t h e best fit a n d i n t e r p r e t a t i o n s without r u n n i n g the risk of c o n t a m i n a t i n g o u r final results. O u r work on the analysis s a m p l e is divided into three stages, each of which is described briefly below a n d t r e a t e d in a s e p a r a t e section of this c h a p t e r . (1) P r e l i m i n a r y searching for v a r i a b l e c o m b i n a t i o n s a n d transf o r m a t i o n s t h a t provided the best fit to the d a t a on regular coffee. T h i s work was a c c o m p l i s h e d by using a stepwise m u l t i p l e regression p r o g r a m for searching over the m a n y c o m b i n a t i o n s of v a r i a b l e s t h a t were j u d g e d t o be potentially i m p o r t a n t . (2) T e s t i n g of the stepwise regression results on coffee with s t a n d a r d fixed v a r i a b l e regression runs, extending the experim e n t s to include t h e o t h e r two p r o d u c t categories, a n d m a k i n g a p p r o p r i ate m o d i f i c a t i o n s in the models. (3) Investigation of b r a n d and store specific e f f e c t s to d e t e r m i n e w h e t h e r the a n a l y s i s s h o u l d be e n l a r g e d to include these d i m e n s i o n s of p u r c h a s i n g b e h a v i o r . T h e first p h a s e of e x p e r i m e n t a t i o n w a s c o n f i n e d to d a t a on regular coffee p u r c h a s e s because t h a t was t h e d a t a base for which the largest s a m p l e size was available a n d the highest p u r c h a s e rate exhibited. This m e a n t t h a t t h e r e were a large n u m b e r of families in the analysis s a m p l e for coffee, a n d t h a t the p u r c h a s i n g statistics for each family were based on a larger n u m b e r of s h o p p i n g r e c o r d s t h a n was t r u e for the other p r o d u c t classes. F u r t h e r m o r e , t h e a u t h o r s h a d p r e v i o u s experience in w o r k i n g with panel d a t a for regular coffee ( F r a n k , 1962, a n d F r a n k a n d Massy, 1963), which could be expected to be useful in t h e i n t e r p r e t a t i o n of the present findings. T h e results of these e x p e r i m e n t s were applied to the tea a n d beer d a t a in p h a s e t w o of o u r work on t h e a n a l y s i s s a m p l e . T h e third s t a g e of analysis was i n s t i t u t e d in o r d e r to check the possibility that p e r s o n a l i t y a n d s o c i o - e c o n o m i c s t a t u s a r e related to the p u r c h a s e of specific b r a n d s or the p a t r o n a g e of certain classes of stores. T h e m a i n focus of this b o o k is on the prediction of p u r c h a s i n g variables o b t a i n e d by a g g r e g a t i n g the d a t a over all t h e b r a n d - s t o r e c o m b i n a t i o n s included in the p a n e l . O n t h e o t h e r h a n d , it is p o s s i b l e t h a t c e r t a i n p e r s o n a l i t y t y p e s might be a t t r a c t e d to p a r t i c u l a r b r a n d s or stores, p e r h a p s because of i m a g e c o n f i g u r a t i o n s , even in the a b s e n c e of a n y general personality-loyalty relation. O u r work on this question is r e p o r t e d in its entirety in the third section of this chapter. Since d a t a c o n s t r a i n t s limited the n u m b e r of b r a n d s and s t o r e s t h a t could be tested a n d the results of the available runs a p p e a r e d to be singularly u n p r o m i s i n g , this p a r t of our analysis was not extended to the validation s a m p l e a n d h e n c e is not t r e a t e d in C h a p t e r 6.

E X P F R I M E N T S ON T H E A N A L Y S I S S A M P L E

91

STEPWISE MULTIPLE REGRESSION: R E G U L A R C O F F E E DATA Three experiments were tried using stepwise multiple regression based on regular coffee purchasing behavior. Each experiment was aimed at sharpening our formulation of the personality and socio-economic variables. T E S T S ON SUBSETS O F EXPLANATORY VARIABLES.

The first experiment consisted of running nine stepwise multiple regressions, each of which was based on the initial set of forty-five personality and fourteen socio-economic variables. Each of the nine regressions was aimed at predicting a different dimension of regular coffee purchasing behavior, namely: 1. Overall loyalty principal component factor score. 2. Activity varimax factor score. 3. Brand loyalty varimax factor score. 4. Store loyalty varimax factor score. 5. Consistency varimax factor score. 6. Proportion spent on first loyal brand. 7. Proportion spent on first loyal store. 8. Pounds per trip. 9. Proportion of pounds purchased on a deal. The coffee data base provided a substantially larger sample size, 629 households, than did that for either tea (376) or beer (237). The first five dependent variables are the summary measures generated by the factor analyses described in Chapter 2. The next two are the raw purchase variables which received a major weight in determining a household's brand and store varimax loyalty scores, respectively. They were included as an internal check on the results associated with the varimax scores. The last two dependent variables are dimensions of purchasing behavior which did not fall neatly into one of the factors for which scores were generated. In this and the two experiments which followed, we make use of a stepwise multiple regression program developed by the Health Sciences C o m p u t i n g F a c i l i t y a t t h e U n i v e r s i t y of C a l i f o r n i a , Los A n g e l e s (Dixon, 1964). The program is basically a least squares searching mechanism. It starts by computing the F ratio (the square of the more familiar t-ratio) that would be associated with each variable if it were the only one included in the equation aimed at predicting a given dependent variable. It chooses the variable for inclusion in the equation that has the highest F ratio, providing that the ratio is larger than the one set by the user as the minimum F required for inclusion. After it has made this first choice it computes the F ratio for each of the

92

P U R C H A S I N G BEHAVIOR A N D P E R S O N A L A T T R I B U T E S

remaining fifty-eight variables a s s u m i n g t h a t the new variable will be added to the e q u a t i o n which already c o n t a i n s the first variable. O n c e m o r e the v a r i a b l e chosen for addition is the one associated with t h e highest F, provided that it is over the m i n i m u m specified by the user. Each time a new variable is added the F ratios for each of the variables in the e q u a t i o n a r e tested to see if they have fallen under t h e F level required for deletion. If they have, they a r e deleted. T h e p r o g r a m keeps on cycling — adding variables a n d c h e c k i n g for deletion — until it reaches a point where no variable passes the F r a t i o for inclusion and those which have r e m a i n e d in the e q u a t i o n a r e in excess of the F ratio specified for deletion. T h e p r o g r a m can be t h o u g h t of as searching the least squares g r a d i e n t looking f o r the best set of predictors, subject to the c o n s t r a i n t s on statistical significance set by the user. In our analyses the F-ratio for inclusion was .250 and that for deletion was .120. These are relatively lenient tests. We preferred to t a k e a greater risk of including s o m e t h i n g t h a t was irrelevant at this stage, as opposed to excluding a variable t h a t m i g h t later t u r n out to be of importance. The principal objective of this first e x p e r i m e n t was to provide a basis for choosing which m e a s u r e of personality ( s t a n d a r d Z , varimaxprincipal c o m p o n e n t or canonical scores) should be included in the final

model. The burden of proof for inclusion falls primarily upon the Z. varimax-principal component, and canonical scores as they are more difficult to interpret and I or involve more complicated assumptions as to the way in which personality characteristics influence buyer behavior than do the standard personality scores. F o r e x a m p l e , the s t a n d a r d scores d o not t a k e into a c c o u n t t h e way in which the h u s b a n d ' s personality characteristics a r e related to those of the wife (i.e., the personality structure of the household). N o r d o they c h a r a c t e r i z e the extent to which the scores for either the h u s b a n d or the wife a r e intercorrelated. The Z a n d the canonical m e a s u r e s represent a t t e m p t s to c h a r a c t e r i z e household s t r u c t u r e , while the v a r i m a x - p r i n c i p a l c o m p o n e n t scores take into a c c o u n t t h e i n t e r c o r r e l a t i o n s of the personality scores for the husband a n d wife, respectively. Given t h a t we have no t h e o r y f o r indicating how household personality s t r u c t u r e or the i n t e r c o r r e l a t i o n s should influence any particular dimension of buying behavior, the best we can d o is observe t h e extent to which t h e searching m e c h a n i s m built into the stepwise p r o c e d u r e picks out t h e alternative personality c h a r a c t e r i z a t i o n s . O u r conclusions a r e based p r i m a r i l y on two s u m m a r y measures describing t h e end result of the stepwise regression process: (1) the proportion of times t h a t each of the f o u r types of personality c h a r a c t e r izations a p p e a r in the stepwise e q u a t i o n s ; and (2) the p r o p o r t i o n of Fratios for e a c h type that a r e significant at t h e 95 percent level of confidence. T h e results are presented in T a b l e 5-1. T h e m e a s u r e " n u m b e r of

93

E X P E R I M E N T S ON T H E A N A L Y S I S S A M P L E

T a b l e 5-1 THE PROPORTION OF CASES INCLUDED IN T H E R E G R E S S I O N , A N D C A S E S WITH SIGNIFICANT F-RATIOS, BY T Y P E O F P E R S O N A L I T Y CHARACTERIZATION FOR THE FIRST STEPWISE EXPERIMENT

Proportion of cases:

Standard scores Varimax-principal c o m p o n e n t scores Canonical scores Z score

N u m b e r of possibilities 270 90 36 9

Included 47% 21%

With significant F-ratio 10% 3%

31% 33%

3% 0%

possibilities" is obtained by multiplying the n u m b e r of individual variables included in a given variable classification by the total n u m b e r of stepwise e q u a t i o n s (for e x a m p l e , there are thirty s t a n d a r d personality s c o r e s which when distributed over the nine e q u a t i o n s yields 270 sepa r a t e possibilities for inclusion). T h e p r o p o r t i o n of inclusions is the ratio of ihe actual n u m b e r of inclusions to the n u m b e r of possibilities. Finally, t h e p r o p o r t i o n of significant F - r a t i o s is o b t a i n e d by dividing t h e n u m b e r of cases where variables in the set a r e in t h e e q u a t i o n with F ratios g r e a t e r than 3.84 by the total n u m b e r of possibilities. B o t h t h e p r o p o r t i o n s of i n c l u s i o n s a n d s i g n i f i c a n t F r a t i o s a r e g r e a t e r for the s t a n d a r d personality scores t h a n for any of the other three c h a r a c t e r i z a t i o n s of personality, a n d by a wide m a r g i n . Since the f a c t o r a n d canonical scores could not be included in a regression that contained the raw personality scores because of collinearity p r o b l e m s , we tentatively decided to d r o p the f o r m e r . T h e Z score was never significant in any of the stepwise runs a n d p r o d u c e d coefficients with signs and m a g n i t u d e s t h a t could not be i n t e r p r e t e d , so it was also considered f o r deletion pending the results of the o t h e r stepwise e x p e r i m e n t s .

T E S T S FOR H L S B A N D - W I F E PERSONALITY INTERACTIONS.

W e suspected t h a t the effect of any o n e personality d i m e n s i o n , f o r e x a m p l e the d e g r e e of need for c h a n g e on the p a r t of t h e wife, might depend on the similarity of the h u s b a n d ' s and wife's personalities as m e a s u r e d by t h e Z - s c o r e . For e x a m p l e , a m o n g h o u s e h o l d s where t h e h u s b a n d and wife have relatively similar personalities those with a wife with a high c h a n g e score might exhibit a lower degree of b r a n d loyalty

94

P U R C H A S I N G BEHAVIOR A N D PERSONAL A T T R I B U T E S

than those households comprised of wives with low change scores, whereas among households consisting of husbands and wives with dissimilar personalities, differences in the wife's need for change might have little effect on the degree of loyalty exhibited. In other words, we hypothesized that there might be an interaction between the extent to which the personalities of the husband and wife coincided and the effect of any need on the part of one member of the pair on buying behavior. We tested this hypothesis by running three more stepwise regressions based on regular coffee purchases to predict the varimax factor scores for brand loyalty, store loyalty, and consistency. The explanatory variable set included the fourteen socio-economic scores together with all of the canonical and varimax-principal component personality scores. Based on the results of the first stepwise run twelve of the standard personality scores and the Z score were deleted. The primary reason for deletion was that their F-ratios were lower than the remaining eighteen personality characteristics. Eighteen new interaction terms were created by taking the cross products of the Z-score with the remaining eighteen personality scores. The eighteen personality dimensions were: Wije Husband Deference Exhibition Autonomy Affiliation Change Endurance Aggression

Deference Order Exhibition Autonomy Affiliation Intraception Succorance Nurturance Change Endurance Heterosexuality

The eighteen interaction terms together with the original eighteen personality scores (36 variables in all) were included as explanatory variables in this experiment. In addition, three of the canonical variables (husband's heterosexuality and succorance and wife's succorance) were used, together with the Z-score, to create multiplicative interaction terms. The results are reported in Table 5-2. A s in the case of the first experiment the standard scores outperformed the other personality characterizations. Apparently the extent to which the personalities of the husband and wife tend to coincide has little impact on household buying behavior, at least for regular coffee, and probably for similar types of frequently purchased food products as well. The results of the first two experiments are consistent in that they both favor the use of the standard personality scores.

95

E X P E R I M E N T S ON T H E A N A L Y S I S S A M P L E

T a b l e 5-2 THE PROPORTION OF CASES INCLUDED, AND CASES WITH A SIGNIFICANT F-RATIO, BY T Y P E OF PERSONALITY CHARACTERIZATION THE SECOND STEPWISE EXPERIMENT

N u m b e r of possibilities S t a n d a r d scores Varimax-principal c o m p o n e n t scores C a n o n i c a l scores Z-canonical interactions Z - s t a n d a r d score interactions

54 30

FOR

Proportion of cases: With significant Included F-ratios

69% 0%

12 9

33%

54

33%

17%

4%

T E S T S FOR NONLINEARITIES.

T h e third and last stepwise experiment served to test for suspected nonlinearities in the relations of the socio-economic and personality variables with purchasing behavior. Given our hypotheses as to nonlinear structure, an appropriate transformation was made of the scores for each variable involved. T h e transformed variables were then substituted for the original variables. Three stepwise regressions were then run for regular coffee using brand loyalty, store loyalty, and the consistency varimax scores as the dependent variables. Our hypotheses, the transformations used, and the resulting impact on structure will be described in the following paragraphs. 1. T h e standard personality scores. Earlier work done by the Advertising Research Foundation (1964), using the same panel but based on data for toilet paper purchasing behavior, led them to the hypothesis that a given personality characteristic influences buying behavior only in extreme cases — that is, only where the need expressed is unusually high or low in intensity. This suggests that the relationship between a given personality trait and buying behavior might take the form of the inverted S-shaped curve pictured in Figure 5-1. If this relationship approximates reality, then the following transformation will tend to linearize it:

where X , , is the ith household's score for the j t h personality variable. X j is the mean for the jth personality variable across all households.

96

PURCHASING BEHAVIOR A N D PERSONAL

ATTRIBUTES

PERSONALITY VARIATE

Figure 5-1. S-Transformation of Personality Variate and X*,, is the transformed variable. The burden of proof falls on the transformed personality scores used in this experiment for the same reason that it fell on the Z, varimax-principle component, and canonical scores that were used in the first experiment: namely, because the transformed scores involve more complicated assumptions and are harder to interpret. If the transformation is effective it should increase (1) the proportion of personality scores included in the final equation and (2) the proportion of F-ratios that are significant. The proportion of standard score variates included in the first (untransformed) experiment was .42 while the result for this experiment is .50. In contrast, the proportion of significant F-ratios was .16 in the first experiment compared to only .07 for the transformed data. Thus the pattern of evidence is inconsistent: the first result supports the use of the transformations while the opposite is true for the second criterion. Given the arguments about burden of proof discussed above, we interpreted the pattern of results as favoring the simpler formulation; namely the use of untransformed personality variates as in the first experiment. 2. The Z, Varimax-Principal Component, and Canonical Scores. Given that the standard untransformed personality scores are to be preferred over the transformed scores the decision problem that faced us changed from a comparison of the first and third experiments to a comparison of the results for the standard untransformed personality scores in the first experiment with the transformed versions of the Z, varimax-principal component and canonical scores in this experiment. Table 5-3 presents the relevant data.

EXPERIMENTS ON THE ANALYSIS

97

SAMPLE

Table 5-3 THE PROPORTION OF CASES INCLUDED. AND CASES WITH SIGNIFICANT F-RATIOS, FOR THE FIRST E X P E R I M E N T STANDARD SCORES AND THE REMAINING PERSONALITY C H A R A C T E R I Z A T I O N S IN T H E T H I R D E X P E R I M E N T

P r o p o r t i o n of c a s e s : N u m b e r of First experiment: standard scores Varimax-principal component scores Canonical scores Third experiment: Z score

With signmcant

possibilities

Included

F-ratios

90

41%

30 12

53% 93%

3%

100%

0%

T h e proportion of cases included for all types of third experiment personality characterizations are somewhat higher than those for the first experiment. However, the opposite is true of the probability of generating a significant F-ratio. A s in the case of the second experiment, this pattern of inconsistency led us to prefer the untransformed standard personality scores. One other piece of evidence supported our conculsion; namely a comparison of the R2' s and F-ratios associated with the two experiments ( T a b l e 5-4). T h e F-ratios for the third experiment are consistently less significant than those for the first experiment. T h e apparent increase in the R1 s for the three equations is primarily the result of an upward bias in predictive efficacy caused by increasing the number of variables included in the equation while holding the sample size constant. Pending the results of trials on tea and beer, these results convinced us that standard, untransformed personality scores should be used as the basis for predicting the purchasing behavior variables in the validation sample. 3. T h e S o c i o - e c o n o m i c S t a t u s Variables ( S E S ) . T a b l e 5-5 presents the F-ratios for each of the fourteen S E S variables for the equations c o m m o n to the first and third experiments. T h e third experiment contained transformations of nine S E S variables whose titles are followed by an asterisk. In seven of the nine cases (wife's education, income, occupation, home ownership, life cycle-varimax, education-varimax, and incomeoccupation varimax) we hypothesized that the underlying relationship between the variable and the store loyalty, brand loyalty and consist-

98

P U R C H A S I N G BEHAVIOR A N D P E R S O N A L A T T R I B U T E S

T a b l e 5-4 C O E F F I C I E N T S OF DETERMINATION, F-RATIO, A N D N U M B E R OR VARIABLES FOR EACH STEPWISE REGRESSION EXPERIMENT (Sample Size: 629 Households) Dependent First experiment R2 F N o . of v a r i a b l e s Second

(1)

(2)

(3)

variable (4)

(5)

(6)

(7)

(8)

(9)

.12

.07

.10

.12

.09

.10

.10

.09

.09

3.60

2.33

3.78

3.70

2.74

3.13

2.46

2.13

2.44

24

23

20

24

24

23

28

28

25

experiment R2 F

.12

.14

.09

3.08

2.94

2.81

27

35

22

N o . of v a r i a b l e s Third experiment R2 F

.14

.13

. 11

2.81

3.05

2.29

34

31

32

N o . of v a r i a b l e s (1) (2) (3) (4) (5) (6) (7)

Over-all loyalty principle c o m p o n e n t score. Activity varimax factor score. Brand loyalty varimax factor score. S t o r e loyalty varimax factor score. Consistency varimax factor score. % first loyal b r a n d . % first loyal store.

(8) (9)

P o u n d s per trip. % purchased on deal.

ency purchasing dimensions would have the general functional f o r m pictured in Figure 5-2. ' If our hypotheses were correct then the following t r a n s f o r m a t i o n would result in a linearization of the relationship between Y and X* :

where X , , is the ith household's score for the j t h variable and d corrects for the origin of the process by taking on the value of the m i n i m u m X v in the sample. In the case of family size we hypothesized t h a t the underlying relationship would a p p r o x i m a t e that pictured in F i g u r e 5-3. T h e car ownership variable was coded in a particular fashion which led us to t r a n s f o r m it by taking the raw variable to the .1 power. N o n o w n e r s of cars were given a code of zero, o w n e r s of low-price cars a code of one, etc. By taking the original value t o the power of . I we

o

0\ —

O

o

o U

E

o

s

C/3

SO r^i

o\

2

O TJ-

Tí

CS

S

—

W-J

Tf'

VO

O

UJ

0/5

è

=>

u

< H

9

U

H

u c

00 Tí m

tí-

O

t t

OO m

t 00

vO O

CTv " t

c o U c u

i n u

2 «

E 'E V

o

>. ü

X «

o 2 C/J

« >> o

Ov VO

3> OÑ

VO

H Crt

C

UJ — o UJ U

0¿ UN

I

CJ O

I H

on

o

z

13 C

«

CU o _1 e o «

g O

73 =

5

E D d

2

Oí

c o • c o

"> U

-r I

* ta

jfc

^

Cl

•- o • c ¡y. > . _û . > ? V C »s S

s°

ZT .2 ¡2 > Jü I oO " uo

O U

M I-

E er§

CL 3 o O

o

x

h

u

O

n ' 5

« 8 «•-» Mf « .e i

^ g w tf W O l- • ï S «es u

o.

O u

S

«

• « V

3

u

". " S

S >

§ .—

"m

1

3

- o "

1

«

U

^ £ U

.a on

g S

»

s

°

U

3

S>

JS h: 5 S

g c o

E o u £

« u. « u

.2

i

a 3 u I

x i 3 « 3

(S >

PURCHASING BEHAVIOR

SOCIO-ECONOMIC

VARIATE

Figure 5-2. Quadratic Transformation of Socio-Economic Variate.

PURCHASING BEHAVIOR

FAMILY

SIZE

Figure 5-3. Square Root Transformation of Family Size Variable.

E X P E R I M E N T S ON T H E A N A L Y S I S S A M P L E

101

converted the codes into a dummy variable form (zero for non-ownership and one for ownership). Our hypothesis was that there would be a difference in buying behavior (e.g., store loyalty) between households that owned and those that did not own a car, whereas the price level of the car would be unrelated to household buying behavior. On the basis of the contrast reported in Table 5-5 nine of the fourteen variables were tentatively eliminated from further consideration. Wife's education, TV ownership, and the market size varimax, as well as both the raw and varimax score for home ownership showed little promise of contributing to the prediction of purchasing behavior. The raw scores for both income and occupation were eliminated in preference to the untransformed version of the income-occupation varimax score. The raw score family size variable was dropped in favor of the transformed life cycle varimax variable. The car status varimax variable was dropped in favor of the transformed version of car ownership. S T A N D A R D M U L T I P L E R E G R E S S I O N : ALL P R O D U C T S The purposes of phase two of our experimentation were to try out the independent variable set tentatively selected in phase one in a fixed variable regression model and extend this part of the analysis to include the analysis sample data for tea and beer. In addition, we tried out a number of minor transformations of the dependent variables during this phase of the study. In order to achieve these objectives we ran some forty 35-variable regressions (not stepwise) covering various dependent variable-product combinations. The results obtained from these regressions are summarized in the following paragraphs. COLLINEARITY PROBLEMS.

Our first and most striking finding was that there is very high multicollinearity among the personality traits for husbands, and also for wives. (Here we are referring to within group collinearity; the relation between husbands' and wives' personality scores was treated in connection as part of the discussion of canonical variates in Chapter 4.) Subsequent investigation revealed what should have been obvious at the outset: The scoring procedure used in the E P P S is such that any of the personality variates can be perfectly predicted from the other fourteen. That is, the Edwards Test produces fifteen scales but has rank of only fourteen. One of the scales is redundant. The fact that the two groups are ndt perfectly collinear in our data must be attributed to errors in applying the test scoring procedures to the data from some of the respondents. We have treated the general problem of handling collinearity in the Edwards Test elsewhere (see Massy, Lodahl, and Frank, 1966). For present purposes it was judged best to delete enough of the personality

PURCHASING BEHAVIOR AND PERSONAL ATTRIBUTES

102

s c a l e s to reduce the collinearity to a c c e p t a b l e p r o p o r t i o n s . T h e c h o i c e as t o which variates were to be e l i m i n a t e d involved a t r a d e o f f between the c o n t r i b u t i o n o f each variable to the total collinearity o f the set (see the reference given above) and its relative predictive power with respect to purchasing behavior. In addition, we felt t h a t it was desirable to m a i n t a i n s y m m e t r y between the variables included for the two sexes in o r d e r to avoid difficulties of interpretation. T h e v a r i a b l e s finally chosen for

elimination

were

achievement,

dominance,

and

heterosexuality.

H e t e r o s e x u a l i t y c o n t r i b u t e d the m o s t to c o l l i n e a r i t y and was not strongly related t o the dependent variables. A c h i e v e m e n t and

domi-

n a n c e were not related to purchasing behavior either, and the validity o f the f o r m e r s c a l e is questionable as well (see A p p e n d i x A ) . T h e three scales were e l i m i n a t e d for both husbands and wives, thus reducing the n u m b e r of e x p l a n a t o r y variables by six. E l i m i n a t i o n of these variables dropped the m a x i m u m coefficient o f d e t e r m i n a t i o n for the collinearity of one e x p l a n a t o r y variable with all the others t o a fairly respectable 0.45.

Additional and

analysis led us to d r o p the e d u c a t i o n - v a r i m a x

replace the

life c y c l e - v a r i m a x

variable

variable with the raw scores

for

wife's a g e and f a m i l y size. T h e s e two scores formed the m a j o r input for the life-cycle f a c t o r score, and they seemed to have s e p a r a b l e relations with purchasing behavior. T h a t is, they tend to have differential effects on such dependent variables and total activity and b r a n d loyalty. F i n a l ly, the i n c o m e - o c c u p a t i o n v a r i m a x f a c t o r score was t r a n s f o r m e d in the s a m e m a n n e r a s had previously been used for life cycle. T h e s e c h a n g e s were suggested when the model was applied to the t e a and beer d a t a . They

left us with a final model c o n t a i n i n g t w e n t y - n i n e

explanatory

variables: twelve e a c h for husbands' and wives' p e r s o n a l i t y , and five for s o c i o - e c o n o m i c status. DEPENDENT VARIABLE TRANSFORMATIONS. T h e m a r g i n a l d i s t r i b u t i o n of s e v e r a l of the d e p e n d e n t v a r i a b l e s ( n u m b e r of b r a n d s , pounds per trip, total pounds, n u m b e r of trips, n u m b e r of s t o r e s ) resembles a Poisson distribution. O n the other hand, the regression

model assumes in part that the dependent variable

is

n o r m a l l y distributed. T h i s assumption is particularly i m p o r t a n t for the validity

of

hypothesis

tests. T h e

following

transformation

was

per-

f o r m e d in order t o bring the marginal distributions c l o s e r t o n o r m a l i t y : Y* ij

= J y1 * ij

1

~

'

where Y , , is the value of the j t h dependent variable for the ith household.

103

E X P E R I M E N T S ON T H E A N A L Y S I S S A M P L E

Three of the dependent variables take the form of proportions. They are pounds on a deal, percentage of purchases made for the brand purchased most often, and percentage of purchases made in the store most often shopped. The regression model assumes that the conditional variance of the dependent variable given the independent variable is c o n s t a n t ( h o m o s c e d a s t i c ) . S u p p o s e a positive r e l a t i o n s h i p exists between some independent variable, say X, and one of the aforementioned p r o p o r t i o n s . T h e n some X ' s a r e a s s o c i a t e d with a higher expected value of the proportion than are others. The expected variance of a proportion is a function of its magnitude, and therefore, the assumption of homoscedasticity will be violated. The following arcsin transformation was performed to stabilize the variance:

LOYALTY T O S P E C I F I C BRANDS AND S T O R E S : SOME EXPERIMENTS Because of the importance associated with the study of specific loyalty to brands and stores we decided to conduct a more intensive set of experiments with respect to the prediction and structural analysis of these dimensions. Five experiments were performed. Three of them were concerned with loyalty to specific brands, while two concentrated on the prediction of store choice. All of them were based on regular coffee purchasing behavior. A brief description of the nature and outcome of each one is presented in the following paragraphs. BRAND LOYALTY TO PRIVATE VERSUS NATIONAL C O F F E E BRANDS.

A two-way multiple discriminant analysis was run to determine the extent to which one can discriminate between households whose favorite brand was a chain store's private brand and those whose favorite was a non-private brand (i.e., a manufacturer's or a packer's brand). A household's favorite brand was defined as the brand that it purchased more often than any other. Only those households whose favorite store was one which carried private brands of regular coffee were included in the analysis. The following twenty-two personality dimensions were included in the equation: Husband Exhibition

Wije Achievement

104

PURCHASING BEHAVIOR AND PERSONAL ATTRIBUTES HUSBAND

WIFE

Autonomy Affiliation Succorance Change Endurance Aggression

Deference Order Exhibition Autonomy Affiliation Interception Succorance Dominance Abasement Nurturance Change Endurance Z-score

T h e following s o c i o - e c o n o m i c c h a r a c t e r i s t i c s were also included: 1. M a r k e t Size: not t r a n s f o r m e d . 2. C a r O w n e r s h i p : t r a n s f o r m e d by t a k i n g t h e raw score to the .1 power. 3. Life Cycle V a r i m a x : t r a n s f o r m e d by taking the s q u a r e of the difference between the original variable a n d the m i n i m u m score a m o n g all households. 4. E d u c a t i o n V a r i m a x : t r a n s f o r m e d in the s a m e f a s h i o n as # 3 . 5. I n c o m e - O c c u p a t i o n V a r i m a x : t r a n s f o r m e d in the s a m e fashion as » 3. T h e discriminant analysis provides a multivariate test of the hypothesis that the m e a n s of all the variables a r e the s a m e for the two groups. T h e F - r a t i o for this test (with 27 a n d 194 degrees of f r e e d o m ) was 1.0, whereas that for the .95 level of confidence is a p p r o x i m a t e l y 1.47. T h u s we c a n n o t c o m e close t o rejecting the hypothesis that the two g r o u p s of families could have been d r a w n f r o m p o p u l a t i o n s with the s a m e underlying personality a n d socio-economic s t r u c t u r e . BRAND LOYALTY T O MAXWELL H O U S E COFFEE.

A n e x p e r i m e n t using multiple regression analysis was run in an a t t e m p t t o predict the b r a n d loyalty v a r i m a x stores for those households t h a t p u r c h a s e d Maxwell H o u s e m o r e often t h a n any other b r a n d . If there a r e d i f f e r e n c e s in s o c i o - e c o n o m i c a n d personality s t r u c t u r e a m o n g b r a n d s o n e would expect t h a t a brand-specific analysis such as this would be m o r e likely to p r o d u c e m e a n i n g f u l results t h a n would an analysis using d a t a on all b r a n d s c o m b i n e d . This b r a n d a n d the one for

E X P E R I M E N T S ON T H E A N A L Y S I S S A M P L E

105

which results are reported in the next section were chosen because they were the only ones for which data sufficient for separate analysis were available. The variables included in this analysis were the same as those used in the discriminant analysis of private and national brand users except that the car ownership variable was omitted. The F-ratio for the significance of the whole regression is 1.05 (26 and 62 degrees of freedom) while the critical value at the 95 percent level is 1.70. The observed differences could easily be due to chance, given the available sample size. We are led to conclude that the use of loyalty data based on Maxwell House alone does not significantly improve the performance of the regressions. (For a comparison with the aggregative results, see Chapter 6.)

BRAND LOYALTY TO EIGHT O'CLOCK COFFEE.

The logic and variable specifications for this analysis were identical to that for Maxwell House. The F-ratio was 1.46, while F(26,17) is 2.19 at the .95 level. Once more there appears to be virtually no variation in brand loyalty associated with the socio-economic and personality characteristics.

STORE LOYALTY TO SAFEWAY.

Using the same set of variables upon which the Maxwell House and Eight O'Clock experiments were based, an attempt was made to predict the store loyalty varimax score for those families who shopped in Safeway more often than in any other store. The F-ratio was 2.76 while that for F ( 2 6 , l l ) is 2.57 at the .95 level. The results are encouraging, but the specific personality and socio-economic variables that were most important in this equation are similar to those that were important in the prediction of the store loyalty varimax for all stores combined. In other words, there appears to be nothing that is structurally specific to Safeway.

STORE LOYALTY TO A 4 P.

The same experiment was performed for A & P with disappointing results. The F-ratio was .74 while that for F(26,76) is approximately 1.70 at the .95 level. There apparently is no payoff in segregating the data on A & P shoppers. The two store experiments provide little motivation for breaking out particular chain stores for individual analysis in the validation sample.

106

PURCHASING

BEHAVIOR AND PERSONAL

SUMMARY OF MODEL

ATTRIBUTES

SPECIFICATIONS

T h e first t w o phases of our e x p e r i m e n t a t i o n with the analysis sample led to the specification of variables a n d t r a n s f o r m a t i o n s t h a t are used in the ensuing work on the validation sample. The final model consisted of the following twenty-nine variables. H u s b a n d s ' personality scores (not t r a n s f o r m e d ) Deference Order Exhibition Autonomy Affiliation Intraception Succorance Abasement Nurturance Change Endurance Aggression

Wives' personality scores (not t r a n s f o r m e d ) Deference Order Exhibition Autonomy Affiliation Intraception Succorance Abasement Nurturance Change Endurance Aggression

W i f e ' s age: not t r a n s f o r m e d . F a m i l y size: not t r a n s f o r m e d . C a r o w n e r s h i p : t r a n s f o r m e d by t a k i n g t h e r a w s c o r e t o t h e .1 power. M a r k e t size: not t r a n s f o r m e d . I n c o m e - o c c u p a t i o n v a r i m a x f a c t o r score: t r a n s f o r m e d by t a k i n g the s q u a r e of the d i f f e r e n c e between the original variable a n d t h e m i n i m u m score a m o n g all the h o u s e h o l d s in the sample. T h e p a t t e r n of results g e n e r a t e d f r o m the experiments cased on specific b r a n d s and stores led us to d r o p this line of investigation. T h e r e f o r e . the results for t h e validation s a m p l e reported in the following c h a p t e r involve b r a n d a n d store loyalty without respect to the particular b r a n d s and stores involved. NOTES A positive a s s o c i a t i o n is p r e s e n t e d for i l l u s t r a t i v e p u r p o s e s T h e validity of this a n d t h e o t h e r t r a n s f o r m a t i o n s which f o l l o w is n o t a f f e c t e d by w h e t h e r the a c t u a l a s s o c i a t i o n is p o s i t i v e o r negative.

VI RELATIONSHIPS BETWEEN PERSONALITY A N D P U R C H A S I N G BEHAVIOR In this chapter we will present our findings on the relationships b e t w e e n p e r s o n a l i t y a n d b u y i n g b e h a v i o r . In a c c o r d a n c e with o u r research strategy, the bulk of our results and interpretations will be based on data from the validation sample, since the data in this sample were in no way used in our efforts to develop a reliable set of personality and socio-economic predictors and a parsimonious set of measurements of purchasing behavior. Before proceeding with the interpretation of these results, we will review our findings from earlier chapters on the reliability of the two sets of measures and on the concept of loyalty proneness. We will then examine the utility of personality measures in the understanding of purchasing behavior by assessing the amount by which the accuracy of our prediction of purchasing behavior increases when personality variables are added to socio-economic indicators as predictors. Although we will conclude that the percentage increase in predictive power which is added by personality variables is relatively small, many of the relationships between personality variables and buying behavior appear to be r e a s o n a b l y s t a b l e . S i n c e t h e a m o u n t of v a r i a n c e in p u r c h a s i n g behavior explained by socio-economic data alone is also very small, the more reliable personality data are interpreted and some tentative explanations for the relationships are offered.

107

108

PURCHASING BEHAVIOR A N D PERSONAL

ATTRIBUTES

REVIEW OF FINDINGS ON RELIABILITY F r o m T a b l e 2-10 we have seen t h a t the reliability of t h e d i m e n s i o n s of p u r c h a s i n g b e h a v i o r r a n g e s f r o m .46 t o .93 with a m e d i a n of .78. T h e s e r e s u l t s were o b t a i n e d in a split-half reliability t e s t , w h i c h m e a n s that, on the average, purchasing behavior on these dimensions meas u r e d in o n e s i x - m o n t h p e r i o d p r e d i c t s a p p r o x i m a t e l y 6 0 p e r c e n t of t h e v a r i a n c e in t h e p u r c h a s i n g b e h a v i o r in t h e f o l l o w i n g s i x - m o n t h p e r i o d . T h u s w e c a n c o n c l u d e t h a t a l t h o u g h p u r c h a s i n g b e h a v i o r is n o t e x t r e m e l y s t a b l e o v e r t i m e , it is at least r e l i a b l e e n o u g h to give rise t o h o p e s t h a t it c a n be p r e d i c t e d . It h a s been held by s o m e t h a t c o n s u m e r panel d a t a a r e n o t o r i o u s l y u n r e l i a b l e . It m i g h t be u s e f u l h e r e t o p o i n t o u t t h a t d a t a c a n be r e l i a b l e o v e r t i m e w i t h o u t necessarily b e i n g valid i n d i c a t o r s of t h e b e h a v i o r t h a t is b e i n g r e p o r t e d . F o r i n s t a n c e , if m e m b e r s of t h e p a n e l f o u n d t h e t a s k of r e c o r d i n g a n d c o m m u n i c a t i n g t h e i r p u r c h a s e s a n o n e r o u s o n e , t h e y m i g h t t e n d a t t h e e n d of t h e m o n t h t o j o t d o w n a n e s t i m a t e of w h a t t h e y t h i n k t h e y p u r c h a s e d : t h e s e e s t i m a t e s c o u l d be relatively s t a b l e f r o m m o n t h to m o n t h ( c o m p a r e d to the actual purchasing behavior), since t h e y would p r o b a b l y be s u b j e c t t o t h e s a m e k i n d s of m e m o r y distortion from m o n t h to month. A s noted in t h e a p p e n d i x o n t h e E P P S , t h e test h a s s a t i s f a c t o r y reliability c o m p a r e d to o t h e r m e a s u r e s of p e r s o n a l i t y . T h e split half reliability c o e f f i c i e n t s r a n g e f r o m .60 to .87 with a m e d i a n of .79. T h e r e f o r e , t h e test is a b o u t as r e l i a b l e as t h e p u r c h a s i n g b e h a v i o r , which it is t r y i n g t o p r e d i c t . A g a i n , this a r o u s e s h o p e s t h a t t h e test c a n d o an a d e q u a t e j o b of p r e d i c t i n g p u r c h a s i n g b e h a v i o r , b u t c a u t i o n s a g a i n s t o p t i m i s m o n t h e size of t h e r e l a t i o n s h i p s t h a t c a n be e x p e c t e d . A w o r d a b o u t the c o n c e p t of " l o y a l t y p r o n e n e s s " a l s o s e e m s in o r d e r . In this s t u d y we a r e a t t e m p t i n g t o predict p u r c h a s i n g b e h a v i o r for three beverages: coffee, t e a , a n d beer. W e are a t t e m p t i n g to predict t h i s p u r c h a s i n g b e h a v i o r with p e r s o n a l i t y d a t a . T h e c o n c e p t of p e r s o n ality is d e f i n e d as " a relatively e n d u r i n g set of t e n d e n c i e s t o r e s p o n d in given w a y s t o given classes of s t i m u l i . " T h e r e f o r e , we a r e t e s t i n g w h e t h e r b e h a v i o r is a f u n c t i o n of p e r s o n a l i t y a n d n o t t h e o t h e r w a y a r o u n d . In o r d e r t o be a f u n c t i o n of t h e " r e l a t i v e l y e n d u r i n g " p e r s o n a l ity v a r i a b l e s , t h e b e h a v i o r in q u e s t i o n m u s t also be r e a s o n a b l y c o n s i s t ent. W e c a n t h e r e f o r e a s k : " C a n b e h a v i o r in a given s t i m u l u s s i t u a t i o n be p r e d i c t e d by t h e b e h a v i o r in a n o t h e r , s i m i l a r s t i m u l u s s i t u a t i o n ? " If purchasing behavior for coffee does not predict purchasing behavior for t e a , very well, t h e n we m a y be a s k i n g t o o m u c h of o u r p e r s o n a l i t y d a t a w h e n we r e q u i r e it t o p r e d i c t p u r c h a s i n g b e h a v i o r f o r t e a . F r o m t h e d a t a of C h a p t e r III we s a w t h a t t h e r e is a s i g n i f i c a n t r e l a t i o n s h i p b e t w e e n p u r c h a s i n g b e h a v i o r f o r c o f f c e a n d tea, e s p e c i a l l y f o r b r a n d a n d s t o r e b e h a v i o r . But of t h e r e l a t i o n s h i p s t e s t e d , o n l y e i g h t o u t of

PERSONALITY A N D PURCHASING BEHAVIOR RELATIONSHIPS

109

thirteen were significant; the correlations ranged in size f r o m .01 to .49 with a m e d i a n value of .26; at most these are not very strong relationships. P u r c h a s i n g behavior for coffee and beer is even less strongly related, with c o r r e l a t i o n s ranging f r o m .02 to .24 with a m e d i a n value of .15. W e can conclude that purchasing behavior for coffee, tea and beer do not predict each other very well, although there are unquestionably some significant relationships a m o n g them which indicate s o m e consistency in this behavior. We must t h e r e f o r e not be t o o optimistic in our expectations for prediction of purchasing behavior f r o m personality data. O n e other set of considerations needs to be taken into account as we examine the relationships a m o n g personality, socio-economic variables, and behavior. A s we worked to develop the best of predictors of the personality variables in the analysis sample, we discovered substantial relationships between t h e personality d a t a and the socio-economic data and that there was substantial collinearity a m o n g the set of personality variables. In reducing t h e collinearity in the set of personality variables in o r d e r to render t h e m usable in a multiple regression analysis, we removed those variables which were most highly correlated with the remainder of the set. These turned out on subsequent analysis to be personality variables which were most highly correlated with socioeconomic characteristics. This is consistent with our strategy of asking, " H o w much variance does personality add to socio-economic data in the prediction of buying behavior?" But it also means that if we were using personality scores alone (with the socio-economic components left in) the total prediction for personality data would seem better than when we separate it in this way from the pure socio-economic variables. In order to facilitate the reader's interpretation of the results, T a b l e A-4 (Appendix A ) presents an abbreviated dictionary of variables. Table A-4 can be folded out f r o m the book to provide a quick reference for the tables that follow. PREDICTIVE POWER A f t e r our e x p e r i m e n t s with the analysis sample, the set of predictors was reduced to t w e n t y - f o u r personality variables and five measures of socio-economic status. For measures of purchasing behavior, there remained thirteen for the coffee sample, fourteen for the tea sample, and twelve in the beer sample. Using an ordinary multiple regression p r o g r a m , the twenty-nine predictors were regressed against each of the dependent variables. T h e results of this analysis are s u m m a r i z e d in terms of predictive power in Tables 6-1, 6-2, and 6-3. T u r n i n g first to T a b l e 6-1. we note that in general the values of R for the total predictive battery against the purchasing behavior variables are quite low (Columns I, 5, and 9 of Table 6-1). While the

-a

_

— Orsi — _ _ o O

OOfNP- sC M » «O m r^i — r-i rn — — — O O O O O O O C

n • H •

XI

«

h-

S^g

UJ

co

0 C/5 01 UJ 0.

do —

c/o UJ < _i cu < O

z

^ or c/3 o Z H< UJ > a Q < J P C/5 < U LU > -J UJ m 01 Cu s


3

O — 30 >/•> — ri m ©— ©r-i© —© — © ©

>-Q o

> o

_1 yj UJ C/2 Q H (/) Z CDi Z ai i > < >ti Z >• < — — — a. O (- z a: Z H cc et X 1- H X 0m CD 75 co !/)V) < 3 H D