7- . 3 {71.9111 "171111111 lI111 1111111111111111111’,‘ 1111 '1“, ‘11:; PJIJWIEEI‘H‘ 1‘4: 1,111.1“? "“3." 1,3ng ‘ ’ . ‘1‘ In“. . u - .I’ a - "r fin'ufis -" I 1 . {I ,III'III 11,". II 11‘ III II ’1111111111111'9 IIIEII ' . Ii IIIIIIMX ‘ *5 "IIIIIIIIIIIIIT'I’MIIII I 1111111111 III III 1111"" I” I" II” {I}? “‘ 1.: 113.11. 11:11’111111311'1:”1111,111111 11111 1111111 111111‘11I ; I“; "I". 17".“ I, ‘11“! '1" 'L‘ 11 31111031113571 ' ’IV I ' *IIIII" ”1' ' " ‘1 l II I ‘ 111.11. I 1‘ .111 ‘ {III 1111' 1111. II 1 1 1.111111 11. . «III II‘I'I' 11' 15’1" 1" 1:1: ”1‘11 1' I III 1 I I III I M '."1 3%.}. : cu... '1 I I 27‘ {1:1111‘ ' ' " ‘11. ‘ ru 11" :I' .1 .; .q ": ., 2% ” ,z. : '111 111‘- 411‘ 1.E'-“ 1 ‘ j:' g a”) $11.: v . | a' ' 3?. "I"! ‘_ I"n'1' .1 I. .. v" ..'F 3. I’ll)“ {.111 ”111'; ‘1‘. ' .1 .f’ I ‘ . '. A.‘ l .I J . . v.- 1M1! I «é. . . n . I 11I1I 2* ' - . ' : 'II‘ '83.: . “ . ‘1“?! I: 1", . g “11:11 1 ‘I . . , .33. 3 I ' ‘ ' ' - ,I IIIIIIWIIII'III .111 I 11 “IL... a... “ 11 ' 11 ”1111111 11111111 I ——"—'=::n‘; W.‘ ~— M ———_ m —.‘~_ .1 ,.__Z— * ~~——~.—. — m “‘4— ,_. M; :12?“ —_-‘__._._ ‘_ ‘zgmrg .—_-.~:~.__ - —:—.‘3" Mt... _'— _ . .. . __.‘- - . r ‘ " _ 1“ -.a :‘a‘ 5".“" u as. E . ' ~45; . . 12"1‘ W‘h xqi.‘ Ii . o .I In; J. ‘ - > It" 42 ~11 {‘11: H1}! 11 ' . ’ 3:. I2 . \X, ' ‘ 3;; I 1:14]; l figl.” "" : ' . III" {‘83: “I???” ‘11111111'1 r111 ‘1‘ 1111 111111 1111 M11, 11 11111 111.11%! 1111 1 111 11.11111“. 11I‘II 7 . I 1'11, .113" $3". " ‘ : ‘ .1 w“: .' . 1131' 2 . ‘ ' 1 . ‘3: “:u‘i"1.uv1‘.‘¢: ‘11 ' 1111N1 II“ 111111111I 1%1111 {.IM; . I 'Ifi1WI I :1 1411111 ,_ 1‘. I,” ,1." I .1 IIII IIIII IWII11~I‘. z.vaI 11.1.1” ‘ 4 1'1,.’"' 1 A": II . ' . .‘I.*Q' I i 1 11‘1” .1111 i "I?“ I; . . f 111 11,1111i 1111110331111 r1 1111111 111111111 1111 ‘ ’1? 411111?) I I . l‘ H .l ' 1 111111111 - I II I I .. , II 1111.1I1111I1I 11n’“111111,wh I1 11,111 M 111111 11-111 31111111113111.i I; H I 1 11‘{ 11111 111 1111'1111 11 1: 11111111". 1 11111111111111 I ":1? '1 .:--:I M” ’ I! I,“ ' I III? II. '1'. 1'13» . i 3’. ll .1 {1.1 '..'!." . t . I 1" 1111' I. " ‘.' .. . 11‘ «gIIII'111 2 ' a. ‘ , . :I'III. . . ; no ._, , . ,.. - -.." 1"111 P1.H‘1I1111I' , ' 1 _ 1' " .1. 1111111111131"! “11. ' ' ‘ 1' ~ I . I . . 1 - 1. 1“'11:”1 ‘ " . ::.-' .:"-"'- 111111111121 1111111311131 : I'M; . '. E1131. r I .' v ‘ I :M I _ ‘ ‘- .... m ~—. 5.“. ‘ ,_._. Wm- $211-17.... .. n I - . - , —-—— - .. - - . . m— -.—. m - -- rvm— - - .___ _ - q... a. m. ._ h? --«._.L_.: LIBRARY il/fl/llllillfl l/lll/I/llilllII/l/llll ll/l/l/I/II/I/II/II/II _ Michigan fitate 93 10529 7521 University This is to certify that the dissertation entitled THE USE OF DEMOGRAPHIC CONSUMER VARIABLES AS SEGMENTATION CRITERIA FOR FROZEN FOOD PRODUCTS presented by RICHARD DALE LEININGER has been accepted towards fulfillment of the requirements for Ph.D. degreein Marketing 8 Transportation Administration Major essor Dmn% -O'Lm [Date igumt, R32, IRS/9x MS U is an Affirmative Action/Equal Opportunity Institution 0-12771 MSU LIBRARIES —__ ‘0 RETURNING MATERIALS: Piace in book drop to remove this checkout from your record. FINES wiii be charged if book is returned after the date stamped beiow. THE USE OF DEMOGRAPHIC CONSUMER VARIABLES AS SEGMENTATION CRITERIA FOR FROZEN FOOD PRODUCTS By Richard Daie Leininger A DISSERTATION Submitted to Michigan State University in partiai fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of Marketing and Transportation Administration 1982 Copyright by RICHARD DALE LEININGER 1982 © ii ABSTRACT THE USE OF DEMOGRAPHIC CONSUMER VARIABLES AS SEGMENTATION CRITERIA FOR FROZEN FOOD PRODUCTS By Richard Dale Leininger The most common operational source of market segmentation identification, demographic consumer characteristics, has not generally been established to be an effective delineator of purchasing behavior. But while there exists a significant amount of doubt, controversy, and confusion, with regard to the use of demographics in market segmentation research and practice, there would also seem to be both opportunity and potential for the demographic—based research of segmentation relationships. The primary purpose of this study was to investigate the use of demographic consumer variables, as a means of segmenting and predicting the purchase of consumer products. A further purpose of the study was to investigate the value of demographic and inter-product purchasing variables in relationship to the purchases of frozen food products, thus providing analysis of the value of demographics for the specific industry of frozen foods, far beyond reported prior studies relating to the market for frozen food products. Using a data base provided by the Akron Beacon Journal, relationships were examined using the analysis of variance and linear Richard Dale Leininger discrimination programs provided by the Statistical Package for the Social Sciences. Specific tests were made to determine the statistical significance of each relationship, as well as the linearity, bivariate predictive power, and multivariate classification ability of demographic and product-purchase variables, in relationship to each of twelve specific categories of frozen food products. The primary conclusion was that demographic variables ggg_be used to successfully predict and segment the purchasing behavior for frozen food products. A combined use of thirteen demographic variables was able to correctly identify the product purchasers of the twelve product categories at an average rate of 61.3 percent, compared to 65.5 percent by means of product-purchase variables. However, an additional finding was that many of the tested demographic rela- tionships were substantially non-linear, with half or more of the bivariate predictive power often lost via linear statistical measurement. Thus, it is suggested that the use of additional demographic variables and non-linear multivariate discriminant methods could be expected to significantly increase demographic-based buyer classification and segmentation success. ACKNOWLEDGMENTS A special recognition must be made to the exceptional support, cooperation, and encouragement provided by the chairman of this dissertation, Dr. Donald A. Taylor. Without his efforts, this dissertation would not have been possible, let alone completed. I will always be grateful for his contributions, both to this dissertation and to my doctoral program. Particular thanks must be given for the efforts by the additional members of the dissertation committee, Dr. George D. Hagenheim and Dr. David J. Closs. Their professional expertise and continued positive suggestions during the course of the research were very much appreciated and will not be forgotten. The Akron Beacon Journal generously furnished access to the data base used in the study, without which this study could not have taken place. Acknowledgment must also be made of the encouragement and efforts on my behalf by my friends and colleagues at the University of Wisconsin at Oshkosh, particularly Dr. Stanley D. Sibley and Dr. Eugene F. Drzycimski. Mrs. Grace Rutherford typed the final draft of this dissertation. Her outstanding knowledge of dissertation form was both notable and very much appreciated. Finally, a very special recognition is due the personal support that was continually given by Mary Beth, my wife. Her patience, love, and understanding were often tested, and helped make this study possible, as well as worthwhile. TABLE OF CONTENTS LIST OF TABLES ......................... CHAPTER I. INTRODUCTION ...................... An Introduction to the Problem ............ Justification for the Study ............. Hypotheses of the Study ............... Hypothesis One .................. Hypothesis Two .................. Hypothesis Three ................. Hypothesis Four ................. Research Design and Methodology ........... Perceived Limitations of the Study .......... Organization of the Following Chapters ........ Footnotes ...................... II. A REVIEW AND ANALYSIS OF MARKET SEGMENTATION RESEARCH Introduction ..................... The Concept of Market Segmentation .......... Earlier Foundations of the Concept of Market Segmentation .................. Earlier Contributions from the Social Sciences . . A Summary and Conclusion About the Origin of the Concept .................. The Application (and Misapplication) of Market Segmentation .................... The Dilemma of Normative Theory Versus Descriptive Explanation ............ The Suboptimization of Segmentation Efficiencies .................. Alternative Methods of Market Segmentation ...... A Review and Analysis of Some Exemplitive Demographic Segmentation Studies in the Marketing Literature . . Segmentation Analysis of the Frozen Food Consumer-- Prior Studies ................... Specific Discussion and Analysis of the Prior Studies ...................... The 1948 Scott Study ............... iv Page vii CHAPTER The National Frozen Food Association Studies . . . The Akron Beacon Journal Studies ......... Footnotes ...................... III. RESEARCH DESIGN AND METHODOLOGY ............ Introduction ..................... Source of the Research Data ............. Identification of Research Variables ......... Independent Variables .............. Dependent Variables ............... Analysis of the Data ................. Computer Programs Used in the Analysis ...... Statistical Techniques and Measurements ..... Footnotes ...................... IV. PRESENTATION OF FINDINGS ................ Introduction ..................... The Value of Consumer Demographic Variables ..... Analysis of Variance-~Test of Significance . . . . Analysis of Variance--Strength of Significant Relationships ................. The Linearity of the Predictive Relationship . . . Discriminant Analysis: Multivariate Classi- fication of Buyers Versus Non-Buyers ...... The Value of Demographic Variables Compared to Product-Purchase Variables as Predictors of Frozen Food Purchases ............... Analysis of Variance--Test of Significance . . . . Analysis of Variance--Strength of Significant Relationships ................. Summary of the Product-to-Product Relationship Strengths ................... A Comparison of Demographic Versus Product Variables as Bivariate Predictors of the Purchase of Frozen Food Products ........ Discriminant Analysis: Multivariate Ability to Predict and Classify Buyers, Based Upon Product-Purchase Variables ........... The Value of Total Food Purchases as a Predictor of Specific Frozen Food Product Categories ..... The Value of Freezer Ownership as a Predictor of the Purchase of Specific Frozen Food Product Categories ..................... Footnotes ...................... Page 132 133 136 137 138 140 148 151 153 CHAPTER V. CONCLUSIONS AND IMPLICATIONS .............. Introduction ..................... Conclusions ..................... The Use of Demographic Variables as Purchase Predictors ................... A Comparison of Demographic Variables to Product-Purchase Variables as Predictors of Frozen Food Product Purchases ........ The Value of Total Food Purchases as a Predictor of Specific Frozen Food Purchases ....... The Value of Freezer Ownership as a Predictor of Specific Frozen Food Purchases ......... Research Implications of the Study .......... The Use of Demographic Variables as Segmentation Criteria .................... Specific Versus General Segmentation ....... Statistical Significance Versus Predictive Strength .................... Linearity of the Demographic Relationships . . . . Management Implications of the Research Findings . . . Demographics as an Operational Management Technique ................... The Need for Basic Statistical Testing ...... The Need for Multivariate and Non-Linear Techniques ................... Some Practical Management Uses for Demographic Segmentation .................. Failure to Find the “Northwest Passage" ..... The Need for New Research Data Collection Methods .................... Suggestions for Future Research ........... Footnotes ...................... APPENDIX A. RELEVANT SECTIONS OF DATA BASE QUESTIONNAIRE FROM 1976 AKRON BEACON JOURNAL SURVEY ............. BIBLIOGRAPHY .......................... vi Page 154 154 154 154 164 167 168 169 169 172 174 177 177 178 179 181 182 183 184 189 190 196 TABLE 4.1 LIST OF TABLES The Statistical Significance of the Demographic Variables, in Relationship to Frozen Food Purchases ......... Number of Demographic Variables Significant ....... The Measured Strength of Demographics as Predictors Total Classification Efficiency of the Discriminant Analysis, Utilizing the Demographic Variables ...... Statistical Strength and Distribution of the Discriminant Analysis Using the Demographic Variables ......... The Significance and Measured Strength of Product-Purchase Variables Used to Predict Other Purchases ........ Total Classification Efficiency of the Discriminant Analysis, Utilizing the Product-Purchase Variables . . . . Statistical Strength and Distribution of the Discriminant Analysis Using the Product-Purchase Variables ...... The Predictive Effect of Total Food Purchases and Home Freezer Ownership on Frozen Food Purchases ........ vii Page 108 114 119 127 128 134 141 142 149 CHAPTER I INTRODUCTION An Introduction to the Problem One of the greatest single needs of the American marketing system is to be better able to understand the American consumer, not only to provide continuing profits for the companies involved, but also to better satisfy the wants and needs of the consumers. Thus, it seems somewhat paradoxical that while marketing literature speaks of “broaden? ing the concept of marketing"1 and advancing beyond mere profits to now include environmental and societal needs, that our knowledge of the con- sumer, and the practical application of what we do know, still remains so strikingly limited. In particular, the "average" or "middle-class" stereotypes of many companies' strategic marketing efforts have been supplanted in the actual consuming public by a vast and increasingly heterogeneous consumer market. As noted by Schwartz (1962), "from a single homo- geneous unit, the mass market has exploded into a series of segmented, fragmented markets, each with its own needs, tastes, and way of life."2 Similarly, Toffler (1971) concludes that the segmentation of the mass market has begun to change the face of American industry beyond recog- nition, with an astonishing diversity of products offered to the consumer.3 The classic article by Wendell Smith, "Product Differentiation and Market Segmentation as Alternative Market Strategies" (1956) pro- vides an explanation for this growing characteristic of the American, via his definition of market segmentation: Segmentation is based upon development of the demand 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 disagregative in its effects and tends to bring about several demand schedules where only one was recognized before.“ Today, the concept of market segmentation has become a basic principle of modern marketing, and is regarded as one of the marketing profession's "hottest products."5 Thus, Frank, Massy, and Wind (1972) suggest that "during the past ten years, it has had as great an impact as any other theoretical construct that has served as a tool for organized thinking about the nature of the marketplace."6 Both academic and industry practitioners of marketing have maintained an active and sometimes frustrating interest in segmentation research and application. For example, Waldo (1973)7 found that a sample of marketing executives indicated that they considered "recog- nizing, defining, and segmenting markets" to be the most important single problem facing the marketing decision maker. Similarly, Springer (1975)8 asserted that "segmentation is the key" for future marketing success. However, despite this indicated interest, and a large number of reported academic and industry studies in the subject area of consumer market segmentation, the available, specific segmentation research, as applied to specific product markets, has been strikingly limited in both scope and practical application. Thus, in a 1976 article, Dhalla and Mahatoo concluded: The poor performance of many segmentation criteria tested so far can be attributed to the fact that too often researchers are anxious to find a magic formula that will profitably segment the market in all cases and under all circumstances. As with the medieval alchemist looking for the philosopher's stone, this search is bound to end in vain. There is no single algorithm that can be applied across all market studies. Each case must be viewed as a unique and potentially different situation.9 Plummer (1974) seems to echo this conclusion, with the observation that "so often, segments developed from a study on one product category have little or no relevance to another category."‘° Likewise, Wind and Green (1972)11 suggest that it may be necessary to think of each segmentation study as an ad hgg_study, having specific validity for only the particular problem, product, and situation. Yet, as noted by Weisenberger (1977),12 a constant temptation exists for a multi-product, multi-brand manufacturer to seek cost savings in consumer segmentation research for a family of prdocuts via one major study, rather than by several individual studies for specific products. He concludes that more product-specific segmen- tation studies would "offer a more finely tuned picture of the con- sumer in the particular situation," although general studies should still be sought to provide some insights to the general segmentation relationships that may exist.13 Thus, in an allegorical sense, failure to find the "Northwest Passage" to universal identification of consumer segmentation relation- ships, should not be perceived as a failure of the research "expedition." Investigations of the existing but still uncharted markets for particular products may prove to be a practical and strategic route toward achieving useful generalizations of segmentation relationships, and may be the ggly_effective route toward utilization of segmentation theory in the actual marketing management of those product categories for which segmentation relationships have not been specifically and accurately "mapped." Similarly, attempts to segment a product market by means of a segmentation map developed for another product market may result in a failure of the marketing program, not caused by the failure of the concept of market segmentation Egr_§g, but rather, caused by the use of the wrong "map" of segmentation relationships. In addition, the conclusions of past segmentation studies are often guilty of semantic "license" in stating that certain segmentation variables were not valid predictors for segmentation purposes, rather than giving a more objective report that the particular statistical method, research design, and data sample of the research did not show useful results. Thus, in a commonly cited study by Frank, Massy, and Boyd (1967), 57 grocery products were examined for demographic ties to product category consumption, with the reported and often quoted conclusion that "socioeconomic characteristics are poor predictors of consumption for a broad range of specific grocery products."’“ They concluded that rates of consumption for other products were usually better predictors of consumption than were demographic variables. However, for the sake of simplicity in using such common, statistical techniques as multiple regression (as was used in the Frank, Massy, and Boyd study)15 assumptions are often made as to the linearity of the test variables, as well as assuming that nominative, categorical variables are interval in nature--two assumptions that §h9g1g_severely limit the use of linear regression and similar techniques, relative to the testing of demographic variables. Moreover, as noted by Bass, Tigert, and Lonsdale (1968), "for market segmentation, the essential question is whether it is possible to identify grggp§_of consumers with different mean puchase rates dependent on certain variables such as income, age, and occupation."16 Thus, they suggest that since multiple regression provides a measure of the variation in individual purchases, this method does not provide adequate measure of the usage rates that are predictable for different market segment grggps, Therefore, they conclude that the low R squares reported in the Frank, Massy, and Boyd study17 implies that variation within the segments is great, but does not necessarily mean that group mean values will not be significant, if properly tested and measured. In addition to questions about the appropriateness of the linear regression methodology, and the assumption, without testing, of a linear data relationship, there are additional questions that might be raised about the Frank, Massy, and Boyd study,18 such as a limited sample size (491 households), admitted problems of coding and multicollinearity within the sample data, as well as the growing age of the data base used (1960); all adding to probable reasons for doubt, and a need for re-examination of this study's conclusions about the usefulness of demographic variables as predictors of consumer purchasing behavior for grocery products. Indeed, much of our current "knowledge" about market segmen— tation would be better classified as dictum, rather than established fact. For example, volume segments are thought to exist, logically believed to be important, dutifully discussed in virtually every mar- keting textbook, but have rarely been identified in the marketplace. Thus, unfortunately, if market segments cannot be properly identified and defined, let alone predicted, segmentation might be better classi- fied as an empty concept, rather than an established principle that is taught as a basic strategic foundation for proper marketing theory and practice. Justification for the Study The most common, operational source of market segmentation identification, demographic consumer characteristics, has not generally been established to be an effective delineator of purchasing behavior. But while there exists a significant amount of doubt, controversy, and confusion with regard to the value of demographics in market segmenta- tion research, there would also seem to be both opportunity and potential for demographic-oriented research of market segmentation relationships. If predictive demographic relationships can be more firmly established, demographic variables might be expected to become the most practical segmentation technique for the marketing practitioner, since demographic data and projections can provide significant advantages of identifiability and accessibility relative to the use of other segmen- tation techniques.19 Moreover, as additional relationships can be established for specific product markets, both the theory of market segmentation and general marketing can be seen as taking an additional step toward scientific generalization of marketing behavior. As noted by Leone and Schultz (1980), "basic answers to questions of h9w_marketing variables are related lead logically to new questions of why they are related. Like our fellow scientists, we must first describe marketing behavior in order to bring us closer to explaining it."2° The specific purpose of this study will be to determine if demographic consumer variables can be shown to be an effective pre- dictor of specific product purchases, and thus demonstrate a practical potential for actual definition of market segments. The study will have the effect of re-examining past conclusions about the effectiveness of demographic variables for market segmentation purposes, particularly for food products, but with probable conclusions relating to all consumer products. In addition, the specific methodology of this study will have the result of providing a more finely tuned picture of the value of demographic segmentation in the specific industry of frozen foods. Despite a number of prior reported studies of the frozen food consumer, the status of existing knowledge about the segmentation potential of this industry remains surprisingly vacant. First of all, most prior studies of the purchasers of frozen foods have limited their measurements of purchasing or consumption to the very broad category of total frozen food consumption, despite the wide variety of product offerings within the total frozen food market. As concluded by Weisenberger, segmentation of more product-specific segmentation relationships are more likely to provide effective seg- mentation relationships than for more generalized, multi-product, multi-brand families of products,21 such as the total frozen food market. Second, among the published and available studies of the frozen food consumer, there seems to be virtually no reported sta- tistical analysis of relationships that may exist. Most reported conclusions seem to have been drawn from general perusal of two-way cross tabulations of survey totals, with no reported tests of statis- tical significance or predictive strength of the relationships between demographic variables and frozen food consumption. Third, within reported frozen food industry research, the use of multivariate analysis of segmentation relationships has been almost totally absent. With the exception of a few three-way cross tabulation tables, without any reported statistical analysis of the results, multi- variate demographic relationships simply have not been investigated, either for the total purchases of frozen foods, or for the separate categories of frozen food products within the total category of frozen foods. Thus, this study of demographic segmentation, via data for the purchases of frozen food products, has the specific potential of being a landmark study for the frozen food industry, since the past reported studies of the frozen food consumer would seem to be describable as generally descriptive, but with very limited statistical analysis of the actual existence or predictive value of segmentation relationships that may exist for this product market. If successful in properly identifying and predicting demographic segments for frozen food products, the results would thus be meaningful for both the frozen food industry and for the general practice of mar- keting. Meaningful explanation of buyer patterns and successful clas- sification into purchasing segments could thus provide the basis for further research, better understanding of the consumer market, more effective operationalization of the principle of market segmentation, as well as more profitable marketing management within the frozen food industry itself. However, if the techniques used and/or the survey data of this study does ggt_demonstrate that demographics can effectively predict and segment frozen food purchasers, that result should be perceived as a valid contribution in itself. If this study fails to establish significant segmentation values for demographics, it would have corroborated earlier studies that have come to this conclusion. As noted by Leone and Schultz (1980), "replication is the key to generalization (in marketing theory) for without it, in the broadest sense, we have no corroboration of research results . . . only by extending findings to other data sets do we perceive the generality of marketing relationships."22 However, if demographics exhibit useful segmentation potentials in the study, both the use of demographics as a marketing tool, and the practical application of marketing segmentation as a management technique, would be greatly enhanced. 10 Finally, the study will provide specific, needed consumer profile information to the frozen food industry, with a method and observed results that may provide specific guidance for marketing practice and future consumer research within this specific industry. Even a finding of poor demographic potentials would in itself be valuable information to marketers in the frozen food industry, as an incentive to seek other segmenting factors for frozen food products, or in seeking other research techniques that may ultimately unlock the way to the effective, practical usage of demographic consumer variables for the market segmentation of frozen food products. Thus, this study would seem to have potential for both the general study of market segmentation, and for the specific industry of frozen foods, as well. Hypotheses of the Study Hypothesis One The major hypothesis of this study will be that demographic variables ggg_be used to successfully predict and segment the purchasing behavior for consumer products. Specifically, this study will focus on the ability to predict and segment the purchasing behavior for specific categories of frozen food products, as a means of testing this major hypothesis. During the course of testing this major hypothesis, four sub- hypotheses will be considered, relative to the degree that the demo- graphic variables can be tested and proven effective in predicting product purchases. These subhypotheses will be as follows: 11 a. There will be a statistically significant, bivariate relationship between the individual demographic variables and the purchases of specific frozen food product categories at the .01 significance level. b. There will be a bivariate predictive relationship between individual demographic variables and the purchases of specific frozen food product categories with a total, non-linear predictive power of 5 percent or more. c. There will be a linear predictive relationship for demographic variables that are significantly related to the purchase of specific frozen food products on a bivariate basis. d. There will be a multivariate predictive relationship between the total set of demographic variables and the purchases of specific frozen food product categories with the resultant ability to correctly classify 75 percent or more of the product purchasers and non- purchasers. Hypothesis Two The second hypothesis of this study will be that demographic variables will prove to be equal to or more effective than the use of product-purchase variables, in the ability to predict the purchases of frozen food product categories, and to classify them into segments of purchasers and non-purchasers of each specific frozen food category. Hypothesis Three The third hypothesis of the study will be that both the demographic variables and the specific frozen food product-purchase variables will prove to be more effective than the total food purchases by the consumer-respondents, relative to the prediction of the purchase of specific frozen food categories. 12 Hypothesis Four The fourth and final hypothesis of the study will be that freezer ownership will not provide either a statistically significant or meaningful predictive relationship to the purchases of the specific frozen food categories. Research Design and Methodology The study will be based upon a survey of frozen food purchases conducted by the Akron Beacon Journal, published in summary, cross- tabulation form under the title, ”Frozen Food Study 1977."23 The Akron Beacon Journal has furnished access to a survey data base of 720 inter- views, conducted on an in-home basis by Bardsley & Haslacher, Inc., a Pala Alto, California marketing research firm. The survey interviews were collected from January 1976 to June 1976, using a random selection process among permanent-residence households in the Akron, Ohio metropolitan area. The survey provides coded data relating thirteen demographic variables to the purchases of twelve specific categories of frozen food products, as well as providing data for total grocery purchases by each household, and for ownership of a separate freezer unit. Survey data will be processed and analyzed via the use of the Statistical Package for the Social Sciences,2“ using the subprograms of "Breakdown" and "Discriminant." Subprogram "Breakdown"25 will be used to apply a one-way analysis of variance, testing statistically whether the group means of frozen food purchasers and non—purchasers are significantly different among 13 the subpopulations of the survey which are represented by the categories within each of the demographic variables in this study. Next, this same subprogram will be used to calculate the total linear and non—linear bivariate predictive strengths of the demographic variables, with respect to each of the bivariate relationships in which a demographic variable is found to be statistically related to the purchase of a specific frozen food product category. Finally, the SPSS subprogram "Discriminant"26 will be used to calculate the linear, multivariate amount of purchasing variance which can be explained by the combined use of the thirteen demographic variables which will be employed in this study. Via the discriminant analysis program provided by "Discriminant," a mathematical function will then be derived that best segments the purchasers from the non- purchasers of each specific frozen food product category, based upon the discriminating, demographic variables. The same subprograms, “Breakdown" and "Discriminant" will also be used to test the predictive value of knowing the purchases of the other frozen food product categories, relative to the purchases of each specific frozen food product category, on both a bivariate and a multivariate basis, comparable to the testing of the demographic predictive variables of this study. Subprogram "Breakdown" will also be used to examine the bivariate statistical strength of total grocery purchases and freezer ownership, relative to the purchase of the specific frozen food product categories. 14 Perceived Limitations of the Study The results of this study are subject to the following limitations, that may or may not significantly affect the results of the empirical analysis. First, the results are based upon a study conducted in only one, limited geographic area, the Akron, Ohio metropolitan area. The possibility that a more national survey sample might give different observed results must be noted. In addition, the single geographic survey area prevents measurement of geographic differences that may exist between regions of the country, and of course, may bias the results via the single, mid-western Ohio area used to collect the survey. However, it may be noted that the primary purpose of this study is to test the ability to effectively segment between frozen food purchasers and non-purchasers, not to study the purchasers of frozen food products p§r_§g, Second, the limited number of demographic characteristics used in the analysis must limit conclusions about the total possible segmentation abilities of all demographic variables that might be available to the academic researcher or the industry-level marketing practitioner. Beyond the variable of geographic region, which has already been mentioned in this section, other demographic variables that might prove useful would include the ethnic background, religion, number of children, age categories of children in the family, and the stage in the family life cycle. Third, the purchase data was collected on a one-time basis, relying on the respondents' memory and truthfulness of recall of 15 the past month's purchases. Purchase data based upon a more longitudinal data base would, of course, be more accurate and desirable. However, the cost of obtaining such a sample over an extended survey period, with a large enough sample size, becomes financially prohibitive to all but the largest companies or govern- mental agencies, and in any case, such data were not available for this study. A fourth factor with regard to the study that should be noted is the age of the data, collected during the first six months of 1976. This possible weakness would certainly be more significant in specific interpretation of the results as they apply to actual, current purchaseS‘ of the specific frozen food products of the study. However, it may be noted that this study will nevertheless provide the most current and comprehensive analysis of frozen food purchases known to exist in the frozen food industry. Moreover, as a basis for demonstrating the potential ability to segment via demographic variables, this limitation seems much less relevant to the possible results of this study. A fifth factor with regard to the data source that should be noted is the possibility of seasonal variation in the consumption of the specific product categories of this study, particularly with regard to the seasonal availability of fresh substitutes. However, the first six months of the year, during which this survey was collected, should present the least distortion in total food purchases, caused by seasonal, fresh substitutes, as well as avoiding possible problems due to hot weather, such as less use of home ovens, vacations, etc., that might possibly cause abnormal shifts from usual purchase patterns. 16 Finally, a sixth factor mgy_be the actual linearity of the demographic relationships in the survey sample. The typical research strategy has been to assume that the bivariate relationships were, in fact, reasonably linear, and thus usable in linear, statistical tech- niques. The factor of linearity will be tg§t§g_in this study, rather than being assumed. While the bivariate predictive relationships of the demographic variables of this study are hypothesized to be reasonably linear, a finding of non-linearity would be expected to limit the accuracy of the linear, multivariate discriminant analysis to be used in this study. To the extent that the demographic variables are examined, and found to lose actual predictive strength via the use of linear measurements, the net conclusions about the multivariate discriminant analysis would thus be limited. Organization of the Following Chapters The remainder of the dissertation consists of four chapters. Chapter II provides a review of prior literature that is relevant to this research study, including the early develOpment of the concept of market segmentation, general application and misapplication of the concept, and a specific review of prior market research studies in the product category of frozen foods. Chapter III presents an explanation and discussion of the research design and methodology used in the study, including the source of the data base, computer programs employed, statistical techniques used, and the statistical criteria used for testing of the hypotheses. 17 Chapter IV contains the research findings of the study, while Chapter V presents the conclusions of the study, including a specific consideration of the research hypotheses. Chapter V also includes a discussion of the implications of this study, as well as presenting some suggestions for further research. FO0TNOTES--CHAPTER I 1Philip Kotler and Sidney J. Levy, "Broadening the Concept of Marketing," Journal of Marketing 33 (January 1969): 10-15. 2Kenneth Schwartz, "Fragmentation of the Mass Market," Dun's Review and Modern Industry, July 1962. 3Alvin Toffler, Future Shock (New York: Bantam Books, 1971), p. 265. “Wendell R. Smith, "Product Differentiation and Market Seg- mentation as Alternative Marketing Strategies," Journal of Marketigg 21 (July 1956): 3-8. 5Ronald E. Frank, W. F. Massy, and Yoram Wind, Market Segmen- tation (Englewood Cliffs, N.J.: Prentice-Hall, Inc., 1972), p. 256. 6Ibid. 7C. N. Waldo, "What's Bothering Marketing Chiefs Most? Segmenting," Advertising Age, 4 June 1973, p. 77. 8Jack Springer, "1975: Bad Year for New Products, Good Year for Segmentation," Advertising Age, 10 February 1975, pp. 38-40. 9Nariman K. Dhalla and Winston H. Mahatoo, "Expanding the Scope of Segmentation Research," Journal of Marketing_40 (April 1976 : 36. 1°Joseph T. Plummer, "The Concept and Application of Life Style Segmentation," Journal of Marketing 35 (April 1971): 35. llYoram Wind and P. E. Green, "Some Conceptual, Measurement, and Analytical Problems in Life Style Research," in Life Style and Psychographics, ed. R. Haley (Chicago: American Marketing Association, 1974), pp. 99-126. 12Terry Mathew Weisenberger, "Generalized Market Segments: A Study Using Selected Convenience Goods in Vigo County, Indiana" (Ph.D. dissertation, Michigan State University, 1977), pp. 123-124. 13Ibid., p. 123. 18 19 1"Ronald E. Frank, William F. Massy, and Harper W. Boyd, "Correlates of Grocery Product Consumption Rates," Journal of Marketing Research 4 (May 1967): 189. 1SIbId., PP. 184-190. 16Frank M. Bass, Douglas J. Tigert, and Ronald T. Lonsdale, "Market Segmentation: Group Versus Individual Behavior," Journal of Marketing Research 5 (August 1968): 264-270. 17Ibid. 18Ibid. 19Johan Arndt, Market Segmentation: Theoretical and Empirical Dimensions (Bergen, Norway: Universitetsforlaget, 1974), pp. 24—25. 20Robert P. Leone and Randall L. Schultz, "A Study of Marketing Generalizations," Journal of Marketing 44 (Winter, 1980): 15. 21Weisenberger, pp. 123-124. 22Leone and Schultz. 23"Frozen Food Study, 1977," Akron Beacon Journal, Akron, Ohio, 1977. 2"Norman H. Nie, C. Hull, J. Jenkins, K. Steinbrenner, and D. Bent, Statistical Package for the Social Sciences (New York: McGraw-Hill Book Co., 1975). 25Ibid., pp. 249-275. 26Ibid., PP. 434-467. CHAPTER II A REVIEW AND ANALYSIS OF MARKET SEGMENTATION RESEARCH Introduction The purpose of this chapter is to present a review and analysis of the marketing research and literature in the area of market segmentation. First, the early foundations and literature relating to the develOpment of the concept of market segmentation will be examined. Second, the general application and misapplication of the concept will be examined, focusing on the observed discrepancy between the academic ideals of segmentation and the observed problems that have occurred in actual marketing research and practice. Third, alternative methods of market segmentation will then be generally examined, with particular focus on the value and use of demographic indicators of segmentation. A review and analysis will be presented, with regard to some exemplitive studies and viewpoints about the use of demographic segmentation methods. Fourth, a specific review and analysis will be given, with respect to prior market research studies in the product category of frozen foods, as they may relate to the empirical research of this study. 20 21 The Concept of Market Segmentation The origin of the concept of market segmentation is usually traced to the publication of Wendell Smith's classic article, "Product Differentiation and Market Segmentation as Alternative Market Strat- egies" in 1956.1 In this article, Smith defines market segmentation as "viewing a heterogeneous market (one characterized by divergent demand) as a number of smaller homogeneous markets in response to differing product preferences among important market segments."2 He explains that segmentation is based upon demand—side adjustments of the product and marketing efforts of the firm, in response to consumer requirements, such that several demand schedules are perceived where only one market demand had been recognized before. In a somewhat prophetic aside, Smith remarks that "to a certain extent, market segmentation may be regarded as a force in the market that will not be denied." Six years later, in 1962, Schwartz noted: It is nothing less than a revolutionary transformation which has come over the mass market during the past five years. From a single homogeneous unit, the mass market has exploded into a series of segmented, fragmented markets, each with its own needs, tastes, and way of life.3 Similarly, Frank, Massey, and Wind (1972) noted that "during the past ten years it has had as great an impact on marketing thought as any other theoretical construct that has served as a tool for organizing thinking about the nature of the marketplace."“ Today, the concept of market segmentation seems to have pro- gressed far beyond the status of an "alternative strategy" as originally suggested by Smith, such that it has become a basic principle of modern 22 marketing. For example, in a current marketing text, Nickels (1978) refers to market segmentation as "g_universal marketing function that enables buyers and sellers to concentrate their efforts on people who are most likely to participate in a mutually satisfying exchange."5 Virtually every marketing text of reasonably current date would be expected to devote at least a chapter to the concept and application of the principle of market segmentation. Moreover, the general topic of segmentation has attracted a plethora of past and present articles, most of which cite Wendell Smith as the originator of the concept. Earlier Foundations of the Concept of Market Segmentation Although Smith can appropriately be credited as instrumental in bringing the attention of marketers to the concept of market segmentation, the origin of the concept and even the term, market segmentation, may be found in the earlier literature of economics and marketing. The origin for the concept of market segmentation might be perceived in marketing literature as early as 1915, when Arch Shaw stated that the market was comprised of "market contours," based upon social and economic differences of people.6 He suggested that a firm should analyze its market by strata, and then adjust its selling strategies to match its market contours. Arndt (1974)7 suggests that the concept of market segmentation can be traced back to the theories of imperfect and monopolistic com- petition, as developed by Chamberlain (1933)8 and Robinson (1933).9 23 Thus, Arndt suggests that Robinson's decision rule (1948)10 for price discrimination in price adjustments to different price elasticities that apply to mutually exclusive segments of the market, might be generalized to other market mix ingredients as well-~thereby providing a conceptual base for market segmentation. Schyberger (1977)11 argues that the originator of the market segmentation concept seems to be Joel Dean. In the textbook, Managerial Economics (1951),12 Dean stressed the need to analyze separate market segments, relating to homogeneous characteristics of pricing, promotion, and distribution. These segments were to be defined by geographic region, product use, channel of distribution, price sensitivity of the customer, etc. In this book, Dean specificially used the term, market segmentation, as well as a sound description of the concept. Moreover, while few marketing academicians would dispute the designation of Smith's 1956 article as a classic addition to marketing theory and practice, Wroe Alderson's text, Marketing Behavior and Execu- tive Action, published in 1957, is typically considered a landmark con- tribution to marketing literature.13 Within this text, Alderson presents a parallel discussion, of product matching and positioning of the firm in the market, that may have been as much or more important in drawing the attention of marketing theorists and practitioners to the concept and practice of market segmentation. For example, discussing his theory of market behavior, Alderson describes consumer demand in the United States market as being radically heterogeneous, with this heterogeneity being substantially independent of the actions of marketers. He noted that marketers develop their 24 activities (i.e., product, location, price, package, etc.) to appeal to consumer needs better than competitors do, thus leading to hetero- geneity of supply as well as demand.”' Thus, marketing middlemen attempt to match the heterogeneous consumer demand with heterogeneous segments, via the process of sorting (i.e., accumulation, sorting-out, allocation, and assorting) resulting in a set of homogeneous supply segments being marketed by the seller-sorter.15 In a similar theory, Alderson suggests that firms seek a position of safety and power, similar to a biological being, which he labels an "ecological niche."16 He suggests that this ecological niche is made up of a core and a fringe, with the core being that part of the market which is best suited to the activities, safety, and power of the organization. Thus, a firm should develop a niche in the market, and protect its survival by finding and meeting the needs of a selected core of customers. Therefore, Alderson must be credited with providing a theoretical analysis and logic for the concept of market segmentation, even though he did not label it by those exact words. Early Contributions from the Social Sciences In addition, prior to Smith's conceptual article, there were numerous examples from the literature of the social sciences, including economics, sociology, and psychology, that provided data and scholarly observations relating to the basic concept that consumers were sub- stantially heterogeneous, and thus, could be expected to exhibit different but predictable buying behavior. 25 For example, Thorstein Veblen, educated as a traditional economist, was influenced by the new field of social anthropology in his description of leisure class buying behavior. His 1899 book, Ih§_ Theory of the Leisure Class,17 is noteworthy for the concept of "con- spicuous consumption," in relation to the theory that economic consump- tion is more by prestige-seeking than by economic-oriented purchasing behavior, emphasizing the strong emulating factor in the purchase of conspicuous, socially important products such as clothing and houses. In a more "modern" application of this concept, Whyte (1954)18 observed that some buyers avoided the purchase of conspicuous consumption items such as air conditioners or similar appliances until their neighbors did, thus exhibiting a form of "conspicuous non-consumption" to try to "fit in" rather than stand out among their relevant reference group. In a current update on this concept, it may be observed that "con- spicuous non-consumption" of energy has become something of a fad beyond probable economic or patriotic motivations, with solar heating and four-cylinder Cadillacs being sold to consumers who would seem to be the least sensitive to the price of energy. Another prominent pre-1956 contribution from the social sciences would be W. Lloyd Warner's social class system, an important contribu- tion to sociology as well as marketing literature. Warner and Lunt (1941)19 described a system of six social classes in our society: the upper-upper, or old, well established socialite families; the upper- middle, or the business and professional families; the lower-upper, consisting of recently affluent families; the upper-lower class, 26 consisting of the wage earner and skilled tradesman; the lower-middle class, made up of white-collar salaried workers; and finally, the lower-lower class, consisting of unskilled laborers. To determine the social class ranking, Warner developed a classification system based on the type of income, rather than income per_§g, occupation, type of residence, and place of residence. Under Warner's guidance, and based upon his social class stratification system, the Chicago Tribune conducted a series of studies regarding family buying patterns in the Chicago area.2° As discussed by Martineau (1958),21 the Chicago studies brought together a body of evidence that showed that the lower-class person is signif- icantly different in where he buys and what he buys, differing in the symbolic value of purchases, as well as economic considerations. In this analysis of the Warner studies, Martineau develops three basic premises which are highly significant to the application of the market segmentation concept: 1. There is a social class system operative in metro- politan markets, which can be isolated and described. 2. It is important to realize that there are far-reaching psychological differences between the various classes. 3. Consumption patterns operate as prestige symbols to define class membership, which is a more significant determinant of economic behavior than mere income.22 Sibley (1972) notes that while market researchers have applied a wide variety of independent and dependent variables in testing and applying the concept of market segmentation, the most prevalent group of variables found in segmentation studies have been demographic and socioeconomic classifications of the consumer.23 From this perspective, it becomes rather fascinating to note that a number of studies, relating 27 to the standard of living and general consumption have been conducted in the past 200 years, as relating to demographic changes within the consumer population.2“ For example, Zimmerman (1936) noted that Frederick LePlay and his followers had affected the study of the social sciences for nearly 150 years, relative to consumption and societal well being.25 According to Zimmerman, LePlay and his associates had conducted family budget and expenditure studies in Algeria, Arabia, Austria, Bali, Belgium, Canada, China, Corsica, Egypt, Great Britain, the United States, and several other countries. Expenditures were classified by general product or expenditure type, and the family was classified by variables such as income, type of income, rank of the family, age, sex, and similar demographic variables. His studies provided an important foundation for further work by economists, sociologists, and other social scientists that followed, as well as contributing a strong influence upon present methodological procedures by the Bureau of Labor Statistics, even today.26 Thus, Ernst Engel, an eighteenth century German statistician, strongly influenced by LePlay's earlier work, conducted budget studies while director of the Saxony Statistical Bureau from 1850 to 1858, leading to systematic relationships being established between income and expenditures.27 Based upon the earlier work by Engel, Wright (1875) interpreted the basic relationships relating to consumption, and published the following four statements that have become known as "Engel's laws": 28 1. That the greater the income, the smaller the relative percentage of outlay for substinance. 2. That the percentage of outlay for clothing is approximately the same, whatever the income. 3. That the percentage of outlay for lodging, or rent, and for fuel and light, is invariably the same, whatever the income. 4. That as income increases in amount, the percentage of outlay for sundries becomes greater.2 Although Engel's original "law" ggly_related to food, this relationship, as expanded and interpreted by Wright, has often been further "reinterpreted" for other products purchased by the family, and incorrectly attributed to Engel.29 Such was the case in an early statistical study of consumer buying power conducted by Berridge, Winslow, and Flinn in 1925.3° While many of these basic relationships, as developed by Engel and Wright, may have altered for present consumer segments, since they were developed to describe a more subsistence-oriented economy, they provided a theoretical foundation for later income- consumption theories by economists such as Keynes (1936)31 and Duesenberry (1952),32 and thus provided continued academic foundations for the broader concept that consumers were not all alike, and therefore could be expected to exhibit different patterns of consumption. However, even though economics may have helped spawn, or at least lay, the theoretical foundations for the marketing principle of market segmentation, it seems somewhat paradoxical that many traditional economists continue to regard the applied technique of segmentation as 29 an imperfection within the market that is more likely to cause undesirable social consequences, rather than resulting in an increase in heterogeneous consumer need satisfaction.33 However, despite this past and continuing normative bias against perceiving the existence of a heterogeneous consumer market, the 19305 brought a growing body of analytic evidence relating to the effect of demographic differences upon consumption. In particular, the 1935-36 Consumer Purchase Study,3“ with its extensive tabulations, served as a catalytic springboard for analysis of the effect of demographic factors such as age, income, marital status, family size, and family type upon the consumption of the American family.35 Significant differences detected in the income and consumption patterns by age and family type served as the ground- work for later, post-war study of these variables. Several such studies were contributed by the Survey Research Center at the University of Michigan, based upon the 1947, 1948, and 1949 Surveys of Consumer Finances, which served as a data base for later researchers. For example, Fisher (1952) combined these surveys, made avail- able by the Survey Research Center, along with the 1935-36 Consumer Purchase Study,36 to develop a hypothetical family life cycle, and attempted to tie this empirically to consumption and saving by the family.37 Another early study of consumption behavior was conducted by Scott (1948).38 In this study, eight categories of frozen food products were examined, in relationship to demographic categories of income, race, occupation, age of homemaker, and whether or not the homemaker 30 was working outside the home. In addition, a set of attitudes were also tested, by income group, about the use and desirability of frozen food products. The method of analysis was simple two-way cross- tabulation, without statistical analysis, but definite differences were noted between demographic groups. For example, the average family which purchased frozen foods in the study, bought 2.4 pounds per week within the low income family set, while the high income consumers purchased an average of 7.9 pounds per week.39 This particular study will be discussed in more detail, later in this chapter, with regard to the literature background relating specifically to the segmentation of frozen food products. The continuing interest and growing body of evidence about consumer consumption differences stimulated the organization of a conference by Consumer Behavior, Inc. (the legal name adopted by the Committee for Research on Consumer Attitudes and Behavior) relating to economic, marketing, and sociological behavior of consumer family buying units,”° which resulted in the book which grew out of this conference, edited by Lincoln Clark (1955)."1 In a paper presented at this conference, Fisher outlined some of the research potentials relating to the effect of the family life cycle, including the study of shifting patterns of consumption."2 Based upon the 1953, 1954, and the 1947-50 Surveys of Consumer Finances, from the University of Michigan, Lansing and Morgan presented a study which treated family life cycle as an independent variable with respect to family income, savings, and consumption patterns.“3 They noted a bi-modal stream of income, caused by a dip in family earnings when the 31 wife typically stopped receiving wage or salary income after the birth of the couple's first child. An analysis of the purchases of homes, cars, and television sets indicated a sequential order for family purchase of these items, relating to the stage in the family's life cycle. In a related study, Arton (1955) used data from the Marketing Research Corporation of America to provide specific examples of con- sumption differences between families of different age and family demographic composition, with respect to products such as strained baby foods, cereals, various prepared food items, and personal care items.““ Similarly, Miller (1955) discussed the effect of the family life cycle upon advertising impact."5 A Summary and Conclusion About the Origin of the Concept The conceptual foundation for the marketing concept of market segmentation is based upon the hypothesis (or observation) that con- sumers are different, and therefore, that they will exhibit different consumption responses. The strategic principle of market segmentation then suggests that marketers should aim at the different needs of these different consumers with an offering of different products, aimed at targeted consumer groups in the marketplace. While the classic 1956 article by Wendell Smith“5 is commonly cited as the origin of this strategic concept, there is ample published evidence that Smith neither originated the term of "market segmentation" or the concept. In fact, a close reading of the Smith article, and a 7 contemporary book by Alderson,” shows that both of these early 32 marketing writers were reporting and explaining a marketing practice that was becoming more common in the American market. Both of these writers sought to point out and explain the strategic process of market segmentation as based upon ex post facto descriptions of observed practices by marketing firms of their time. Perhaps a fairer assess- ment of the literary roles of both Smith and Alderson is that they served as catalysts to marketing academicians, and to marketing practitioners, in recognizing, understanding, and applying a process that was fast becoming a characteristic of the burgeoning American marketplace after World War II. As noted by Brandt (1967) the strategy of segmenting a market had been naturally applied since the application of mass production, earlier in the century."8 He notes the natural transition from "open markets" (e.g., geographic segmentation) to "fragmented markets," based upon progression in both the communication and transportation methods available to the marketing practitioner. Brogowicz (1977) similarly expresses the viewpoint that the basic ideas expressed by the Smith article were implicit in the writings and marketing practices of many early marketers."9 He states the opinion that the concept and practice of market segmen- tation might have become popularized long before the mid-19505, if it had not been for the lack of discretionary purchasing power during the depression, and the restraints upon consumer purchases caused by the consumer goods shortages that were characteristic of the market, during and after World War 11. Thus, the conceptualization and practice of market segmentation was postponed by marketers until 33 the transition had been completed from the seller's market, caused by the pent—up demand of the depression and World War II, to the buyer's market that began to develop in the 19505. Thus, the conclusion might be that the Smith article did not gag§g_the increasingly heterogeneous supply and demand schedules that followed its publication, but rather, pointed out to marketers a process that was well under way. Therefore, in 1956, market segmentation was a strategy whose time had come, and was pointed out at a time when it was increasingly characteristic or applicable to many product markets. Moreover, the basic postulate of market segmentation, that consumers are heterogeneous and thus exhibit heterogeneous demands, had been espoused on both a theoretical and empirical basis, within both business and social science literature, prior to Smith's catalytic article. Thus, the citations of pre-l956 contributors to this basic postulate are not suggested as being all inclusive, nor are they intended to take credit away from the early conceptualizers of the concept and practice of market segmentation, within the academic literature of marketing that was to follow and expand upon these initial foundations. Rather, it is to the credit of the relatively new but fast maturing discipline of marketing, that such a diverse foundation was to be perceived, integrated, and applied so forcefully, and so contradictory to the normative, homogeneous perception of the market, as had been typically postulated by marketing's academic ancestor, the field of economics. 34 The Application (and Misapplication) of Market Segmentation In the twenty-five years since Smith's classic article about the "alternative strategy"5° of market segmentation, Wind (1978) observes that market segmentation as a concept has become a dominant part of marketing literature and practice, and notes that "besides being one of the major ways of operationalizing the marketing concept, segmentation provides guidelines for a firm's marketing strategy and resource allocations among markets and products."51 However, Michman (1971) has concluded that although the concept of market segmentation has spawned a plethora of papers and complex studies, an assessment of these efforts suggests that "policies of market segmentation have not been as successful as imagined."52 Brogowicz (1977) similarly concludes that "the fact is that all too often the published results of segmentation studies have been either discouraging, inconclusive, or suggestive of shrewd ex post facto analysis."53 Echoing these conclusions, Arndt (1974) observes that while increasing numbers of increasingly high-powered statistical techniques have been applied to more and more "candidates" for predictor variables, "the search for identifiable, sufficiently strong correlates of purchase behavior has mainly turned out to be a wild goose chase."5“ Moreover, in his review of the state of segmentation research, Wind (1978) states that a discrepancy is revealed between academic developments and real world practice, with a substantial gap between academic-oriented research and the actual application to current 35 industry research and marketing practice.55 He also notes that many industry studies, designed as baseline studies, are simply ignored when firms undertake other marketing studies such as product or concept testing, since many firms are unable to perceive relevant results to marketing operationalization. In a similar conclusion, Mack (1968) notes that "when market segmentation knowledge exists at all, it is likely to be a good deal less than recent, a great deal less than accurate, and highly mis- trusted--even if trustworthy."56 Thus, he surmises that in such a corporate climate, segmentation may be an idea that everybody talks about, but nobody does anything about, relative to actual, segmented marketing efforts. In fact, the concept of market segmentation has not been universally accepted in marketing practice or academic marketing theory. In a sometimes quoted but typically ignored article, Reynolds (1965) concludes that much of what has been perceived as segmentation in the market may, in fact, simply be the result of a "variety strat- egy," with a heterogeneous array of products consumed by a typical American consumer, rather than different products consumed by separate, heterogeneous segments.57 Thus, while not disputing the idea that seg- ments may sometimes exist, he observes that even products quite differ- ent from each other may still be purchased by the same general group of consumers, and that this diversity of product may lead the marketer to falsely conclude that segmentation is present in the market. A salient conclusion from Reynolds' logic might be that the failure of certain methodologies to define distinct market segments for a specific product 36 may not mean a failure of the predictor variables tested, or the statistical technique used, but rather that true market segments g9 ngt exj§t_for that particular product! Thus, Wind notes that it is not uncommon to find markets in which no significant differences are found in the response elasticities to marketing variables,58 relative to the observed differences in con- sumer characteristics.59 Thus, doubt must be cast upon the premise that afly given market is heterogeneous in demand and can be effectively segmented for all product markets. Yet, numerous academic articles and studies heroically conclude that a particular predictor variable or statistical method was valueless for segmentation purposes, based upon observed consumer responses for a single product, or a product category so broad, and so generally con- sumed that distinct segments could not be expected (i.e., purchases of food or gasoline) beyond general purchasing Quantity differences. Moreover, such "learned conclusions" are often made on the basis of data samples that are dated, relatively small, and nonrepresentative of the consumer market,5° in studies that are typically not verified or repeated,“'compounded by the "lack of a systematic effort (by both academicians and practitioners) to build a cumulative body of substan- tive findings about consumer behavior"62 as it relates to market segmentation operationalization. As discussed in a recent article by Leone and Schultz (1980), studies that emphasize "new" products and new research techniques are often regarded more favorably in the academic world than those which 37 "merely" duplicate previous studies.63 Thus, for example, Boyd (1976) in a research editorial for the Journal of Marketing, complained that "too often manuscripts tend to replicate earlier studies, with but a small difference in either the research design or the product class involved,"5“ thereby presenting a clear message to marketing acade- micians, who wished a publication in this journal, ngt_to replicate earlier studies! Yet, as noted by Leone and Schultz: Replication is the key to generalization for without it, in the broadest sense, we have no corroboration of research results. We are left with one-shot studies that represent historical facts. Only by extending findings to other data sets do we perceive the generality of marketing relationships.65 Therefore, the gggg_segmentation studies are in need of replication, both for verification of validity for particular products, and also to determine if similar relationships can be established for other products, with additional data sets, and at different points in time, if the marketing literature is to provide an effective, "cumulative body of substantive findings"66 about segmentation relationships that may (or may ggt)57 exist in the real-world market. Moreover, the pgg[_ studies, unless re-examined, may continue to function as misleading guideposts, to academic researchers and marketing practitioners alike, thus blocking the route to more effective utilization of the market segmentation concept, or by continuing to point in less productive directions. Thus, Wind notes that to date (1978) he is aware of only a few unpublished industry segmentation studies in which the same market was 8 studied for segmentation relationships during a period of a few years.6 Researchers often assume not only that a limited sample was 38 representative and valid, but also that the relationships that were defined and measured would remain stable over time, despite limited evidence to support this assumption. Although Wind mentions, for example, that some stability in benefit segments over a two year period has been found,69 and that some behavioral segment stability had been documented over a three to four year period,70 a recent study by Calantone (1976) noted that in a study of benefit segmentation for banking services, that while the relative desirability of persons wishing certain benefits remained somewhat constant, the individuals seeking those benefits had changed.71 Thus, a marketer assuming that benefits which attracted customers would also maintain them, may be in for a rude surprise by ignoring the possible dynamics of segmentation relationships. In a follow-up study, Calantone and Sawyer (1978) observed that even when the existence of benefit segments remained constant, the importance of these benefits may change for selected targets, and that demographic characteristics of the benefit segments were likely to change, even over a few years.72 Therefore, media strategies, as well perhaps as message content, may require alterations to continue to effectively reach targeted segments. A logical, allegorical infer- ence might be that the use of segmentation study "roadmaps" that are seriously limited in detail, as well as obsolete for the current market, may well be a more significant factor in the operationalized ineffi- ciency of the segmentation concept than the effectiveness of the concept pgr_§g, assuming that a defined "map" even exi§t§_for a given product. 39 In a related vein, a recent study by Weisenberger (1977) observed the ongoing temptation for a multi-product, multi-brand marketer to seek cost savings in consumer segmentation research by seeking segmentation relationships for a family of products, via one major study, rather than conducting separate studies for separate products.73 However, Weisenberger's conclusion was that this type of general, "all-purpose" segmentation is subject to much error and interpretive difficulty, and he suggests that the more micro-approach of product specific segmentation studies "offer a more finely tuned picture of the consumer in the particular situation."7“ While mar- keters might continue to seek insights to general segments that may exist, in an allegorical sense, failure to find the "Northwest Passage" to universal consumer segmentation should not be seen as a failure of the segmentation research "expedition." Investigation of the existing but uncharted aspects of particular products may prove to be a more practical and useful route toward achieving eventual segmentation generalizations, and may be the only route to the practical utilization of segmentation theory in actual marketing management of those product categories for which segmentation relationships have not been effectively "mapped." This conclusion is echoed by Dhalla and Mahatoo (1976), who state: The poor performance of many segmentation criteria tested so far can be attributed to the fact that too often, researchers are anxious to find a magic formula that will profitably segment the market in all cases and under all circumstances. As with the medieval alchemist looking for the philosopher's stone, this search is bound to end in vain. There is no single algorithm that can be applied across all market studies.75 40 Similarly, Wind and Green (1972) suggest that it may be necessary to think of each segmentation study as an ad_hgg_study, having validity for the particular problem, product, and situation.76 Likewise, Plummer (1974) states that "so often, however, segments developed from a study on one product category have little or no relevance to another category.77 Thus, the failure to define "universal" relationships in market segmentation should not be regarded as a failure of the concept of market segmentation; nor should a failure of a particular set of segmenting variables be regarded as a universal failure in segmentation research, based upon limited value for a few product-market situations. As noted by Leone and Schultz (1980): Only by extending findings to other data sets do we per- ceive the generality of marketing relationships. . Basic answers to how marketing variables are related lead logically to new questions ofw __y_they are related. Like our fellow scientists, we must first describemarketing8 behav1or 1n order to bring us closer to explaining it. Thus, they argue that "replication is the key to generalization, for without it, in the broadest sense, we have no corroboration of research results,"79 thus we are limiting research to "one-shot" studies, rather than working toward a generalized body of marketing knowledge. The Dilemma of Normative Theory Versus Descriptive Explanation However, perhaps the most significant factor relative to the observed "efficiencies" of various segmentation techniques may be that unrealistic, normative criteria, for what an effective market segment "should" be, have gone beyond practical description of the actual 41 behavior of a normal (i.e., average or typical) market segment in the current United States consumer market. Thus, the concept of market segmentation may have fallen into a semantic and/or philosophical- theoretical trap, similar to that of micro-economic theory. For example, the concept of the "perfect" market, as developed by Marshall (1890), suggests a market of homogeneous supply and demand, for purposes of explaining market behavior with all other factors held constant.8° Yet this simplified "ideal" market is interpreted by some economists and social critics as a normative standard of what the market shgglg_become; thus, many economists object to the process of market segmentation as causing the market to become more "imperfect,"' leading to undesirable social consequences.81 Thus, this normative standard has become an ideal, or sought-after goal for some, despite the fact that the "perfect" market concept is far from descriptive of the actual market. In particular, the Marshallian model assumes away customer differences, and ignores the fundamental question of how product and brand preferences are formed in an imperfect, real-world market.82 In a similar, perhaps over-reactive fervor for the "ideal" usage of the segmentation concept, marketers may have assumed away much of the consumer similarity_that exists for a typical consumer market. For example, Mayer (1963) notes the impressive uniformity of American spending patterns, and concludes: "By and large everybody . wants to buy the same things that everyone else buys. Americans exhibit a remarkable homogeneity of tastes, attitudes, and buying habits, regardless of occupation."83 Moreover, it might be concluded 42 that mass media of the modern market, including television portrayal of the "typical" family home, etc., has done much to homogenize the tastes of the various geographic and ethnic segments of the United States, rather than create differences in consumer tastes, particularly in comparison to cultures that do not have access to such mass media of selling and communications. Thus, Reynolds (1965) concludes that there are obvious, true examples of market segmentation, but that most markets are character- ized by consumer preferences for variety, rather than exhibiting neat preferences for particular brands and products.°“ The purpose of this discussion, however, is not to dispute the value of the concept of market segmentation peg gg, but rather to suggest that the idealized, normative standards of what "perfect" segmentation shgglgibg_is not a realistic, descriptive standard, relative to practical observation and judgment of segmentation efficiency or utility in a typical consumer market. As concluded by Wind (1978), the normative segmentation models have rarely, if ever, been implemented, beyond limited academic studies, due to the difficulties in operationalizing such standards in the market.85 Tolleffson and Parker (1978) have determined that the normative prescriptions for segmentation have been advanced and used in research as though ideal measurement and conceptual standards, and have been included without qualification, in statements about segmen- tation, relative to pricing, promotion, and other marketing activities.86 But when tested, they note that such assertions of "ideal" differences between segments are generally ngt_ideal, and they question if such 43 normative assertions can be demonstrated ideal in any setting. For example, they observe that similarity of response elasticities and response coefficients have been advanced in the literature for a number of years as ideal performance indicators of segmentation activity, but failed to demonstrate this in their study. In a similar vein, numerous presumptions, or postulated standards have been advanced as requirements for proper use of the segmentation concept, but these "ideals" may not be reflected in reality, and thus may not be fair standards for the practical application or measured efficiency of segmentation for a typical product market, with expected, realistic degrees of consumer segmentation heterogeneity, variability, agg.homogeneity. For example, in a current, marketing management text, Bell (1979) states that to meet requirements for meaningful segmentation, a segmented group mg§t_meet the following requirements: 1. A segment must be specifica11y_identified and measured. A segment must be reasonably stable over time. A segment must react uniquely to marketing efforts. #WN A segment must evidence adequate potential. 5. A segment must be economically accessible.87 In this same text, Bell declares that "all_markets are composed of segments,"88 and that these segments are usually composed of further sub-segments--a presumption that has not been universally accepted 89 (i.e., Reynolds) or effectively proven. In particular, mere observance of consumer categorical differences, such as age, or 44 consumer brand preference for a single purchase decision, may not constitute a true segment. Moreover, Bell, as well as other academicians, seems to have adopted segmentation standards for "meaningful segmentation" in a logic similar to the micro-economic standards to define a "perfect market." If just ggg_of the normative presumptions of the perfect market is absent, that market is defined as being "imperfect" in micro-economic theory. Similarly, Bell declares that to be a "meaningful" segment, that segment must satisfy each of the five requirements cited above,9° with the parallel, implicit logic that a segment that does not, is not meaningful. Unfortunately, double entendres of semantic meaning of terms such as "imperfect" or "not meaningful" may give the broader semantic interpretation of being defective, undesirable, or not useful to meet the needs of the firm and the consumer. Thus, perhaps a more practical, liberal, and descriptive classification is needed to evaluate the desirability of particular segmentation techniques, such as "useful" versus "not useful," relative to the applied, rather than theoretical, segmentation goals of the marketing practitioner. As noted by Wind, there is a critical need for reexamination of the nature and value of normative segmentation, relative to understanding management's "use of and difficulties in operationalizing segmentation."91 Moreover, in a review of segmentation literature, no published segmentation study was found or mentioned that actually considered and tested for performance in each of the segmentation criteria, as set by Bell,92 let alone for the range of surrogate measures for these criteria that have been proposed and singularly tested in academic segmentation 45 studies. In fact, some of the criteria are rarely tested at all; for example, it has previously been discussed in this chapter that few segmentation researchers have attempted to validate the stability of a segment over time,93 despite this being a stated requirement for meaningful segmentationlg“ In a sense, strict standards and requirements are like strict laws; if they are not perceived, applied, or enforced, they cease to have real significance, and thus, new, practical standards may be needed, to provide actionable parameters for the application of market segmentation. The Suboptimization of Segmentation Efficiencies Another significant weakness in the operationalization of the marketing segmentation concept, both by marketing practitioners and academic writers, has been the tendency to suboptimize the segmentation effort, both with regard to overall, systemic efficiency of the firm, and with regard to the tgtal_process of marketing to a targeted segment or segments in the market. Mahajan and Jain (1978) have concluded that most current approaches to normative segmentation have involved two separate steps, the development of segments, and then the allocation of resources to 95 reach these defined segments. They suggest that the result may be nonfeasible segmentation schemes, suboptimal definition of the market, and probable inefficient utilization of the firm's marketing resources. It is suggested that segmentation programs should be aware of managerial, institutional, and environmental constraints, as well as resource constraints of the firm and the segmented consumer. 46 Similarly, Winter (1979) points out that segmentation per se_ should not be the only goal of the segmentation process, but rather, the firm should be aware of meeting company goals, such as profit, rather than merely maximizing sales to a particular segment.96 He suggests that even if it is assumed that revenues would be very high from a company policy of providing a separate marketing mix for each segment, the costs of doing so may be prohibitive. Thus, the disag- gregative process may need to be modified substantially, to become more aggregative, relative to cost-benefit criteria in meeting the firm's profit goals. A similar warning was stated earlier by Mack (1968), when he concluded that an extensive market segmentation policy by a marketing firm was likely to substantially increase the cost of market research efforts, as well as the investment base and direct marketing cost increases likely to follow a strategy of multi-segment marketing, which should be weighed against the value of such a strategy to the firm.97 Likewise, Mossman, Crissy, and Fischer (1973) argue that "profit contribution is the relevant measure of a segment's per- formance."98 They observe that the American Accounting Association committee on marketing costs has suggested a format for analyzing segmental profit contribution,99 that emphasizes cost attachment to each segment, along with cost control and behavior. In this article and a later book (1978),‘“’Mossman, Crissy and Fisher recommend that a "return on assets committed" should be 47 used to measure and evaluate the contribution of a market segment relative to the market value of the company assets that can be attributed to marketing expenditures and investments that directly relate to that specific market segment. They suggest that detailed, disaggregated analysis of each segment's actual performance is imperative to measure the success or efficiency of segmented marketing efforts. Yet, they conclude that past attempts to measure the success of segmenting strategies has taken the form of special statistical studies, imposed over aggregated accounting data, which can be expected to provide broad indicators at best, let alone to provide the continuing, integrated information system that is needed for marketing decisions. Even if a consumer segment can be sufficiently defined, relative to a predictable difference in demand, unless the marketing practitioner can achieve a reasonably accurate measurement of the profitability of this segmentation effort, both for the specific segment purchases, and for the profits of the tgtal firm, the true value of segmentation may prove to be illusory. Thus, segmentation might be perceived as a suboptimizing factor within the "marketing concept" philosophy of marketing management, to the degree that sales may_be increased, but the net effect on profits is generally unknown for most firms, and therefore, increased segmentation may well decrease profits in many instances. Thus, the principle of market orchestration, as explained by Kotler,”1 warns that a firm which employs a multi-segmented strategy must take care that the segments it chooses to market toward are reasonably compatible. The danger must be observed that marketing 48 toward an additional segment, beyond the incremental costs of the effort pg: sg, may cause a net drop in company sales and profits. As an extreme example, even though a definite market for X-rated cartoons is discernable, such a product line extension by Walt Disney studios might be predicted to have catastrophic results on the marketing success of the G-rated movies and cartoons that are the core of Walt Disney marketing success. Even though a profitable market segment might be served with X-rated Bambi features, the ngt_profitability for the total firm might be predicted to decline. The effect of violating this principle of market orchestration was dramatically demonstrated in an industry experience of this author's. A Chicago-area grocery chain, following good marketing strategy and practice, had observed an attractive concentration of older, high-income consumers who were not presently buying on a frequent basis at any of the company's other area stores. After conducting an extensive customer survey among this group, the company attempted to "localize" a new store to specifically appeal to this segment of con- sumers, with store layout, product line, and services oriented to the stated desires and complaints of this target group. A special consult- ant was hired to help develop special circulars for distribution only in the high-rise apartments where these potential customers lived, emphasizing the gourmet foods, special services, and other advantages of shopping at this new store. Engraved invitations to a special open house and tea were sent gnly_to these people, to introduce the store and the employees before the new store opened to the rest of the public. 49 However, in examining the demographic makeup of the area, large concentrations of orientals were noted, so an extensive line of oriental foods were added. Southern ethnic foods were added to appeal to consumers from the south and Appalachia. Then, to make sure the store had enough business, another circular was blanketed for miles around the new store, emphasizing giveaways of cheap bread, jelly, and similar items for several days after the opening. Special prices were offered, lower than other stores operated by the chain, on items such as round steak, private label ice cream, and bananas. The result was total chaos, with more than 25,000 customers the first week, and consistent totals of 20,000 or more customers attracted in following weeks by a continuing series of opening specials. But despite the large numbers of people, the store was not able to achieve sales totals high enough to make a profit, since the store was so crowded by bargain-hunters, people were not able to shop. An innovation of roof-top parking, with escalator service to the store was totally overloaded by the volume of customers, and literally became a barrier to entry, in the minds of many older customers. In particular, the attempt to reach many other people, as well as the high-income segment that was the core segment targeted by the company, resulted in the attractive customers being literally crowded out, frightened away, or offended by the poor service and hordes of bargain-hunting customers, many of whom were merely cannabalized from other nearby stores that the company operated in the area. Moreover, as the store floundered, special promotions were continued, to try to boost sales, which 50 continued to attract volumes of customers, but did not result in profitable sales volume. Thus, the principle of segmentation did not fail, as this store did, but rather, the company tried to appeal to too many segments at once, thus destroying the appeal to the segment that may well have made this store very profitable, and therefore resulting in a suboptimized result, as measured by the resultant store profit deficits, and the store's ultimate closing. A similar warning against too much segmentation was made by Mack (1968) when he cautioned that additional segmentation efforts may carry some risks of cannabalism, whereby the sales of a product osten— sibly aimed at only one segment, may also encroach upon the sales of another intra-company product offering, with possibly disastrous results 2 A classic example of this upon total company sales and profits.10 would be the introduction of Alka-Seltzer Plus by the marketers of the already successful Alka-Seltzer. The new "Plus" product was perceived by many loyal Alka-Seltzer users as being a new form of the existing product, with added or improved features. As a result, most of the sales of the new product came at the expense of the older, regular Alka-Seltzer, resulting in limited additional sales to the company, and the danger of legal liabilities from consumers using the new "Plus" for inappropriate reasons, as well as probable confusion in the mind of consumers about the benefits or values of using "Alka-Seltzer." Trout and Ries (1972) refer to this type of additional product offering as the "line extension trap,"‘°3 but perhaps a more appropriate term would be the "additional segmentation trap," caused, in the ultimate sense, 51 by inappropriate definition of market segment parameters, due to a lack of mutual exclusivity between "defined" market segments. Another type of suboptimization within the concept of market segmentation might be perceived in much of the academic research and literature. By the nature of the resources available, most academic studies have tended to focus on single aspects of segmentation, while often failing to explicitly study or analyze the integrated needs of the total segmentation concept. Ideal efficiencies have often been sought and sometimes proclaimed with regard to single components of segmentation, even though increased "efficiency" within one aspect, such as measurement or identification of segmented groups of consumers in a research study, while at the same time reducing the practical ability to identify or reach segments in the actual marketplace. Thus, an increased ability to predict segmented behavior may be something of a moot point, or even a negative practical contribution to the application of the segmentation concept, if the "defined" segment cannot be identified or accessed with a reasonable degree of success. Similarly, Mahajan and Jain have noted that a segmen- tation model should develop schemes of segmentation simultaneously with constraints of managerial, institutional, environmental, and resource constraints of the market.'m' Wind similarly reinforces this idea by stating that academic work on segmentation should be examined relative to management's information needs, to incorporate the separate developments in segmentation theory, to narrow the gap that exists between academic segmentation research and the practical application by marketing managers in the real world.”5 52 Thus, by focusing too sharply on individual aspects of segmentation theory, the result may be impractical definition of the market, infeasible segmentation, and suboptimization of the operationalization of the segmentation concept. Similarly, seg- mentation techniques that have particular weaknesses in individual situations or aspects of segmentation are often rejected as "ineffi- cient" segmentation methods, without consideration of the tgtal_value to segmentation theory and practice; this tendency seems especially apparent in many academic researchers' evaluation of the usefulness of demographic consumer variables in segmentation applications. The specific value of alternative methods of delimiting market segments, as reported and analyzed in previous market segmentation literature, will be examined in more detail in the next section of this chapter. In particular, attention will be focussed upon the use of demographic variables as segmentation criteria, since demographics have been the most frequent”6 albeit controversial, form of market segmentation, and also because demographics will serve as the heart of the empirical research of this dissertation. Alternative Methods of Market Segmentation As a general rule, market segments have been defined and researched on the basis of two basic types of descriptor variables: 1. General consumer variables, which seek to classify the market consumer by descriptive characteristics such as demographics, personality, life style, and social class. 2. Product-consumption variables, which seek to classify consumers on a basis related to consumption and/or purchasing in the 53 market. Such variables include the measurement of usage frequency, product benefits, brand loyalty, marketing mix factor sensitivity, or physical characteristics of the market. Plummer (1974) refers to these two general classifications as the "people-oriented approach" versus the "product-oriented approach,"‘°7 which will be the terminology used henceforth, in this discussion, to refer to these two basic segmentation alternatives. However, as might be expected, a continuing, but basically inconclusive, debate has ensued within the academic literature, as to which method of segmentation should be regarded as the "best" or "worst" general or specific method of segmentation. As concluded by Brogowicz (1977), "there is no simple answer to this question,"108 since no one method is likely to be the most effective in all situations. Thus, as noted early in the development of market segmentation by Yankelovich (1964), "the marketing director should never assume in advance that any one method of segmentation is the best."’°9 He suggests that the marketer's first step should be to consider all probable segmentation methods, then choose the most meaningful alternatives to work with. Yankelovich observes that this approach is similar to the logic for research in the physical sciences, where the researcher chooses the hypothesis that seems to best explain the particular phenomena under investigation."° Wind (1978) similarly presents the conclusion that: In contrast to the theory of segmentation that implies that there is a single best way of segmenting a market, the range and variety of marketing decisions suggest that any attempt to use a single basis for segmentation (such as psychographic, 54 brand preference, or product usage) for all marketing decisions may result in incorrect marketing decisions as well as a waste of resources.111 Moreover, Wind concludes that "surprisingly, despite thousands of academic and commercial studies by marketing and consumer researchers, one can draw very few generalizations as to which variables would have what effect under what conditions."112 Young, Ott, and Feigin (1978) deduce that segmentation studies have often been disappointing, because study results are simply not actionable from a marketing standpoint, due to researchers' common preoccupation with the techniques and methods of their favorite methodological approach per se, rather than developing a study method that is relevant to the competitive environment of a specific market.113 Thus, even if "results" are obtained, they are often not useful for practical, applied market segmentation programs. They conclude that: no single approach can be used to segment all markets as the specific competitive structure and environment determine the appropriate approach. Thus, the marketing researcher must analyze the market carefully to insure that only approaches relevant from a marketing standpoint are used. In many cases, markets should not be segmented at all as appeal to all segments is required for marketing success. . . .'1“ Arndt (1974) concludes that in determining the usefulness of a particular method of segmenting markets, the researcher must take care to examine the identifiability and the accessibility of the segmentation variables, as well as their predictive efficacy.115 While predictive ability is considered an analytical consideration, the identifiability and accessibility of the segmentation variables employed may significantly affect the operational value of such a 55 method. In an admittedly subjective and informal evaluation of the usefulness of some of the common types of segmentation variables, he ranks these methods as follows, in Figure l.116 Predictive Predictor Group Efficacy Identifiability Accessibility Demographic and socio- economic variables Low High High Social variables Low to Medium Medium Low to Medium Personality Low to Medium Medium Low Activities, interest, and opinion variables Low to Medium Medium Low to Medium Consumer learning indicators Medium Medium Medium Product perception variables High Medium Low Figure 1. Evaluation of Usefulness of Predictor Groups in Segmentation Analysis. Thus, product-oriented variables, such as product perceptions, may rank high in predictive efficacy, but rank very poor in accessi- bility, thus making this method a difficult or almost impossible segmentation strategy to operationalize beyond limited, academic research studies. However, demographic variables may rank low in observed predictive efficacy, but be the most efficient method of identifying and accessing the market segments that are targeted by 56 demographic predictors, due to operational advantages to the marketing practitioner. Thus, the t9tal_value of a method should be considered, rather than a suboptimized focus on only one aspect, such as ggly_ predictive ability, without proper consideration of the operational aspects upon the overall ability or value of a specific segmentation method. Therefore, a product-oriented method may provide better predictive segmenting abilities, but the operational aspects, such as accessibility, may be so low as to provide an absolute veto for its practical application, and thus may be a moot alternative for most segmentation purposes by a marketing practitioner. For example, the heavy-half theory, or volume segmentation, as popularized by Twedt (1964), suggests that one-half or less of the total market commonly accounts for 80 percent or more of the consump- tion in most product categories.117 The principal problem with this "technique" of segmentation, as pointed out by Haley (1968), is that not all heavy users are typically attracted to the same brand,118 thus resulting in a very heterogeneous volume "segment." Moreover, the heavy users of one product are not necessarily the users of another. Thus, as Twedt himself notes, "the heavy user is not readily identi- fiable in terms of characteristics other than being a heavy user,"119 thus defining the volume segment by the tautological basis of being a heavy, volume user of a specific product. Thus, volume segmentation may not be an operational method of segmentation. Moreover, other segmentation methods, such as benefit seg- mentation, or other aspects of product-oriented segmentation, often employ consumer characteristics, such as demographics, as "descriptive" 57 variables to identify and access the purchasers of specific product segments. Thus, Haley (1968) suggests that demographics be used as "secondary" variables to identify those who have been found to purchase specific product benefits.'2° Another interpretation of Haley's logic might be to perceive his "benefit segments" merely as more accurate definitions of the relevant product market, thus allowing demographics or other consumer-oriented criteria to define and predict the users of certain categories of products in a more effective way than for a broad product class such as toothpaste. Thus, no method should be expected to "segment" the market for food or health and beauty aids into reason- ably homogeneous segments; failure to do so should not be used to indi-‘ cate the "failure" or inefficiency of the tested, predictive segmenting criteria, but rather, indicates that segmentation is not very opera- tional for such a broad grouping of products and consumers. However, "sex-appeal" or "cavity prevention for kids" may present a more logical basis for partitioning the market into realistic, defined sub-sets of brand alternatives, such that demographic consumer segments might then be defined, predicted, and accessed in the market. Thus, the predictive problems of consumer characteristics, such as demographics, may not be due to their inherent lack of pre- dictive, segmenting ability pgy_§g, but rather the failure of many marketers to properly define or measure a relevant market. Relative failure to predict the purchasers of a cavity-fighting toothpaste such as Crest, may not have meant that demographics were poor segmentation predictors, but rather that the marketing researcher may have myop- ically defined the market for toothpaste as being merely "toothpaste," 58 thus exhibiting "marketing myopia"121 in defining the relevant indicants of purchasing behavior. Similar questions about the validity of demographic-based segmentation, or any segmentation method for that matter, might also be raised via the issues of data validity, data age, and the lack of verification over time,122 relevance to particular products that have had limited or no significant testing of such relationships, statis- tical methodology or research design,123 as well as questions about the interpretation of prior demographic segmentation studies, both in terms of predictive efficacy and also the relevance to total, practical application of a segmentation strategy in the actual market.'2“ These issues will be examined, as they relate to specific, exemplitive, demographic segmentation studies that have been reported in the marketing literature relevant to the value of demographic consumer variables as segmentation criteria. A Review and Analysis of Some Exemplitive Demographic Segmentation Studies in the Marketing Literature One of the most influential and commonly cited studies of the value of demographic variables in segmentation has been the 1967 study by Frank, Massy, and Boyd.125 In an often quoted conclusion, this study reports that "socioeconomic and demographic characteristics are poor predictors for a wide range of specific grocery products."126 However, the study continues, that "49 regressions were sta- tistically significant [but] . . . not as useful . . . for these products as had been hoped." The demographic variables were reported 59 as explaining half or more of the consumption variation in 46 of the 57 product categories studied.127 Next, it can be observed that the product categories tested in this study included such universally consumed products as toilet paper, catsup, laundry detergent, and toilet soap--all products that would be much more noticeable by their absence than by their presence in the typical shopping basket. It is similarly not surprising that family size and the age of the youngest child were the two variables most highly correlated with such products. Thus, an interpretation of these results might be that the results were merely providing a consumption function for products typically consumed by most households, rather than determining true segmentation for such products. The demo- graphic variables used should not be regarded as defective for failing to define segments that were not likely to be there. Moreover, the relatively close substitutibility among many of the categories, such as canned corn, pork and beans, hot cereal, and cold cereals, may well show a product-to-product consumption relationship due to a general substitution and/or variety effect, rather than establishing homo- geneous segments that could be identified and accessed in the marketplace. Another major concern relative to this study is the age of the data (1960-61) and the comparatively small size of the sample (491 families) for the imposition of a multi-variate regression involving a total of 14 demographic variables, including some, such as occupation, religion, and race, that are nominal scale variables, which are not easily measured by regression analysis. 60 Another major concern, relative to the validity of the conclusion of this study has been expressed, and researched by Bass, Tigert, and Lonsdale (1968)128 and by Wheatley, Chiu, and Stevens (1980).129 Both of these studies assert that regression analysis may not give adequate measurement of the true segmentation relationships that may exist within some of the product categories. In particular, both studies suggest that gyggp_behavior, rather than individual household behavior, may be a more relevant and predictive basis for assessing the strength of demographic predictive relationships. Thus, Bass, Tigert, and Lonsdale state: "For market segmentation, the essential question is whether it is possible to identify groups of consumers with different mean purchase rates dependent on certain variables such as income, age, and occupation."'3° Thus, even though the within-group variance, as observed in multiple regression, may be large, both studies argue that the relevant choice-differences are between groups, not individuals, and effective segmentation can be performed based upon gyggg_means. Based upon a data base of 6,264 responses, Bass, Tigert, and Lonsdale demonstrate that a regression analysis of ten grocery and health and beauty aid products resulted in a uniformly low R2 pre- dictive value, consistent with similar studies, such as Frank, Massy, and Boyd,131 but their conclusion is that this merely indicates that the variances within regression cells is great, ggt_that the relation- 2 Thus, an analysis of cross- ships are weak or valueless per se.13 classification tables was reported to effectively discriminate between patterns of group behavior. For example, the probability that a college 61 graduate will buy five or more cans of frozen orange juice is noted as twice the likelihood that the household head will be a high school graduate.’33 This study also indicates that the consumption of beer is a non-linear variable, a factor which is likely to lower or even destroy the ability of linear regression analysis to predict consumption based upon demographics. When the analysis of beer is extended to two or more independent, demographic variables, the difference in conditional means and probabilities, and thus the ability to segment, is further increased.lm' Thus, Bass, Tigert, and Lonsdale argue that the yery strength of demographic group means of consumption may reduce the reported strength of segmentation in a method such as linear regression analysis. Similarly, Wheatley, Chiu, and Stevens report the study of demographic segmentation criteria for the consumption of coffee and soft drinks, based upon a sample of 252 respondents. When tested via standard, linear multiple regression, as used by Bass, Tigert, and Lonsdale, they reported an R2 of 14 percent for coffee, and about 20 percent for soft drinks, in this 1980 study.135 However, when measured on a non-linear, group basis, they reported a total R2 explanation of .846 for coffee, and an even more impressive .984 R2 for soft-drink consumption, with the primary factor being the age of the consumer.136 Thus, they suggest that target markets for some products may effectively be segmented by means of readily available demographic information about the consumer. Therefore, they conclude that the 62 usefulness of demographic variables as a basis for segmentation should be reconsidered, due to the demonstrated efficacy shown via their study, and that of Bass, Tigert, and Lonsdale. Additionally, Wheatley, Chiu, and Stevens observe that the formation of segmentation strategies and the development of target- audience profiles within the media has been a long standing practice among marketing practitioners. Thus while not seeking to totally discredit earlier studies, which suggested that demographics were poor segmentation criteria, they suggest that this issue seems far from conclusive. They conclude that "the abandonment of the more accessible, less ambiguous, and less expensively obtained demographic and socioeconomic data seems premature,"137 and suggest that other researchers seek to replicate the group approach with other products and with larger samples. Another pioneering and often cited study about the value of demographic segmentation variables was the 1959 study by Evans, regarding possible differences between the purchasers of Ford versus Chevrolet automobiles.138 It is reported by Michman that this study emphatically contended that demographics were not of practical use in market segmentation strategies,139 but Arndt points out that this study a1§g_showed that the objective, demographic variables yeye_better pre- dictors than the personality-variables that were studied as a compara- tive technique for segmentation.1“° Thus, there seems to be a semantic confusion, between reporting that demographic variables were poor pre- dictors, because they did not provide eggggh_prediction in a given study and research design, versus perceiving that demographics were 63 better than other tested alternatives. This logic seems somewhat akin to complaining that a cup is half-empty versus regarding the benefits of a cup that is half-full; the value of a particular study may be perceived for what it lacks from the ideal of total, desired explanation, or it may be recognized for the contribution relative to practical, alternative methods, and prior knowledge. Moreover, an actual finding of significant demographic dis- tinctions between Ford and Chevrolet owners, if_it pag_been documented, could have been regarded as a failure of Ford and Chevrolet marketing strategies, since the general basis of both companies' strategies has been to aim at the same segments of the auto market, in price, features, and other basic appeals. Perhaps a more distinctive, demographic seg— mentation could have been defined for Ford versus Cadillac, for example. In addition, the variety in the product lines for Ford and Chevrolet are the broadest in the auto market, as they both seek to apply a differentiated strategy of offering different products to separate segments of the market. Thus, expecting the demographic profile of a Chevrolet pickup to effectively predict the sales of a Corvette, beyond very limited accuracy, would seem to be very unlikely. These particular models are aimed at identifiable segments of consumers versus comparable brand and model alternatives, in reasonably distinct niches. Within the very broad and expectedly heterogeneous consumer market, supplied by an observed, differentiated and heterogeneous product line for both Ford and Chevrolet, it seems necessary to identify reasonably homogeneous product market categories, as suggested by 64 1 concept of benefit segmentation, before seeking to Haley's'“ accurately identify a reasonably homogeneous consumer segment. Thus, the limited ability to predict the purchase of a Ford versus a Chevrolet automobile might be interpreted not as a failure of demographics, but rather because the measure of segmentation pur- chasing behavior was too heterogeneous and indistinct to correlate with the demand patterns for possible demographic segments. Moreover, the relevant measurement of demand within a segment may not be a brand per §e, but rather a set of brand and model alternatives that satisfy the needs of that segment, distinct from the alternative-set of another distinct segment. Thus, a common complaint against demographic segmentation in particular, and against segmentation efficiency in general, is that the segmented groups are too broad or indistinct to accurately distinguish between the users of various brands.”2 However, as Dhalla and Mahatoo (1976) observe, instead of being confined to a single brand, segmenta- tion behavior may extend to a group of brands in the segment's "evoked set."“'3 This would seem to be especially true if several brands were actively marketing to the same consumer segment. Therefore, rather than demographics "failing" to measure the purchase of a particular brand, another interpretation might be that purchase of a particular brand is likely to be a limited measurement of a segment's purchasing behavior in a real-world, competitive market. Thus, an indistinct measurement of segment response cannot be expected to correlate to a higher degree with a demographic definition of segment membership; if the measurement of the response is ambiguous or 65 inaccurate, the predictive power of the segmenting method, such as demographics, will be "measured" inaccurately, as well. A recent example of the problem of measurement, caused by ambiguity within the factors of demand response, is the study presented by Assael and Cannon (1979), in which demographics are studied for their ability to properly distinguish product usage and interest, and to match consumers to these "user" segments, via appropriate media selections.'““ The conclusion of this study is that demographics decreased the efficiency of the media-product match, when compared to the direct match of the media with consumers who are known to be members of the target group. However, as conceded by Assael and Cannon, direct matching may not be possible if data, on media readership and direct matching of product usage, are not available from the same source, which they typically are not. Nevertheless, the most important question about the derived conclusions of this study relate to the usage and interest factors that were utilized to determine such efficiencies. Five criteria were used to indicate the dependent response of consumer market segments, including: 1. Purchased furniture in the past two years; Purchased a color TV in the past two years; Interested in attending plays and musicals; Interested in attending movies; and 01-500“) Interested in gardening.”5 66 Thus, the categories of user purchases and interests might be perceived as unlikely "measurement" criteria, since they are so broad and expectedly heterogeneous in scope, that they tend to become almost meaningless as effective delimiters of segmentation behavior. For example, color TV purchase and ownership has become almost a universal response in American household purchasing behavior, and virtually every household is likely to buy some furniture, either new or used, on a semi-regular basis. Thus, the very breadth of expected purchasing interest and behavior in these "defined" groups is more likely to be distinguished by observed similarity between demographic groups, rather than by distinguishable, significant differences. Seventeen demographic variables were utilized, to pick the most effective media choices to reach these target segments, utilizing media alternatives such as Forbes, Golf Digest, Fortune, Scientific American, Sports Illustrated, and Gourmet, a very puzzling set of potential media to reach the listed usage and interest targets. Moreover, it is highly unlikely that an advertiser, wishing to reach the defined usage and interest targets, would even have considered most of the media alternatives that were tested, even to reach "elite" parts of the target groups. For example, upper income furniture buyers or gardeners might be much more effectively reached in magazines like Better Homes and Gardens than Business Week or Smithsonian, let alone Sports Illustrated. It has long been noted that younger consumers form the heart of the movie-goer market, at least in-theater, yet there are no media tested that could be expected to effectively contact or interest young movie-goers. 67 As concluded by Arndt (1974), many segmentation studies seem to give the impression of "being no more than attempts to use data conveniently available, for instance in the form of consumer panels, and see what comes out of the computer.““ Data for this study were obtained via Starch Inra Hooper, Inc., which provided data about the interests of a defined "elite" consumer group, relative to subscription and readership of a similar "elite" set of print media,'“7 and surveyed pply_families with an income of $15,000 or more (in 1974), in which the household head was employed in a professional or managerial capacity. Thus, the survey data cannot be expected to reflect the consuming habits of the normal population, or for the expected consumer demographics of these "product and interest categories." In all probability the data was used, because it was the pply available source of such direct- matching data, otherwise not available to the researcher or market practitioner. Thus, it might justifiably be perceived that Assael and Cannon were "successful" in reaching a ppprrepresentative group of consumers, with direct-matching techniques based upon data that is rarely available to marketing practitioners, via media choices that typically would ppt_ even be considered relevant alternatives to reach the "defined" user and interest segments. Nevertheless, based upon the questionable consumer sample, media alternatives tested, and the ill-defined behavior segments, Assael and Cannon boldly conclude that it is "clear" that "the use of mediating (demographic) variables substantially reduces the efficiency of media selection."‘"8 68 However, the detailed analysis and critique of this study, and the others that have been noted in this discussion of exemplitive segmentation studies, has not had the primary purpose of lauding or criticizing the results of these particular studies peyy§e, Rather, beyond the literal results or validity of these specific studies, or any other segmentation study, care must be exercised in the consid- eration and interpretation of their purported, "proven" conclusions. Significant questions often exist about validity of prior study sampling, measurement criteria, statistical techniques, and inter- pretive logic, in addition to continuing issues of the age of prior studies, and the general relevance of findings for one type of product to another type of product or segmentation classification. Thus, "caveat emptor" seems a valid strategy for consideration of the conclusions that are cited in marketing literature, based upon prior, and perhaps questionable, segmentation research. Segmentation Analysis of the Frozen Food Consumer--Prior Studies Despite six reported studies of the frozen food consumer, the status of existing, operational knowledge about segmentation of the frozen food market remains surprisingly limited. First of all, the results of most of the prior studies have been limited by the research limitation of considering the frozen food market as one basic product category. Yet, a 1975 survey of supermarkets showed an average of 816 frozen items, with the total 9 number of items expected to show continued increases.‘“ Moreover, significant evidence exists that segments for the purchase and 69 consumption of frozen food products are likely to vary substantially between separate product categories. Thus, "frozen food" may not prove to be a relevant measurement of purchasing behavior, relative to the definition of market segments, or the measurement of segmentation efficiency. Secondly, among the published, available studies of the frozen food consumer, there seems to be virtually no reported statistical analysis of consumer data. Most conclusions seem to have been simply drawn from general perusal of two-way cross-tabulations or simple averages of survey responses, with no reported statistical testing of the significance or predictive ability that may exist within these tabulated survey responses. Beyond some broad generalizations about the possible meaning of survey responses, data are typically reported in raw, cross-tabulated tables, from which the reader is apparently expected to infer what he can. As noted by Arndt about so many seg- mentation studies, "the uncomfortable questions of relevance, and "so what" keep arising,150 questions which seem commensurate with the prior studies of the frozen food consumer. It seems especially frustrating to note the same basic questions asked again and again in the six reported studies, but with no significant conclusions derived, no recognition of prior study conclusions, and for the most part, no reported reference to the existence of prior studies. For example, all six studies specifically ask about the availability of freezer space, and/or the ownership of a separate freezer unit; yet, only the earliest 70 study, done in 1948, attempts to analyze the meaning of the survey response to this question. Third, the use of afly_multivariate analysis technique has been almost totally absent in these studies. With the exception of a few three-way cross-tabulation tables (again, without signif- icant analysis of any results), multivariate relationships simply have not been investigated in these studies for the purchases of "frozen foods" as a total, or for the separate categories of frozen categories within the total frozen food category of observed purchasing response. It may be noted that specific products, such as orange juice, have occasionally been examined as part of total food purchasing, but with limited reported results or conclusions relative to the specific frozen items, or for total purchases of frozen foods. Thus, reported and published studies of frozen food segmentation would seem to be described, at best, as generally descriptive, with virtually no reported statistical analysis of study results, and with very limited operational guidelines as to the effectiveness of any segmentation technique for the actual defi- nition or prediction of segmented purchasing patterns for frozen food products. Thus, beyond being a valid contribution to general market segmentation research, this study has the potential of being a landmark study of frozen food market segmentation. 71 Specific Discussion and Analysis of the Prior Studies The 1948 Scott Study The earliest known study of the frozen food consumer was conducted by Raymond Scott, in Syracuse, New York. A consumer survey of 1,089 persons was conducted during the summer of 1947, and the results were published in February 1948.151 Beyond the early analysis of the frozen food market, via demographic consumer characteristics, the study may also be noted in the history of the concept of market segmentation as taking place well before the classic segmentation article by Wendell Smith in 1956.152 From a research standpoint, the study is also noteworthy for the painstaking detail spent in collecting a valid, demographic sample that would be reasonably representative for characteristics such as race, income, and nationality of the survey respondents. A block sampling technique was used, with specific follow-up instructions to interviewers to reach homemakers who were not at home. Considering the age of the study, and the lack of tools, such as computer analysis, this study is amazingly well designed and analytic in its approach. Data results are presented in a series of two-way cross- tabulation tables with no statistical analysis of the results, but Scott attempts to analyze each table to a considerable degree, relative to observed differences in demographic consumption patterns for the consumption totals of all frozen foods, as well as consumer attitudes about the then relatively new category of frozen foods. Moreover, the effect of the 19305 depression, and the product shortages of World War 72 II, must be considered in terms of the effect on consumer incomes, family demographics, available product variety, and even the pent-up demand for electric refrigerators that was characteristic of this period. However, despite these constraints on the scope of the frozen food market, the study is also notable for the fact that it surveyed consumption of specific frozen food product groups, rather than merely a "total" frozen food consumption measure. Thus, in many ways, this study is the most complete, best designed, and most analytic of all the known studies of the frozen food consumer. In terms of specific conclusions, Scott noted that income, nationality or race, occupation, and age were shown to be the most important factors in frozen food consumption. In particular, income proved to be a two-edged factor, since lower-income groups were observed as the least likely to own or have access to proper storage facilities, as well as being, by far, the least likely to consume frozen items, both in percentage of such households using frozen items, and in the total amounts used. Per-capita purchases decreased as family size increased, and it was noted that working homemakers purchased tyjpe_the quantity of frozen items as a non-working homemaker-consumer.153 Another interesting observation was that 96 percent of the families whose chief wage earner was a professional or an executive had used frozen food products of some kind during the past two years, while only one-third of the unemployed and two-fifths of the laborers had used any frozen food items. Scott suggested that these figures 73 were likely to be influenced (intercorrelated) by the income levels associated with these occupational groups,'5“ as well as by the educational levels of such consumer groups. Convenience was listed as the most important attitude factor in using frozen foods, which became increasingly important in frequency of mention as income of the family increased, varying from 34 percent mention by low-income families, to 68 percent for upper-income families. Primary reasons for ppt_using frozen items were price, preference for fresh, lack of storage space, and home-canning.155 This last factor seems especially relevant, since home-canning was much more common in this era, but was reported less frequently among high-income families, 3 and more frequently among low-income families. Survey response may also have been affected by the fact that it was collected during June and August, when the greater availability of fresh items, home-canning, and lack of freezer space could all have been expected to reduce the incidence of frozen product usage, as compared to a winter survey's responses. While the age of the survey, the vast change in consumer buying patterns, and the extraordinary growth in the size and variety of the frozen food market, all make literal usage of the Scott conclusions rather tenuous for a modern-day frozen food marketer, the general impression remains that many of the observed patterns between demo- graphics and frozen food purchases pppld_remain amazingly consistent in explaining modern characteristics of frozen food purchasing between these same demographic groups.- In any case, the Scott study seemed to definitely establish that important differences did exist between the 74 demographic groups studied in his survey, relative to their observed purchases of frozen food products. The National Frozen Food Association Studies Another major and continuing sponsor of research, about the frozen food consumer-purchaser, has been the National Frozen Food Association, Inc., an industry trade association. This group has sponsored and published three major studies of the frozen food purchaser, in 1973, 1975, and 1981, with reportedly, another major study under consideration, thus providing evidence of the continuing interest by frozen food marketers to better understand consumer differences within the frozen food market. The first study, conducted in 1972 by B. J. LaLonde, was published in 1973, based upon 1,610 panel respondents furnished by National Family Opinion, Inc.156 This study was intended to act as a benchmark to provide information to the industry about the frozen food consumer, including the specific issue of the influence of consumer demographic variables upon consumer frozen food purchasing patterns. The study also had the stated objective of examining the attitudes of consumers about frozen foods, as well as the impact of demographic characteristics upon frozen food usage patterns. Although the study reports that survey data were analyzed "using standard statistical cross tabulation programs,"157 the spe- cific statistical methods used in the study were not disclosed, nor was any specific statistical test, measurement, or predictive value ever mentioned in the published study. The study goes on to report 75 that 600 tables were produced during the course of the research, with the most significant and interesting tables reproduced in the study report. However, most of the tables presented were merely distributional reports of specific question responses, such as the percentages of responses among the entire population, with regard to purchases of groceries, frozen food total, and ice cream, as well as raw data scores and distributions on attitude questions. The reasoning for inclusion of such tables was not given, nor was the relevance of them to the stated objectives ever really explained. In a section labeled "detailed findings,"158 cross tabulations on a percentage basis were reported between the purchase expenditure categories of the previously printed tables, relative to bivariate distribution among demographic factors such as age, income, education, and occupation. Similar bivariate tables were presented for the data responses to questions about frozen food attitudes. However, the simple distribution of consumption response that could be attributed to the raw, percentage distribution of demographic traits within the survey sample was generally not considered, nor were any statistical measures of predictive value ever presented in the study for any table. Moreover, there was no attempt demonstrated to correlate the attitudes about frozen foods to any pattern in frozen food purchases. The only measure of frozen food purchases that was reported in the study was total frozen food purchases. Even though the original survey questionnaire asked specific questions about the purchase and home inventory of a relatively long list of specific frozen food products, this information was not even key-punched, and was totally 76 ignored in the published study. Similarly, even though the study had included a question about home freezer ownership, no results were mentioned in the study results. Within the 32 published pages of the report, two pages explained the methodology and goals of the study, one page gave implied implica- tions for segmentation (that seemed to have no direct tie to the study data), and ppe_page was presented of very general conclusions. The rest of the study was simply page after page of tables. Moreover, the "conclusions" that were cited were not explained, substantiated, or measured via survey results, at least as reported in the study. LaLonde notes that the "summary report of the research points clearly in the direction of market segmentation,"159 but from the tables that were presented, and the broad, unexplained or unsub- stantiated conclusions, this study seems to add little usable knowledge about the demographic segmentation of the frozen food market. A follow-up study was conducted by LaLonde, Hansen, and Scott in 1975, resulting from 1,400 panel respondents, again furnished via National Family Opinion, Inc.‘6° For the most part, this study dupli- cated the questionnaire of the 1973 study and tends to apply the same basic "analytical" techniques as the original study. For the most part, the same cross-tabulation tables are presented, with the results from the 1975 survey listed beside the results of the survey collected in 1972. No statistical testing or measurement is reported in the study, either for the validity of demographic variables as segmentation variables, or to determine if the observed differences between the 77 two study's responses were statistically different or meaningful to the marketer. By simple perusal of the compared results, many of the observed differences could easily have been attributed to mere sampling variation, rather than indicating a meaningful change in segmentation response or attitudes. Moreover, the value to the marketer of noting the changes that were observed, even if valid, is very unclear, since most behavior patterns of purchase and attitude indicated substantial heterogeneity of response within every bivariate table presented. In such cases, mere averages of response may be very misleading, without statistical analysis of the variation. Moreover, while "multivariate analysis" was supposedly applied,‘ this apparently consisted of three-way cross-tabulation tables, with no reported statistical analysis, and with very limited, perceptual rather than empirical, conclusions as a result. Thus, the results and conclusions of both LaLonde studies can be characterized as broadly descriptive of general patterns that were observed, but not statistically tested or measured, at least as reported in the published study reports. Nevertheless, with this caveat in mind, the 1975 report did arrive at some interesting conclusions. Perhaps most important, the 1975 report concluded that frozen food usage is clearly product specific, based upon some testing of specific product purchasing that was surveyed, but otherwise not presented in the report. Thus, the measurement of tptpl_frozen food consumption seems to be less valid for the definition of frozen food market segments. 78 Next, education was noted as being positively associated with frozen food purchases, at least until the level of a doctor's degree. Similarly, income is noted as having a positive relationship relative to the purchase of frozen foods. Age was noted as having an important but curvilinear relationship to frozen food expenditures, having higher levels in the thirty to sixty year age brackets, with lower levels above and below this category; this age group was characterized as being the heavy user segment for frozen items.161 Thus, this study does develop some general, inferred relationships between demographics and frozen food consumption, but does not present statistical testing of the validity or the predictive efficiency of such relationships. Finally, the National Frozen Food Association also sponsored a 1981 study via a Better Homes and Gardens panel of 500 reader- respondents, for each of two surveys, combined in the 1981 report to association members.162 However, even though this panel data contained demographic data about the respondents, survey questions were not cross-tabulated to the demographic factors, and thus this study did not make significant contributions to demographic segmen- tation of the frozen food consumer. Here again, the "report" con- sisted mainly of percentage breakdowns to survey questions, with little attempted analysis of the meaning of such responses. However, even if such responses had been correlated to demographic character- istics the nature of the panel, as well as the questions asked, would have made the results of the study very suspect. 79 In particular, the panel is observed to have 46.1 percent of its respondents with an income of $25,000 or more, compared to 15.7 percent of the population in general. Moreover, the panel members were significantly older than the normal demographic distribution of the market, was much more highly educated, nearly always married, and owned their own homes in 91 percent of the cases.163 Thus, the panel may have been representative of Better Homes and Gardens' readers, but did not represent the total market for frozen food items. Next, the measured indicants of purchase response seemed very questionable in many cases. Again, the survey considered pply_the estimated total expenditures for all frozen foods, rather than testing ' for the purchases of specific products. In addition, the survey asked, "during an average week three years ago, how much did you spend on frozen foods,"16“ a question that is almost impossible to expect a typical consumer to accurately reply to. The survey also asked, "during an average week three years ago, how much did you spend for food," the response for which was dutifully tabulated, but not interpreted as meaning anything to this study's resulting conclusions. While some attitudes about frozen food consumption, roughly comparable and often identical to the questions on the 1972 and 1975 studies, were asked in this recent study, only raw distributions and summary scores were presented, with no significant analysis of results. No attempt was made to compare these results to studies of these attitudes in the prior National Frozen Food Association reports. No statistical testing of any kind was reported. Thus, the 1981 80 Better Homes and Gardens'study did not appear to make a significant contribution to knowledge about the frozen food consumer, or to the practice of market segmentation. The Akron Beacon Journal Studies Finally, the last major reported studies of the frozen food consumer have been the two Akron-area studies conducted by the Akron Beacon Journal, as part of its regular surveys of buying patterns in its publication area. The same basic study was done in 1971-72 and in 1977,165 reporting the consumption patterns for twelve categories of frozen food products, in relation to the variables of income, family size, and the age of the head of household. Ten other demographic categories, such as race, education, and occupation were collected as part of the survey data base, but were never reported in the published format of the studies. The results of these studies are simply presented in cross- tabulation format, with absolutely no analysis or conclusions of any kind given in the published report. Thus, these two reports are best considered as data to be analyzed, rather than information published about the segmentation factors of frozen food products. But, in par- ticular, one factor that is consistently observable in both reports is that for most products, a majority of the survey sample did not purchase ppy_of the product category being studied. In fact, for some products, such as frozen fruits, frozen pizza, and poultry products, 29 percent or less of the study group, in each of the two reports, accounted for almost all of the observed purchases in these product categories. 81 However, unless these "heavy-user segments" can be identified or classified from the total population, no segment can be targeted by the frozen food marketer of these products. Thus, the practicality of such segmentation remains somewhat illusory at the present knowledge level about such consumer segments. We find again, as noted by Twedt, that volume segments seem to exist, but if they can only be identified by the tautological fact of pejpg_heavy users, such users cannot be identified and targeted as a viable market segment.166 Thus, while the data base used in the Akron studies seems to be quite adequate with regard to survey collection methods, as well as the fact of responses collected about specific categories of frozen foods, rather than merely total frozen food purchases, the reports remain basically raw data tabulations, which have made no significant contribution at this point in time. The data base used in the 1976 Akron Beacon Journal study will be employed as the empirical foundation for the research of this dis- sertation. As explained in the following chapter, the 720 questionnaire responses from this study will be employed to test the ability of demo- graphic variables to segment the purchases of frozen food products, and thus operationalize the concept of market segmentation. FO0TNOTES-—CHAPTER II 1Wendell R. Smith, "Product Differentiation and Market Segmentation as Alternative Marketing Strategies," Journal of Marketing 21 (July 1956): 3-8. 21bid., p. 6. 3Kenneth Schwartz, “Fragmentation of the Mass Market,” Dun's Review and Modern Industry, July 1962. “Ronald E. Frank, William F. Massy, and Yoram Wind, Market Segmentation (Englewood Cliffs, N.J.: Prentice-Hall, Inc., 1972), p. 246. 5William G. Nickels, Marketing Principles (Englewood Cliffs, N.J.: Prentice-Hall, Inc., 1978), p. 75. 6Arch W. Shaw, Some Problems in Market Distribution (Cambridge: Harvard University Press, 1915), pp. 101-103. 7Johan Arndt, Market Segmentation: Theoretical and Empirical Dimensions (Bergen, Norway: Universitetsforlaget, 1974), p. 7. 8E. H. Chamberlin, Theory of Moggpolistic Competition (Cambridge: Harvard University Press, 1933). 9Joan Robinson, The Economics of Imperfect Competition (London: McMillan, 1933). 1°Joan Robinson, The Economics of Imperfect Competition (London: McMillan, 1948), pp. 179-188. uBo W. Schyberger, Theory and Methods of Market Segmentation. Research Report No. 64 (Stockholm: Department of Business Administra- tion, Stockholm University, 1971). 12Joel Dean, Managerial Economics (Englewood Cliffs, N.J.: Prentice-Hall, Inc., 1951). 13Wroe Alderson, Marketing Behavior and Executive Action (Homewood, 111.: Richard D. Irwin, Inc., 1957). 1"Ibidu PP. 232-233. 15Ibid. 82 83 16Ibid., p. 59. 17Thorstein Veblen, The Theory of the Leisure Class (New York: The MacMillan Co., 1899). 18William H. Shyte, Jr., "The Web of Word of Mouth," Fortune 50 (November 1954): 140. 19W. Lloyd Warner and Paul Lunt, The Social Life of a Modern Community (New Haven: Yale University Press, 1941), p. 88. 20W. Lloyd Warner, Marchia Meeker, and Kenneth Ellis, Social Class in America (Chicago: Science Research Associates, 1949). 21Pierre Martineau, "Social Classes and Spending Behavior," Journal of Marketigg 23 (October 1958): 212-230. 22Ibid. 23Stanley 0. Sibley, "The Identification of Behavioral, Knowledgeable, and Demographic Market Segments of Purchasers of Household Durables" (Ph.D. dissertation, Michigan State University, 1972), pp. 64-65. 2"David L. Wilemon, "An Analysis of the Age-Occupation Matrix as a Criterion for Vertical and Horizontal Market Delineation" (Ph.D. dissertation, Michigan State University, 1967), p. 36. 25Carl C. Zimmerman, Consumption and Standards of Living (New York: 0. Van Nostrand Co., Inc., 1936), p. 418. 26Wilemon, p. 40. 27Zimmerman, p. 39. 28Massachusetts Bureau of Statistics and Labor, Sixth Annual Report of the Bureau of Statistics of Labor, Public Document’3l (Boston: Wright & Potter, State Printers, 1875), p. 438. 29Wilemon, p. 43; and Zimmerman, p. 101. 3William A. Berridge, Emma A. Winslow, and Richard A. Flinn, Purchasing Power of the Consumer: A Statistical Index (New York: A. W. Shaw & Co., 1925), p. 168. 31J. M. Keynes, The General Theory of Employment Interest and Money (New York: Harcourt, Brace, & Co., Inc., 1936). 32J. S. Duesenberry, Income, Saving, and the Theory of Consumer Behavior (Cambridge: Harvard University Press, 1952). 84 33Frank, Massy, and Wind, pp. 177-184; and Andrew A. Brogowicz, "Race as a Basis for Market Segmentation: An Exploratory Analysis" (Ph.D. dissertation, Michigan State University, 1977), p. 70. 3"U.S., Department of Agriculture, Family Incomes and Expenditures, Misc. Pub. No. 339-489 (1939-1941). 35Carol Warwick Shaffer, "Income and Expenditure Patterns Related to the Life Cycle" (Ph.D. dissertation, Michigan State University, 1964). 36Janet Fisher, "Income, Spending, and Saving Patterns of Consumer Units in Different Age Groups," in National Bureau of Economic Research Studies in Income and Wealth (Princeton, N.J.: Princeton University Press, 1952), vol. 15, pp. 75-102. 37U.S., Department of Agriculture. 38Raymond C. Scott, Frozen Food Purchases in Syracuse, New York (Ithaca, N.Y.: Department of Agricultural Economics, New York State College of Agriculture, Cornell University, February 1948). 39Ib‘id., p. 11. “°Shaffer, p. 11. “‘Lincoln H. Clark, Consumer Behavior, vol. 2 (New York: New York University Press, 1955). “ZJanet Fisher, "Family Life Cycle Analysis in Research in Consumer Behavior," in Consumer Behavior, vol. 2, ed. Lincoln Clark (New York: New York University Press, 1955), pp. 28-36. I""John B. Lansing, and James N. Morgan, "Consumer Finance Over the Life Cycle," in Consumer Behavior, ed. Lincoln Clark, pp. 36-53. 1"'S. G. Barton, "The Life Cycle and Buying Patterns," in Consumer Behavior, ed. Lincoln Clark, pp. 53-57. l'sDonald Miller, "The Life Cycle and the Impact on Advertising," in Consumer Behavior, ed. Lincoln Clark, pp. 57-59. “55mith. “7A1derson. 1’83. C. Brandt, "Dissecting the Segmentation Syndrome," Journal of Marketing 30 (October 1966): 22-27. ”9Brogowicz, p. 70. 85 5°Smith. 51Wind, p. 317. 52Ronald D. Michman, "Market Segmentation Strategies: Pitfalls and Potentials," American Marketin Association Proceedings (Chicago: American Marketing Association, 1971 , p. 322. 53Brogowicz, p. 71. 5"Arndt, p. 25. 55Wind, pp. 317-318. 56Mack Hanan, Market Segmentation: The Basis for New Product Innovation and Old Product Renovation, American Marketing Association Bulletin No. 109(Chicago: American Marketing Association, 1968), p. 4. 57William H. Reynolds, "More Sense About Market Segmentation," Harvard Business Review, September-October 1965, pp. 107-114. 58Wind, pp. 320-321. 59Frank, Massy, and Wind, p. 176. 60Wind. 61Arndt. 62Wind. 63Robert P. Leone and Randall L. Schultz, "A Study of Marketing Generalizations," Journal of Marketing 44 (Winter 1980): 15. 6“Harper W. Boyd, Jr. "The JMR's Editorial Objectives," Journal of MarketingyResearch 13 (February 1976): 1-2. 65Leone and Schultz. 66Wind. 67Reynolds, pp. 107-108. 68wind, p. 326. 69Ibid., p. 327. 7°Robert C. Blattberg and Subrata K. Sen, "Market Segmentation Using Models of Multidimensional Purchasing Behavior," Journal of Marketing 38 (October 1974): 17-28; and Blattberg and Sen, “Market Segments and Stochastic Brand Choice Models," Journal of Marketing Research 13 (February 1976): 34-45. 86 71Roger J. Calantone, "Research Methodologies for Benefit Segmentation Analysis" (Ph.D. dissertation, University of Massachusetts, 1976), Abstract listed in "Marketing Doctoral Dissertation Abstracts," Bibliography Series No. 29 (Chicago: American Marketing Association, 1977), pp. 26-27. 72Roger J. Calantone and Alan G. Sawyer, "The Stability of Benefit Segments," Journal of Marketing Research 15 (August 1978): 395-404. 73Terry Mathew Weisenberger, "Generalized Market Segments: A Study Using Selected Convenience Goods in Vigo County, Indiana" (Ph.D. dissertation, Michigan State University, 1977), pp. 123—124. 7"Ibid. 7SNariman K. Dhalla and Winston H. Mahatoo, "Expanding the Scope of Segmentation Research," Journal of Marketing 40 (April 1976): 36. 76Yoram Wind and P. E. Green, "Some Conceptual, Measurement, and Analytical Problems in Life Style Research," in Life Style and Psychographics, ed. R. Haley (Chicago: American Marketing Association, 1974), pp. 99-126. 77Joseph T. Plummer, "The Concept and Application of Life Style Segmentation," Journal of Marketing 35 (April 1971): 35. 7“Leone and Schultz, p. 15. 79Ibid. 8°Alfred Marshall, Principles of Economics (London: The Macmillan Company, 1890). 81Frank, Massy, and Wind, pp. 177-184. 82Phillip Kotler, "Behavioral Models for Analyzing Buyers," Journal of Marketing 29 (October 1965): 37-45. 83Kurt Mayer, "Diminishing Class Differentials in the United States," in Marketing and the Behavioral Sciences, ed. Perry Bliss (Boston: Allyn & Bacon, Inc., 1963), p. 203. 8"Reynolds, p. 114. 85Wind, p. 319. 86John O. Tollefson and V. Parker Lessig, "Aggregation Criteria in Normative Market Segmentation Theory," Journal of Marketing Research 15 (August 1978): 346-355. 87 87Martin L. Bell, Marketing Concepts and Strategy (Boston: Houghton Mifflin Co., 1979), pp. 124-125. 88Ibid., p. 123. 89Reynolds. 90Bell, p. 124. 91Wind. 92Bell. 93Wind, pp. 326-327; and Calantone and Sawyer. 9"Bell. 95Vijay Mahajan and Arun K. Jain, "An Approach to Normative Segmentation," Journal of Marketing Research 15 (August 1978), pp. 338-345. 96Frederick W. Winter, "A Cost-Benefit Approach to Market Segmentation," Journal of Marketing 43 (Fall 1979): 103-111. 97Mack, pp. 25-27. 98W. J. E. Crissy, Paul Fischer, and Frank H. Mossman, "Segmental Analysis: Key to Marketing Profitability," M.S.U. Business Topics 21 (Spring 1973): 42-49. 99American Accounting Association, “Report by the Committee on Cost and Profitability Analysis for Marketing," Supplement to the Accounting Review, 1972. Cited in Crissy, Fischer, and Mossman, pp. 45-46. 1“Frank H. Mossman, W. J. E. Crissy, and Paul M. Fischer, Financial Dimensions of Marketing Management (New York: John Wiley & Sons, 1978), pp. xi, 1-6. 101Phillip Kotler, Marketing for Nonprofit Organizations (Englewood Cliffs, N.J.: Prentice-Hall, Inc., 1975), pp. 114-120. “’2 Mack, p. 24. 1“Jack Trout and Al Ries, "Positioning Cuts through Chaos in the Marketplace," Advertising Age, May 1972, reprinted in Marketin Classics, 4th ed., ed. Ben M. Enis and Keith K. Cox (Boston: Allyn & Bacon, Inc., 1981), pp. 422-423. 1“Mahajan and Jain, pp. 339-340. 88 105Wind, pp. 317-318. 1“Sibley, pp. 64-65; and Henry Assael and Hugh Cannon, "Do Demographics Help in Media Selection?" Journal of Advertising Research 19 (December 1979): 7-11. lo7Plummer,p.34. 108Brogowicz, pp. 71-72. 109Daniel Yankelovich, "New Criteria for Market Segmentation," Harvard Business Review 42 (March-April, 1964): 84. 110Ibid., pp. 83-84. 111Wind, p. 319. 112Ibid., p. 320. 113Shirley Young, Leland Ott, and Barbara Feigin, "Some Practical Considerations in Market Segmentation," Journal of Marketing Research 15 (August 1978): 405-411. 11"Ibid., p. 411. 115Arndt, pp. 24-25. 116Ibid., p. 24. 117Dik Warren Twedt, "How Important to Marketing Strategy 15 the Heavy User," Journal of Marketing 28 (January 1964): 71-72. 118Russel I. Haley, "Benefit Segmentation: A Decision-Oriented Research Tool," Journal of Marketing 32 (July 1968): 30-31. 119Twedt, p. 72. 120Haley, pp. 30-35. 121Theodore Levit, "Marketing Myopia," Harvard Business Review. July-August 1960, reprinted in The Great Writings in Marketing, ed. Howard Thompson (Plymouth, Mich.: The Commerce Press, 1976), pp. 36-57. 122Wind, pp. 319-321. 123Frank M. Bass, Douglas J. Tigert, and Ronald T. Lonsdale, "Market Segmentation: Group Versus Individual Behavior," Journal of Marketing Research 5 (August 1968): 264-270; and John J. Wheatley, John S. Y. Chiu, and Andrea C. Stevens, "Demographics to Predict Consumption," Journal of Advertising Research 20 (December 1980): 31-38. 89 12"Young, Ott, and Feigin, pp. 405-412; Arndt, PP- 15'273 30d Sibley, pp. 64-76. 125Ronald E. Frank, William F. Massy, and Harper W. Boyd, "Correlates of Grocery Product Consumption Rates," Journal of Marketing Research 4 (May 1967): 184-190. 126Ibid., p. 189. 127Ibid. 128Bass, Tigert, and Lonsdale. 129Wheatley, Chiu, and Stevens. 130Bass, Tigert, and Lonsdale, p. 265. 131Frank, Massy, and Boyd. 132Bass, Tigert, and Lonsdale, p. 267. 1331bid., p. 268. 13"Ibid., pp. 268-270. 135Wheatley, Chiu, and Stevens, p. 33. 1351bid., pp. 35-36. 137Ibid., p. 31. 138Franklin B. Evans, "Psychological and Objective Factors in the Prediction of Brand Choice," Journal of Business 32 (October 1959): 340-369. 139Michman, p. 322. 1“Arndt, p. 17. 1“Haley, pp. 30-35. 1”Ibid; and Dhalla and Mahatoo, p. 34. l”Dhalla and Mahatoo, p. 35. 1“"‘Assael and Cannon, pp. 7-11. 1H5Ibid., p. 8. 1"‘5Arndt, p. 25. 1”Assael and Cannon. 90 1"81bid., p. 10. 1"""Orange Juice, Dinners, Vegetables, and Pizza Head List for Fastest Moving Frozen Food Items," Supermarketing, 1 April 1975, p. 42. 15°Arndt, p. 26. 151Raymond C. Scott, "Frozen Food Purchases in Syracuse, New York," Department of Agricultural Economics, New York State College of Agriculture, Cornell University, 1948. 152Smith. 153Scott, pp. 6, 10-12. 15"Ibid., pp. 8-9. 155Ibid., pp. 19-22. 156Bernard J. LaLonde, "A Profile of the Frozen Food Consumer," Frozen Food Factbook, 1973 (Hershey, Pa.: National Frozen Food Association, 1973), pp. 61-91. 157Ibid., p. 64. 158Ibid., p. 74. 159Ibid., p. 88. 16°Bernard J. LaLonde, Robert A. Hansen, and Carol A. Scott, "Profile 1975: The Consumer and the Frozen Food Industry," Frozen Food Factbook, 1975 (Hershey, Pa.: National Frozen Food Association, 1973). 161Ibid., "Conclusions and Implications," no pages numbered. 162"Better Homes and Gardens Surveys the Frozen Food Markets," Frozen Food Factbook, l98lTTHershey, Pa.: National Frozen Food Association, 1981), pp. 40-62. 1631616., pp. 47-48. 16"Ibid., p. 60. 165"Frozen Food Study, 1971-72," Akron Beacon Journal, Akron, Ohio, 1972; and "Frozen Food Study, 1977," Akron Beacon Journal, Akron, Ohio, 1977. 166Twedt, p. 72. CHAPTER III RESEARCH DESIGN AND METHODOLOGY Introduction The purpose of this chapter is to explain the research methodology employed in this study. The first section explains the source of the research data base, and explains the methods used in the collection of the survey data. The second section identifies the independent and dependent variables that will be utilized in the course of testing the research hypotheses. In the final section, the actual process of analyzing the data is explained, including the computer methodology and the statistical techniques and testing standards used in the empirical research. Source of the Research Data The research will be based upon a survey of frozen food purchases conducted by the Akron Beacon Journal, the results of which were published in non-analyzed, cross-tabulation form under the title, "Frozen Food Study, 1977."1 The data base for the original, 1977 report was collected on a continuous twelve month basis from January 1976 to December 1976 by Bardsley & Haslacher, Inc., a Palo Alto, California marketing research firm. The survey consisted of 1,439 interviews, with each monthly sample designed to comprise a smaller sample of the total 91 92 survey sample. However, after receiving permission from the Akron Beacon Journal for the use of the total survey sample, it was discovered that the last six months of the data had been destroyed, and thus the data sample would be reduced to the results of the interviews collected from January 1976 to the end of June 1976. Therefore, the data base for this study will consist of 720 interviews. Interviews were collected in the metr0politan area of Akron, Ohio, including the Ohio counties of Medina, Summit, and Portage, with a sampling universe consisting of all permanent-residence households with installed cooking facilities, in these counties. The total sample was allocated among the counties in proportion to the number of house- holds with listed telephone numbers in each county, based upon the current telephone directories covering the survey area. Using a random procedure, addresses were selected from each telephone directory in the survey area. Next, using current maps, the nearest street intersection was located for each address obtained from the phone directories. Maps for each start-point were then reproduced to assist interviewers in locating the selected households to be interviewed. Interviewers were then instructed to apply an interval to each original start-point, to select a cluster of eight households for interview. Interviews were collected in person by the interviewers, in the home of each survey respondent. Media-audience and demographic characteristics were asked or recorded from all respondents, in addi- tion to questions about specific product purchases. Relevant sections 93 of the questionnaire, relating to data used in this study, are reproduced in Appendix A. Identification of Research Variables Independent Variables The primary set of independent variables to be tested in this study will consist of demographic characteristics of the survey respondents. Thirteen specific demographic variables will be examined, including measures of the following respondent characteristics: 1. Race, 2. Age of the head of household, 3. Occupation of the head of household, 4. Sex of the head of household, Sex of the respondent, Household income, Education of the head of household, oowaim Rent or own the household residence, 9. Length of residence at the same address, 10. Number of persons in the household, 11. Number of persons employed in the household, 12. Type of residence structure, and 13. Length of residence in the Akron metropolitan area. Two commonly surveyed consumer purchasing characteristics will also be considered as independent variables, relative to their predictive relationship to the purchase of the dependent categories 94 of frozen food purchases. These consumer purchasing independent variables are: 1. Total grocery purchases, and 2. Ownership of a separate freezer unit. In addition, the purchase responses for each of twelve specific frozen food categories will also be examined as independent, product- purchase variables, to determine the degree that one specific product purchase can be predicted by means of prior knowledge of the respon- dents' purchases of other frozen food product categories. These twelve product-purchase variables will also serve as the dependent variables, and will be listed in more detail, as dependent variables. Dependent Variables The dependent variables of this study will consist of the reported purchasing activity of the survey respondents in twelve specific categories of frozen food products. These specific frozen food purchase categories include: 1. Breakfast-baked goods, Dessert-baked goods, Seafoods, #0)“) Uncooked poultry, whole or parts, 5. Cooked (heat and serve) poultry, 6. Potatoes, 7. Vegetables, 8. Fruit juices, 9. Fruit or berries, 95 10. Pizza, 11. Complete dinners, and 12. Main courses, not including poultry. Analysis of the Data Computer Programs Used in the Analysis Survey data will be analyzed by the use of the Statistical Package for the Social Sciences,2 primarily via subprograms "Breakdown"3 and "Discriminant."“ First of all, data distribution will be examined for all variables used in the study, and then recoded by SPSS to eliminate obvious coding errors, non-responses, etc. Next, using the SPSS program "Breakdown,"S a one-way analysis of variance will be used to determine if the group means between the categories of the independent variables of the study are different in terms of their reported purchase patterns within the dependent, specific product purchase variables. For example, the data would be tested to determine if there was a statistically significant difference in the mean purchase of breakfast baked goods between white and non-white purchasers (race of the respondent). Actual testing of this statistical relationship would be done by computation of the "F-ratio," which is then compared to statistical tables of this measurement, to determine if a statistically valid relationship is present. Thus, the F-test will be calculated for each of the thirteen demographic variables, in relationship to each of the twelve dependent variables of product purchases, for a total of 156 bivariate tests of significance. In addition, this same computer program would be used 96 to test the relationship of each product-purchase variable, in relationship to each of the other product-purchase varibles of the study, for a total of 132 additional tests of statistical significance. "Breakdown"6 will also be used to test the significance of the relationship of total grocery purchases and freezer ownership to the purchase of the twelve product-purchase dependent variables of the study, resulting in 24 more bivariate tests of statistical significance, via the F-test. Next, the statistic Eta squared will be calculated by means of "Breakdown,"7 for each of the bivariate relationships which are found to be statistically significant in the F-test, to determine the total, non-linear bivariate variance in each of the dependent, product-purchase variables of frozen food purchase that can be explained by means of the tested, independent variable. Each significant relationship will then be examined to determine the linear predictive power, as measured by the statistic R squared. A specific test of linearity for each statistically significant relation- ship will then be conducted, by comparison of the non-linear Eta squared value to the linear R squared value for each tested predictive relation- ship. If R squared is subtracted from the Eta squared value of a rela- tionship, a measurement is thus obtained of the explanatory power that would be lost by an assumption of linearity for the bivariate, predictive relationship. In the event that a meaningful difference in predictive power is excluded by use of the linear R squared measurement of the relation- ship, rather than the non-linear Eta squared value, that independent 97 variable will be recoded, when possible, to allow a closer fit between the linear R squared value for that predictor variable, and the total non-linear predictive Eta squared value of the variable, to provide a better linear predictive input for the linear multivariate methodology that will follow in this study. Finally, the SPSS subprogram "Discriminant"8 will be used to determine the total, linear multivariate amount of purchase variance that can be explained by the combined use of either the total set of demographic variables or the product-purchase variables, in relation- ship to each of the dependent, product-purchase variables of the study. For each category of frozen food purchases, a linear discrim- inant function will be calculated, which will be used by "Discriminant"9 to classify each consumer respondent as a purchaser or a non-purchaser of that product category, and then to calculate the accuracy of this predicted classification, relative to the actual, reported purchases by these respondents. Based upon the accuracy of the consumer classifications observed in this classification test, the effectiveness, in total, of demographic variables will be compared to the relative effectiveness of the product- purChase variables, in determining which set of predictive variables was best able to correctly segment respondents into the proper classifica- tions of purchasers and non-purchasers for each of the specific frozen food categories of this study. In addition, "Discriminant"1° will be used to calculate the multivariate statistic of the canonical correlation, for each dis- criminant function, to determine the linear, statistical predictive 98 value that can be attributed to the demographic and the product-purchase variables, in relationship to each of the frozen food product categories. The canonical correlation, when squared, presents the statistical equivalent of a multivariate R squared measurement, for each tested multivariate relationship. Statistical Techniques and Measurements Statistical significance: One-way analysis of variance. The initial statistical technique to be employed in the study will be a one-way analysis of variance, a technique to determine if the group means of subsamples within a sample base are statistically different from each other.11 In the case of this study, the goal would be to determine if purchasing of specific product categories of frozen food products differ among experimentally altered categories of the inde- pendent variables of the study, primarily the demographic variables of the survey sample. Two variances are pitted against each other; one variance, presumably due to the experimental, independent variable, such as race, is tested against another variance that is presumed to be due to random error or chance'2--in this case, the purchase of a frozen food product, such as frozen breakfast-baked goods. If nothing beyond chance is causing group mean scores to vary, then the variance in observed purchases of breakfast goods between-gnoups (between the racial categories) will not vary beyond the measure of chance error, that is the expected variance among any group, or the within-group variance. If an experimental manipulation has been influential, the 99 variance between-groups should show an increase, beyond the differences that can be attributed to chance; in the case of the given example, the test would be if group consumption means of breakfast goods vary sig- nificantly among different categories of race within the survey sample. Stated in formal null hypothesis form, the research would be testing the proposition that there is no significant difference between racial categories, relative to their reported purchases of breakfast-baked goods. If the between-group means are found to be significantly different, this null hypothesis can be rejected, and the variable of race would be regarded as showing differences in purchasing, between the categories of race, that is statistically significant. The test method provided by subprogram "Breakdown" is the F-test, where the F-ratio (F==between group means square/within-groups mean square) is computed and compared to the known sampling distribution of the F-ratio.13 If the F-ratio so computed from the sample data is as great or greater than the tabled entry of the F-table, the differ- ences between the experimental group-means are regarded as significantly different; that is, the difference is greater than can be reasonably attributed to chance distribution of expected within-group variations between the categories of the independent, experimental variable. The specific standard regarded as "statistically significant" typically is reported as the .05 or the .01 level. That is, an .01 level of null hypothesis rejection means that an observed between-group variation could be attributed to chance only one time in 100 samples, given the sample's within-group variation. 100 For purposes of this study, the defined standard for a finding of statistical significance for a relationship will be the more strin- gent level of .01. However, to allow comparability with other past and future studies using the more common, and more uncertain, sig- nificance standard of the .05 level, relationship levels at .05 will be reported in the findings, and also tested for their bivariate predictive strength. However, it should be noted that a finding of statistical significance, even at the more stringent .01 level, does not provide a measure of the strength of the relationship measured pey_§e; it only indicates that a relationship exists that is not likely due to mere chance--a nuance in interpretation of results that seems to have eluded far too many researchers. Therefore, the next task of the research methodology of this study is to measure the strength of the relationship that may exist within statistically significant associations between the independent and the dependent variables of the study. Strength of the relationships: Eta squared and R squared. Subprogram "Breakdown"1“ is again utilized to calculate the measurable strength of the observed, statistically significant relationships between the independent variables of the study and the dependent measures of product purchases. The methods used will be the general correlation ratio of Eta squared, and the linear correlation ratio of R squared. 101 Eta squared provides a general, non-linear measurement of the strength of the variance shared by the dependent and the independent variables being tested, and gives a measure of the proportion of vari- ance in the dependent, product-purchase variable that can be attributed to variation in the independent, demographic variables that are the heart of this research, as well as for the additional variables tested as independent variables. This statistic is a bivariate measure of association calculated by dividing the between-group sum of squares (between categories of the tested independent variable) by the total sum of squares for all purchases of the tested dependent variable of the specific product purchased.15 For purposes of marketing application, a meaningful degree of predictability of the variation in the dependent, product purchase variables is needed, rather than the limited knowledge that "some" relationship exists. While no specific measurement of what strength of relationship is "meaningful" has been established, the rule that will be adopted by this study is that for a relationship to be regarded as meaningful, a bivariate Eta squared value of .05 or more must be obtained for statistically significant relationships within the survey data. Linearity of the relationships. Next, a specific test of linearity will be conducted by calculation of the linear, Pearson's R statistic, which when squared provides a measurement of the linear proportion of variance in the dependent variable that can be attributed to the effect of the tested, independent variable.16 If R squared is subtracted from the Eta squared measure for the same independent- 102 dependent variable relationship, a measure is thus obtained of the degree of explanatory power that would be lost by use of the linear R squared measurement, as compared to the total degree of predictive power provided by Eta squared. The reason for this test is to determine if the R squared statistic provides a reasonably accurate measurement of the predictive power of the tested independent variables of the study, for further use in the multivariate, linear discriminant analysis to be used in the subsequent stage of this study. If a major disparity is observed between Eta squared and R squared, the commensurate independent variable would then be recoded, when possible, to derive a variable, or combination of variables,that comes closer to the non-linear, Eta squared predictive power of that variable, for use in the discriminant analysis to follow. Discriminant analysis. The purpose of discriminant analysis can be generally described as providing a statistical ability to distinguish groups from one another on the basis of linear sets of measures. It can also be used to assign or segment group members on the basis of their scores on a combination of independent, predictor variables.17 Both uses will be applied in this study. First of all, a multivariate function will be determined for each of the product purchase variables of the study, in such a way as to provide the maximum ability to distinguish or "discriminate“ between study groups, in the sense of being able to tell them apart by means of the predictive variables, relative to the dependent trait or activity. In this study, discriminant analysis will be used to determine which combination of variables is best able to distinguish or segment users 103 from non-users of specific frozen food products. Additionally, the combined predictive measure of the multivariate function is determined in a statistic similar to the linear R square statistic of the bivariate analysis; a canonical correlation is calculated, which when squared indicates the proportion of product purchase variance which can be explained by the combined, multivariate function.18 Finally, the total, linear multivariate function is utilized to segment purchasers from non-purchasers for a given frozen food product category, by means of individual respondent scores, which will be calculated from individual traits of the survey respondents, as weighted by the discriminant function. After all respondents have been classified, the degree to which product users and non-users were correctly segmented will then be determined, both as a measure of total effectiveness, and to determine whether demographic or product purchases of other products can provide the best method of segmenting the purchasers from the non-purchasers of specific frozen food product categories. For purposes of this study, the defined standard of classi- fication success, in predicting the correct classifications of specific product purchasers and non-purchasers, was set at a level of 75 percent or greater, to meet the hypothesized requirement for multivariate classification ability. FO0TNOTES--CHAPTER III 1"Frozen Food Study, 1977,’I Akron Beacon Journal. Akron, Ohio, 1977. 2Norman H. Nie, C. Hull, J. Jenkins, K. Steinbrenner, and 0. Brent, Statistical Package for the Social Sciences (New York: McGraw-Hill Book Co., 1975). 3Ibid., pp. 249-275. l'Ibid., pp. 434-467. 5Ibid., pp. 249-275. 6Ibid. 7Ibid. 8Ibid., pp. 434-467. 9Ibid. 10Ibid. 11Ibid., pp. 259-260. 12Fred N. Kerlinger, Foundations of Behavioral Research (New York: Holt, Rinehart, & Winston, Inc., 1973), pp. 220-229. 13Nie et al., pp. 259-260. 1"Ibid., pp. 260-261. 15Kerlinger, pp. 230-232. 16Nie et al., pp. 260-261. 17Ibid., pp. 434-441; and Kerlinger, pp. 650-653. 18Nie et al., p. 442. 104 CHAPTER IV PRESENTATION OF FINDINGS Introduction The purpose of this chapter is to present and discuss the empirical findings of this study. The first section presents research findings relating to the value of demographic consumer variables, relative to their ability to predict and segment the purchasing behavior for specific categories of frozen foods. Following the order of the research hypotheses about demographic variables, as set forth in Chapter I, these consumer characteristics are examined for the statistical significance and the bivariate predictive power rela- tive to the tested categories of frozen food products. Demographic variables are then tested for their multivariate ability to predict the purchase of the frozen food products of this study. The second section of this chapter examines the predictive value of demographic variables, in comparison to the predictive value of product purchase information. Demographic data and product-purchase information are compared by means of their relative bivariate predictive powers, and by their multivariate ability to classify the users and non-users of each specific frozen food product category. The third section of this chapter presents findings about the ability to predict purchasing behavior for frozen food products, by 105 106 means of information about a consumer's total purchases of grocery products. Finally, the fourth section of this chapter presents findings about the ability to predict frozen food purchases by means of infor- mation about the consumer's ownership of a separate freezer unit. The Value of Consumer Demographic Variables The major hypothesis of this study, as discussed in Chapter I, was that demographic consumer variables ppn be used to successfully predict and segment purchasing behavior for consumer products. Specifically, this section of the study reports on the observed ability to predict and segment twelve specific categories of frozen food products, by means of thirteen demographic consumer characteristics. During the process of testing this major hypothesis, demographic variables are tested in three stages, or subhypotheses. First, an analysis of variance is performed, to determine if a statistically significant relationship exists for any of the demographic variables, in relationship to the surveyed purchases within each of the twelve frozen food product categories. Second, these same bivariate rela- tionships, between the demographic and product purchase variables, are examined to determine if a meaningful degree of predictive power exists. The predictive power of the demographic variables are examined on both a total, non-linear basis, and also on a linear basis. Third, the multivariate value of all thirteen demographic variables is tested, relative to the ability to predict and segment the purchases of each of the frozen food categories. 107 Analysis of Variance--Test of Significance Utilizing subprogram "Breakdown," within the Statistical Package for the Social Sciences,1 an analysis of variance was performed to determine if the relationship of any of the demographic variables was statistically significant to any of the frozen food product categories. Actual statistical testing was done by computation of the F-ratio between group purchase means within each demographic variable category, in relationship to each of the product purchase categories. Thus, for example, purchases for breakfast-baked goods were examined to determine if group purchasing means differed signif- icantly between racial categories of white versus non-white consumers. The F-ratio which is thus obtained is compared to statistical tables of this measurement, to determine if statistically valid differences in product purchases can reasonably be attributed to demographic differences in the consumer's race. This analysis of variance, via the F-test, was performed for each of the thirteen demographic variables in the study, in relationship to the purchases within each of the twelve product categories. The results of these tests are presented in Table 4.1. The specific standard regarded as statistically significant for this study has been set at a null-hypothesis level of .01. How- ever, since many studies choose to accept relationships at the .05 level as being significant, Table 4.1 presents the significance levels up to the .05 level. In addition, it may be noted that several Table 4.1 108 The Statistical Significance of the Demographic Variables, in Relationship to Frozen Food Purchases Frozen Food Categories m m 3 .313 1’ 3 .8 as .s.§ a a). r z a a... a... .. z I- O .‘KL ‘0: 8 Q U 9‘- OJ 1: 01: 0 cu m» u u u aq- ua Pa) CO 01m “- 01-- .KO— Q Q w- ?L N C at . 2.: mat on u: 8: u as E at- N g: '3 Demographic Factors up 8 8 8 191’ 58 u 8 a? 3 u u. :3 E u ‘5 2'? 3 Race NS6 NS NS NS NS NS .006 NS NS NS NS NS Age .008 NS .000 .035 NS .000 NS .009 NS .000 .025 NS (NS) (NS) Occupation of head of .015 NS .004 NS NS .000 NS .009 NS NS NS NS the household (NS) Sex of head of the .037 NS NS NS NS NS NS .008 NS .049 .038 NS household (NS) (NS) (NS) Sex of respondent NS NS .017 NS NS NS .001 NS NS NS _ .047 NS (NS) (NS) Income .006 .004 .001 NS NS NS NS .000 NS .012 NS NS (NS) Education NS .023 .001 NS NS NS NS .000 NS .012 NS NS (NS) (NS) Rent or own residence NS NS .026 .015 NS NS .018 .020 NS NS .028 NS (NS) (NS) (NS) (NS) (NS) Length of residence NS NS .002 .026 .030 .000 NS NS NS NS .000 .038 in home (NS) (NS) (NS) Number in household .008 NS .000 NS NS .000 NS z02? NS .000 NS NS NS Number employed in NS NS .008 NS .003 .002 NS .000 NS .001 NS NS household Type of home NS NS NS NS NS NS NS NS NS NS NS NS Length of residence .016 NS .001 .019 NS .011 NS .000 NS .004 .026 NS in the area (NS) (NS) (NS) aSignificance levels less than the .05 level are reported as NS. Per requirements of the research hypotheses, an .01 level is required for determination of a significant relationship in this study. However, significance levels between .01 and .05 are reported in the above exhibit, but are noted as being not significant (NS) at the .01 level. 109 relationships were significant at the level of .012, .015, or similar levels that were extremely close to the .01 level. Therefore, relationships that were found at levels of significance between .01 and .05 are reported, but are also noted in parentheses as being not significant at the .01 level, as demonstrated in the survey sample used in this study. Based upon the results presented in Table 4.1, the statistical validity of each of the demographic variables will be discussed in turn, as they relate to the purchase of the frozen food product categories of this study. Statistical significance of race. Within the twelve frozen food product categories studied, the variable of race was shown to be significant for only one product category, that of frozen vegetables, at the .006 level. Even at the more liberal significance level of .05, no additional product category showed a significant level of relation- ship to the race of the purchaser, beyond group mean differences that could reasonably be attributed to mere chance in sampling. Thus, race did not seem to be an important variable, relative to most frozen foods. Statistical significance of age. The variable of age was significantly related to five of the twelve frozen food categories at the .01 level, and exhibited a relationship to two additional products at the .05 significance level. At the .01 level, age was significantly related to the purchase of breakfast-baked goods, potatoes, seafoods, fruit juices, and pizza, with weaker .05 ties shown to uncooked poultry and complete dinners. However, even beyond the seven frozen product categories where at least a .05 level of significance was noted, it 110 seems somewhat curious that seemingly related items did ngt_show a significant tie. For example, breakfast-baked goods was significant at the .008 level, but dessert-baked goods was not significant at all. Uncooked poultry was significant at the .035 level, but cooked poultry did not exhibit a statistically significant link to the variable of age. Similarly, frozen potatoes was significant at the .000 level, but frozen vegetables exhibited no significant relationship at all. Pizza was significant at the .000 level, as was frozen complete dinners at the .025 level, yet, main courses did not exhibit significant ties to the age of the purchaser. Statistical significance of occupation. The occupation of the head of the household was significant at the .01 level with three product categories: seafoods, potatoes, and fruit juices. In addition, significance was noted at the level of .015 with breakfast-baked goods. Statistical significance of the sex of the household head. Sex of the household head was significant for only one product, fruit juices, at the .01 level. However, three additional product categories were significant at the .05 level: breakfast-baked goods, pizza, and complete dinners. Thus, this variable also seemed to be one of the weaker demographic indicants of frozen food purchasing behavior. Statistical significance of the respondent's sex. Sex of the respondent was significant for one product, frozen vegetables, at the .01 level, and was linked to seafoods and complete dinners at the .05 level. This variable might be regarded as somewhat of a control vari- able within the survey, where differences in reported purchases might be due more to bias, or even lack of knowledge on the part of the 111 respondent, especially regarding substantial differences from the purchase traits previously noted in relationship to the sex of the head of the household. Thus, strong differences were not actually expected relative to the sex of the respondent, and were not observed in the sample responses for most of the products. Statistical significance of income. Income was shown to meet the .01 test of significance for four frozen product categories: breakfast-baked goods, dessert-baked goods, seafoods, and fruit juices. In addition, one additional product, pizza, barely missed the .01 level of significance, at .012. Statistical significance of education. The educational level of the household head was shown to be significantly related at the .01 level with only two products: seafoods and fruit juices. One addi- tional category, pizza, barely missed the .01 level, with a .012 test of significance. Dessert-baked goods also was significant at the .05 level. Nevertheless, eight of the twelve categories failed to demon- strate even an .05 degree of statistical significance with the variable of education. Statistical significance of renting or owning the residence. Renting or owning the residence did not exhibit any significant ties to the frozen products of the study at the .01 level of significance. However, five product categories showed a significance level between .01 and .03. These product categories were seafoods, uncooked poultry, vegetables, fruit juices, and complete dinners. 112 Statistical significance of the length of residence in the ppme, The length of residence in the respondent's current home was significant at the .01 level for three product categories: seafoods, potatoes, and complete dinners. Three additional products were shown to have a significant tie at the .05 level: uncooked poultry, cooked poultry, and main courses. It might be noted that length of residence in the home was the only demographic variable, even at the .05 level, to exhibit a significant relationship to both uncooked and cooked poultry products, and was one of only two factors to exhibit even a .05 significance with the category of cooked poultry. Statistical significance of the number living in the household. The number of persons living in the household was significant at the .01 level with four product categories: breakfast-baked goods, seafoods, potatoes, and pizza. Fruit juices was also significant at the .05 level. Statistical significance of the number of persons employed. The number of persons employed was statistically significant at the .01 level for five product categories: seafoods, cooked poultry, potatoes, fruit juices, and pizza. No additional relationships were observed, even at the .05 level of significance. Statistical significance of the type of home. The type of home, such as single-family, apartment, etc., was not significant for any of the twelve product categories of the study, the only demographic vari- able within the study to have this distinction. 113 Statistical significance of the length of residence in the apea. Length of residence in the metropolitan area of the survey was significant at the .01 level for three product categories: seafoods, fruit juices, and pizza. Additionally, four other product categories exhibited relationships that were very close to the .01 level. Breakfast-baked goods showed a .016 level of significance, uncooked poultry was .019, potatoes barely missed the .01 cutoff with a .011 reported significance level, and complete dinners exhibited a .026 significance level with length of residence in the area. Thus, this variable, along with length of residence in the present home, seem to be two of the strongest links to frozen food product purchases. Number of significant demographic variables per product. Table 4.2 shows the number of demographic characteristics that proved to have a significant statistical relationship to each of the frozen food product categories, at both the .01 and the .05 level of significance. As can be seen in this table, the number of significant demographic variables varies substantially within the set of frozen product categories. For example, the product category of seafoods exhibited a total of ten of the thirteen demographic variables with a significance of .05, and eight of these ten demographic variables exhibited statistical significance at the hypothesized .01 level. However, npne_of the tested demographic variables exhibited statis- tical significance in relationship to the purchase of the product category of fruit or berries. Similarly, only one demographic factor, length of residence in the current home, exhibited as much as a .05 114 Table 4.2 Number of Demographic Variables Significant No. of Variables No. of Variables Significant Significant Product Categories at .01 at .Oll-.05 Total Breakfast-baked goods 3 3 6 Dessert-baked goods l l 2 Seafoods 8 2 l0 Uncooked poultry 0 4 4 Cooked poultry l l 2 Potatoes 5 la 6 Vegetables 2 l 3 Fruit juices 7 2 9 Fruit or berries O 0 0 Pizza 4 3b 7 Complete dinners l 5 6 Main courses 0 l 1 3At the .Oll level. bTwo were at the .0l2 level. llS significance level in relationship to the purchase of frozen main courses. Even apparently similar product categories exhibited markedly different numbers of significant relationships to the tested demographic independent variables. Thus, breakfast-baked goods had a total of six demographic variables exhibited a significant relationship at the .05 level, and with three factors at the .Ol level. However, dessert-baked goods had a total of only two demographic variables demonstrate signif- icance at the .05 level, and only one at the .Ol level. Moreover, only one demographic factor, income, was significantly related to the pur- chase of both breakfast-baked and dessert-baked goods. Similarly, frozen potatoes had five demographic variables exhibit statistical significance at the .0l level, and a sixth factor at the .Oll level. Yet, frozen vegetables had only two variables sig- nificant at the .Ol level, and one additional tie at the .05 level. Moreover, none of the demographic variables that exhibited a statistical significance, even at the .05 level, exhibited statistical significance for both frozen potatoes and frozen vegetables. Another odd quirk is that both variables that showed a .0l level relationship to the pur- chase of frozen vegetables, race and the sex of the respondent, did not exhibit a .Ol relationship to any other product. In fact, race did not exhibit agy_other significant relationship, even at the more liberal .05 level, to any other frozen product category in the study. ll6 Significance of the relationship between demographic variables and frozen food products--a summary of findings. Based upon the analysis of variance, as measured by the F-test, and as presented in Tables 4.l and 4.2, a basic summary of the results of the significance testing phase of this study would be that there are significant rela- tionships between demographic consumer variables and the purchase of specific frozen food product categories. However, the significance of the relationship between demo— graphics and the frozen product purchases seems to vary substantially between product categories, both in the number of demographic factors that proved significant per product category, as well as which specific demographic variables proved to be significantly related to the purchase of each specific frozen product category. There was no universal demographic variable that was significant at either the .Ol or the .05 level for all frozen product categories. In fact, the maximum number of significant relationships at the .01 level was exhibited by the factors of age and the number of persons employed in the household, each of these two factors showing an .Ol relationship to five frozen product categories. Several other demo— graphic variables exhibited .Ol relationships to three or four product categories, with additional relationships noted if the accepted significance level is expanded to .05. Demographics exhibited a total of ten significant relationships at the .05 level for the purchase of seafoods, nine .05 relationships to fruit juices, and seven relationships to the purchase of frozen pizza, based upon the thirteen demographic variables that were tested. 117 Only the product category of fruit or berries was not linked at the .05 level to at least one demographic factor. Only the type of home was found to exhibit no significant relationship at the .05 level of significance testing, although the factor of renting or owning the consumer's residence failed to show a relationship at the .0l level originally set as the significance level criteria for this study. Yet, even the factor of renting or owning the residence proved to display fivg_relationships, between .Ol and .03 significance levels, among the twelve product categories tested. At the .0l level of significance testing, demographic variables were found to relate significantly to nine of the twelve frozen product categories, with at least one demographic factor found to be signifi- cantly related. If the more liberal, and common, .05 significance level is accepted, eleven of the twelve product categories will have at least one demographic variable significantly related. Thus, the preponderance of the empirical testing of the frozen food purchase data indicates that the demographic variables of this study are generally able to define at least one significant relationship, relative to the purchase of specific frozen food product categories. The next stage of this study will be to determine the strength and the relative predictive power of such relationships as were found to exist in the significance testing, as reported in Table 4.l. 118 Analysis of Variance—-5trenQLh of Significant Relationships Subprogram "Breakdown" of the Statistical Program for the Social Sciences2 is again utilized, to calculate the measurable strength of the independent, demographic consumer variables, relative to their predictive value toward the dependent, product purchase categories of the study. The statistical methods used are the general correlation ratio, Eta squared, and the linear correlation ratio of R squared. Eta squared is used to provide the measurement of the total, non-linear proportion of variance in the product purchase categories that can be attributed to variation in the demographic characteristics of the survey sample. Next, a specific test of linearity is conducted by means of calculating the R squared value of the relationship between the demo- graphic variables and the product purchase categories. If R squared is subtracted from the Eta squared value for the same relationship, a measure is thus provided of the amount of explanatory power that would be lost in the multivariate, linear discriminant analysis that is used in the subsequent stage of this study. The results of these calculations are presented in Table 4.3, providing a measurement of the linear and non-linear relationships between each of the thirteen demographic variables of this study, relative to each of the twelve frozen food product categories. Table 4.3 The Measured Strength ll9 of Demographics as Predictorsa Frozen Food Categories .. m x: .58 a a .2 as as a a... >. r z a 3.. s .. ‘0- L O .KL “DS- 3 Q 0 0: OJ 3‘8 QB '00- OH 00 H H U “I o o—m in Ex 33 a 83 33 .3 % S Sc 3 g2 5: DEMOQ'APMC ”am as 83 .2 58 88 a? 3 3; :3 a: gas 28 Race NS NS NS NS NS NS .006 NS NS NS NS NS .010 Age .008 NS .000 .035 NS .000 NS .009 NS .000 .025 NS .029 .048 .023 .055 .028 .056 .024 .010 .043 .015 .044 .013 .029 .003 Occuptation of head .015 NS .004 NS NS .000 NS .009 NS NS N5 of household .029 .034 .056 .028 .017 .029 .026 .013 Sex of head of .037 NS NS NS NS NS NS .008 NS .049 .038 NS household .006 .009 .005 .006 Sex of the respondent NS NS .017 NS NS NS .001 NS NS NS .047 NS .008 .014 .005 Income .006 .004 .001 NS NS NS NS .000 NS .012 NS NS .036 .038 .043 .077 .034 .014 .016 .012 .058 .011 Education NS .023 .00l NS NS NS NS .000 NS .012 NS NS .020 .035 .077 .034 .005 .023 .058 .011 Rent or own residence NS NS .026 .015 NS NS .018 .020 NS NS .028 NS .007 .008 .007 .007 .006 Length of residence NS NS .002 .026 .030 .000 NS NS NS NS .000 .038 in present home .020 .012 .010 .040 .032 .012 .016 .Oll .006 .027 .004 .006 Number in household .008 NS .000 NS NS .000 NS .020 NS .000 NS NS .029 .049 .050 .025 .039 .020 .023 .020 .008 .026 Number employed in NS NS .008 NS .003 .002 NS .000 NS .001 NS NS household .0l9 .020 .023 .029 .025 .001 .011 .006 .012 .0l6 Type of home NS NS NS NS NS NS NS NS NS NS NS NS Length of residence .016 NS .001 .019 NS .011 NS .000 NS .004 .026 N5 in area .0l7 .027 .016 .018 .036 .020 .015 .004 .023 .010 .012 .019 .0l4 .022 aThe first reported figure for each variable refers to the significance level of the analysis of variance. Significance lower than the .05 level is reported as “NS.“ The second value is the Eta squared score, giving the total. non-linear predictive value. The third value, when present, gives the linear, R squared measurement, if different from Eta squared. 120 Within Table 4.3, the first reported value for each of the demographic variable to product category relationships refers to the significance level of the analysis of variance, as measured by the F-test. Significance levels lower than .05 are reported as not significant, or NS in the table, and no further tests are made for the strength of the relationship. The second line in Table 4.3 for each relationship of at least .05 significance level, presents the Eta squared value, giving the total, non-linear predictive strength of the relationship pair. The third line for each relationship pair in Table 4.3, if present, is the linear, R squared measurement of the predictive strength- of the relationship. However, several of the demographic variables are coded into only two responses. In this situation, a separate R squared measurement is not made, since the Eta squared value will be the same as the R squared value. Thus, for example, Race is coded into only two categories of white and non-white respondents. Table 4.3 shows that a relationship was significant at the .006 level for the purchase of frozen vegetables, with an Eta squared value of .010. No additional value is given for this relationship for R squared, since Eta squared and R squared have the same value. Thus, Race provides a statistically significant relationship to the purchase of frozen vegetables, but provides a predictive strength of only 1 percent, although this relationship is linear, due to coding into only two categories of the independent variable. 121 Based upon the Eta squared scores of each significant relationship, as presented in Table 4.3, the evaluation is then made, whether or not such demographic variables can provide a mean; jggjul_degree of predictability about the purchase of the frozen food product categories of this study. As discussed in Chapter III, the standard set for a relationship to be regarded as meaningful is that a bivariate Eta squared value of .05 or greater must be obtained, for statistically significant relationships. Strength of demographic variables relationships to frozen food purchases--a summary of findings. Based upon an examination of the Eta squared scores for the demographic variables tested in this study, demographics did not generally exhibit a meaningful predictive value relative to the purchase of the twelve frozen food categories surveyed. 0f the 156 relationships examined in this study (thirteen demographic variables times twelve frozen food categories) only six relationships were found to meet the stated criteria of providing an Eta squared value of .05 or greater. Two other relationships had measured Eta squared values of .048 and .049, with only one additional relationship testing as high as .040. The highest Eta squared value was found for both income and education, as these two factors related to the purchase of fruit juices, with scores of .077 indicating that either income or education gave 7.7 percent predictive value regarding the purchase of fruit juices. The demographic factors of age, income, education, and the number of persons in the household all had at least one Eta squared 122 relationship score of .05 or larger, with only the factor of age demonstrating two relationships with scores of .05 or larger. The strongest relationships were noted for the categories of frozen potatoes and fruit juices. Potatoes had .05 or larger Eta squared scores in relationship to the demographic variables of age, occupation, and the number of persons in the household. Length of residence in the home also gave a .04 score for the purchase of frozen potatoes. As noted previously, income and education had .077 scores relative to the purchase of fruit juices. The only other product that had a .05 relationship to any of the demographic variables was pizza, which had a .056 score, as related to the age of the household head. However, it should be noted that several other demographic variables showed Eta squared scores of .02 and .04 for other frozen product categories than those cited for .05 relationships, thus providing some potential for predictive power on a multivariate basis in the discriminant analysis performed by this study. The actual predictive Eta squared scores are reported in Table 4.3, for all relationships which proved to have statistical significance at the .05 level. However, results of testing the predictive power of such relationships show that all relationships meeting the predictive criteria of .05 Eta squared scores, as required to be considered meaningful predictors of product purchases, also exhibited significance in the F-test of .000. Relationships that 123 exceeded the .01 significance level rarely showed predictive value as high as 1 percent. The Linearity of the Predictive Relationshig The test of linearity employed in this study was to compare the linear measurement, R squared, to the total non-linear predictive measurement provided by Eta squared. The results of this test may also be seen in Table 4.3. The general conclusion is that many of the relationships between demographic variables and the purchase of frozen food products are non- linear to a substantial degree. Thus, the explanatory power of R squared often is somewhat less than the total predictive value of the non-linear measurement. Moreover, the degree of linearity varied between product categories, making it very difficult to make general- izations about specific demographic factors for all product categories. However, the general pattern that was observed within the group mean consumption scores in raw categorization format was that relation- ships were often curvilinear. That is, consumption tended to increase over the categorical range of most demographic factors, then declined again at the highest or largest classification groups. While the limited sample size within the user groups, combined with relatively small numbers of respondents in some demographic categories make over- generalization very tenuous, the basic conclusion seems apparent that the evaluation of the predictive power of demographic variables on solely a linear measurement basis would be likely to understate or 124 even ignore much of the possible predictive ability from demographics. Table 4.3 shows that half or more of the predictive power was often lost in utilizing the R squared value as the measurement method, as compared to the total, non-linear value provided by Eta squared. For example, the number of persons in the household showed a 5 percent predictive value toward fruit juices, and a 4.9 percent value for predicting the purchase of seafoods. However, the R squared pre- dictive values for these relationships were 2 percent and 2.3 percent, respectively, for a net loss of about 60 percent of predictive power. The basic conclusion would then seem to be that non-linear statistical methods would increase the predictive abilities of demographic vari— ables, and should be employed when possible. The implications and conclusions about past and future demographic research, as a result of this finding, will be discussed in Chapter V. However, the implication for this research study was that reasonable attempts should be made to recode demographic variables to allow a closer fit between the non-linear, Eta squared value and the linear, R squared value, to allow more effective use of the linear, multivariate discriminant analysis to follow in this study. The basic tactic employed was to study the group means of purchases within each category for a given demographic factor, and to combine categories with similar purchase means, thus reducing some of the non-linearity of the observed responses. Thus, for a demographic variable such as age, it was noted that purchase means increased with age, and then generally declined after 125 the age of 55. Over 55 year old purchasers were thus combined with the younger purchasers, who had purchase means similar to that of older consumers for many product categories, thus reducing the curvilinearity of the data, and giving a closer fit with the total, non-linear Eta squared measurement. Similarly, households with three or more persons employed were combined with households with none employed, since it was observed that group means of purchasing usually decreased if more than three persons were employed, to about the same levels of purchase frequency as for families where no one was employed, or the head of the household was retired. While some detail was no doubt lost by such recodings, the intent was not to alter the measurement results of the bivariate analysis of demographic variable predictive power, but rather to utilize a multivariate measurement and classification method, discriminant analysis, in a way that would utilize a greater amount of the predictive power of demographics in a linear methodology. The results of such recodings generally tended to reduce the measured Eta squared value of the demographic factor, but produced an R squared value that was higher than before, although still substantially lower than the Eta squared values reported in Table 4.3. The results of the multivariate use of demographic variables to predict and classify the purchasers of the specific frozen food products of this study are presented in the next section of this chapter. 126 Discriminant Analysis: Multivariate Classification of Buyers Versus Non-Bgyers Using subprogram "Discriminant" of the Statistical Package 3 all thirteen demographic variables were for the Social Sciences, employed in a linear, multivariate analysis to determine the total ability to distinguish between the buyers and the non-buyers of the twelve specific categories of frozen food products. The direct method of discriminant analysis was used, whereby all discriminating variables are employed at once, to determine the total predictive power from all the variables employed in the analysis. Tables 4.4 and 4.5 show the results of this analysis. In Table 4.4, the ability of the demographic variables to correctly distinguish between the buyers and the non-buyers of frozen food products is shown for each of the specific product categories. First, the percentage of all consumers who were correctly classified as buyers or non-buyers, based upon their demographic characteristics, is reported. Next, the specific percentage of buyers correctly classified is presented. Table 4.5 presents the canonical correlation of the combined, multivariate use of the demographic predictors, a value which can be generally interpreted as a multivariate R correlation. This value, when squared, presents the total, linear predictive ability of the demographic variables, relative to observed variation in the purchase of frozen foods. The next figure presented in Table 4.5 is the per- centage of non-buyers who were incorrectly predicted to be buyers. 127 Table 4.4 Total Classification Efficiency of the Discriminant Analysis, Utilizing the Demographic Variables Percentage of Percentage of All Consumers Buyers Frozen Food Category Correctly Classified Correctly Classified 2‘; Z». Breakfast-baked goods 55.9 62.4 Dessert-baked goods 56.8 60.8 Seafoods 61.1 66.8 Uncooked poultry 59.9 57.4 Cooked poultry 58.9 64.7 Potatoes 58.9 66.0 Vegetables 63.0 65.1 Fruit juices 64.8 66.6 Fruit or berries 59.3 56.3 Pizza 58.9 66.5 Complete dinners 64.4 48.5 Main courses 63.8 55.4 Average of all products 60.5 61.3 128 Table 4.5 Statistical Strength and Distribution of the Discriminant Analysis Using the Demographic Variables Percentage of Non-Buyers Percentage Canonical Incorrectly of Buyers Frozen Food Category Correlation Classified in Sample 2 1 Breakfast-baked goods .185 46.3 24.8 Dessert-baked goods .182 44.7 27.7 Seafoods .317 43.3 44.0 Uncooked poultry .176 39.6 17.0 Cooked poultry .189 41.7 9.5 Potatoes .250 47.0 45.1 Vegetables .231 42.1 70.2 Fruit juices .313 38.5 64.2 Fruit or berries .176 40.1 16.6 Pizza .244 43.8 26.2 Complete dinners .202 29.5 27.9 Main courses .185 33.1 27.2 Average of all products .220 40.9 33.3 129 Finally, Table 4.5 presents the actual percentage of the survey sample which was reported to purchase at least one package of each specific product during the 30 days prior to the survey. Classification efficiency of the discriminant analysis. As discussed in Chapter I, it was hypothesized that the multivariate discriminant analysis, using demographic variables as the predictive factors, would be able to provide a meaningful ability to distinguish between buyers and non-buyers of the specific frozen food product categories. For purposes of this study, meaningful was defined as providing at least a 75 percent ability to correctly classify buyers and non-buyers. As shown in Table 4.4, demographic variables were not able to classify the buyers and the non-buyers for any of the twelve product categories with a 75 percent ability. Thus, in this study, the above hypothesis failed for every product category. However, while demographic predictors failed to provide clas- sification ability at the 75 percent rate, Table 4.4 shows that eight of the twelve product categories wgrg_correctly identified as to the buyers of the product, at rates of 60 percent or greater, with four instances where buyers were correctly identified at a rate of 66 percent or greater. This success rate is far greater than mere chance, since the actual proportion of buyers in the sample was less than half of the respondent sample for ten of the twelve categories, with a percentage of buyers as low as 9.5 percent for the product category of cooked poultry, as shown in the last column of Table 4.5. Yet, buyers of 130 cooked poultry were correctly identified, by means of their demographic characteristics, at a rate of 64.7 percent. Overall, the lowest success rate for classifying both buyers and non-buyers was for the product of breakfast-baked goods, with 55.9 percent of sample respondents correctly classified, for a product category where only 24.8 percent of the sample purchased any of the product. Moreover, 62.4 percent of the buyers were correctly identified. Fruit juices showed the highest classification rate overall, with 64.8 percent of buyers and non-buyers correctly classified, and also showed a 66.6 percent rate of success in correctly classifying the buyers of this product category, for a category where 64.2 percent of respondents were known to be buyers. Similarly, the buyers of frozen pizzas were correctly identified and classified as buyers in 66.5 percent of the instances in which survey respondents actually reported the purchase of at least one pizza, even though only 26.2 percent of the sample were actually buyers. On an average, demographic variables were able to correctly classify all consumers as buyers or non-buyers with a success rate of 60.5 percent for the twelve frozen food product categories, and with a 61.3 percent ability to correctly classify product buyers. Thus, while demographics failed to provide the 75 percent rate of classification that was hypothesized and hoped for in this study, the average 61.3 percent ability to correctly identify buyers does seem to be a g§§fgl_segmentation technique, even if it failed to reach the "highly meaningful," 75 percent classification ability. 131 Statistical predictive ability of the discriminant analysis. As shown in Table 4.5, the canonical correlation from the multivariate use of demographic variables was relatively low, averaging .220 for the twelve product categories examined. When squared, this figure indicates an average explanation of the observed purchases of frozen foods of 4.84 percent on a linear basis. The highest canonical correlation was .317, for seafoods, giving a net predictive value of 10.05 percent. The lowest canonical corre- lation was for uncooked poultry and fruit or berries, both categories showing a .176 correlation to the demographic variables, for a net predictive value of 3.1 percent, a value which is not surprising, since these categories showed few statistically significant rela- tionships to the demographics in the bivariate stage of this analysis. Implications and the meaning of this low observed canonical correlation will be discussed later, in Chapter V. However, it should be noted that the net predictive value of the demographics is limited to the ligggr_variation due to demographics, even though Table 4.3 suggests that much of the observed, predictive value of demographic variables in this study was actually non-linear. Moreover, the net predictive value, as measured by the linear, canonical correlation, of all demographic variables used in the study was higher than the total of any one demographic variable, even on a non-linear basis. Thus, even with admitted problems of linearity, the multivariate discriminant analysis gave an improved ability to predict the con- sumption of frozen food products, compared to the use of any single demographic factor. 132 The Value of Demographic Variables Compared to Product-Purchase Variables as Predictors of Frozen Food Purchases First of all, in this section of the study, analyses are made of the bivariate predictive power of using product purchasing infor- mation to predict the purchase of individual, frozen food categories, in comparison to the predictive power of demographic variables utilized for the same function. A test of statistical significance of the rela- tionship between the purchase of a given frozen food product category and the observed dependent relationship to the purchase of the other frozen food categories of the study represents the first stage in this analysis, to determine if such relationships exist. Second, the bivariate predictive power is calculated for any product-to-product relationships that are proven to be statistically significant. ' Third, a comparison is made between the frequency and the strength of predictive relationships that exist between product purchases in relationship to other product purchases, in comparison to the frequency and strength of the relationships found to exist between demographic variables and the purchase of the product categories. Fourth, the multivariate ability of product-purchase variables to predict and classify the purchase and non-purchase of specific products is determined in relationship to each of the twelve frozen food product categories of this study. 133 Finally, the multivariate ability to predict and classify the purchase and non-purchase of each of the specific frozen food product categories, by means of purchasing information about other frozen food product categories, is compared to the equivalent ability of demographic variables to predict and classify the purchase and non-purchase of the twelve frozen food product categories of this study. Analysis of Variance-~Test of Significance Utilizing the same analysis of variance computer program that was used to test the demographic variables, the F-test is used to determine if group purchasing means for each of the specific frozen food product categories is significantly different between the pur- chasers and the non-purchasers of the other eleven product categories of the study. Following the same significance rule that was applied to the significance testing of the demographic variables, an .01 level of significance is required to regard the relationship of any product-purchase category as being statistically significant to the tested, dependent product category. The results of this series of 132 F-tests is presented in Table 4.6 (twelve product categories times eleven product categories), as the first reported value for each relationship pair. As was done in reporting the relationship significance for the demographic vari- ables in Tables 4.1 and 4.3, relationships up to the .05 level are reported, but are not regarded as significant, according to the significance standard of this study, beyond the .01 level. 134 Table 4.6 The Significance and Measured Strength of Product-Purchase Variables Used to Predict Other Purchases Frozen Food Categories m .1"; ‘6 m 8 tag .g .. a .. .2 s t .. e t’ 2 so at s a '= °a a: a .8'0 013 O u o» H u 4» us- an pm Frozen Foods as 03 33 2;, §-; '5'; 3 8. "' '5 i: 3 g2 5‘; Independent Variables £3 33 ,3 g: 88 8 g E :8 E US £8 Breakfast-baked goods --- 000 .000 000 .000 .000 .000 .000 .000 .000 .000 .000 139 .035 018 .040 .051 .020 .043 .057 .047 .061 .022 Dessert-baked goods .000 --- .000 .000 .000 .000 .000 .000 .000 .000 .000 .000 .139 .044 .044 .037 .043 .032 .047 .076 .024 .042 .036 Seafoods .000 .000 -—- .000 .000 .000 .000 .000 .000 .000 .019 .000 .035 .044 .045 .018 .081 .051 .058 .037 .057 .008 .032 Uncooked poultry .000 .000 .000 --- .000 .001 .002 .005 .000 .000 .000 .000 .018 .044 .045 .014 .014 .014 .011 .022 .029 .030 .018 Cooked poultry .000 .000 .000 .000 —-- .000 .000 .013 .000 .000 .000 .000 .040 .037 .018 .104 .021 .019 .009 .056 .048 .050 .063 Potatoes .000 .000 .000 .001 .000 --- .000 .000 NSa .000 .001 .000 .051 .043 .081 .014 .021 .050 .057 .042 .015 .019 Vegetables .000 .000 .000 .002 .000 .000 --- .000 .000 .000 .000 .000 .020 .032 .055 .014 .019 .050 .063 .034 .019 .020 .037 Fruit juices .000 .000 .000 .004 .013 .000 .000 --- .000 .000 .044 NS .043 .047 .058 .011 .009 .057 .063 .029 .022 .006 Fruit or berries .000 .000 .000 .000 .000 NS .000 .000 --- .000 .008 .002 .057 .077 .037 .022 .057 .034 .029 .018 .010 .013 Pizza .000 .000 .000 .000 .000 .000 .000 .000 .000 --— .000 .000 .047 .024 .057 .029 .048 .042 .019 .022 .018 .041 .073 Complete dinners .000 .000 .019 .000 .000 .001 .000 .044 .008 .000 --- .000 .061 .042 .008 .030 .050 .015 .020 .006 .010 .041 .116 Main courses .000 .000 .000 .000 .000 .000 .000 NS .002 .000 .000 --- .022 .036 .032 .018 .063 .019 .037 .013 .073 .116 aSignificance levels less than the .05 level are reported as NS. The first reported figure for each variable refers to the significance level of the analysis of variance. 135 As may be seen in Table 4.7, most product-purchase variables prove to be highly significant to the purchase of other product-purchase variables of this study. Only 9 of the 132 tested pairs of product purchase to product purchase relationships did not meet the .01 sig- nificance level. Only four relationship pairs did not show significance at least at the .05 level. In fact, rather amazingly, 110 of the 132 bivariate relationship pairs tested were significant at the .000 level, with many of the remaining relationships proving significant at the .001 or .002 level. Breakfast-baked goods, dessert-baked goods, and pizza all proved to have relationships at the .000 level, relative to the purchase < of each of the other eleven frozen product categories. The average product purchase variable, used as a predictor of the purchase of the other product categories, was able to show a .01 significance of rela- tionship to ten of the eleven other products in the study, with an average of 9.25 of these bivariate relationships at the .000 level of significance. Thus, a basic conclusion seems apparent, that the buyers of one frozen food product tend to buy other frozen food products, with a strength of statistical relationship that is far greater than could reasonably be attributed to chance distribution of purchases within the total survey sample. 136 Analysis of Variance—-Strength of Significant Relationships Next, again using the analysis of variance subprogram "Breakdown" from the Statistical Package for the Social Sciences,“ the bivariate predictive strength is calculated for all product-to-product relationships that were tested significant in the F-test. The Eta squared statistic is calculated for each significant pair with a minimum of .05 significance, although a .01 level is required by the significance standard of this study. As can be seen in Table 4.6, relationships with a significance level less than the .01 level never displayed an Eta squared, predictive value as great as 1 percent of the observed variation in purchasing patterns of the dependent product category, thus providing verification for the appropriateness of the .01 significance rule within this study. As noted previously in this chapter, Eta squared provides a measure of the total, non-linear predictive value of the difference in the group purchasing means in the dependent product category, that can be explained via differences in the independent, predictive vari- able, in this case the purchase of another frozen food product category. A test of linearity was not necessary for these relationships, since all product variables were coded into two categories, of purchase of any quantity of the product versus non-purchase of the product class. Thus, by definition, the relationship strength measured, as coded, can only be linear. The Eta squared value for each significant pair is reported as the second figure for each relationship pair, in Table 4.6. As may 137 be noted, each relationship value is reported twice within the table, since each product category is tested as both a dependent and an independent variable in relationship to each of the other product categories. Summary of the Product-to-Product Relationship Strengths The strongest relationship shown in Table 4.6, was the link between breakfast-baked goods and dessert-baked goods, with an Eta squared score of 13.9 percent of variation in one of these product categories explained by the predictive variation in consumption in the other product within this relationship pair. Another strong relationship was noted between frozen main courses and frozen complete dinners, where an 11.6 percent predictive relationship was found. Only one other product-to-product relationship reported as much as a 10.0 percent predictive Eta squared score. As might have been expected, cooked poultry showed a 10.4 percent relationship to uncooked poultry. Utilizing the definition of a .05 Eta squared score being required to determine the existence of a meaningful predictive strength of relationship, there were a total of 17 product-to-product relation- ships that showed a predictive strength of .05 or greater, among the 66 exclusive relationship pairs that were tested in the study. Several relationships can be noted as hovering slightly above or below this .05 standard. 138 Another fact that can be observed within the results shown in Table 4.6 is that no one product variable provides a meaningful strength of relationship to all the other product categories. Thus, it seems necessary to know the specific predictive relationship of a given frozen food product category to another specific frozen food category, rather than assuming that the knowledge of any given frozen product can be used to provide a meaningful degree of predictive power to any other frozen food product. A Comparison of Demographic Versus Product Variables as Bivariate Predictors of the Purchase of Frozen Food Products Based upon the results of the analysis of variance, as presented in Tables 4.3 and 4.6, product purchase data, for the specific purchases of the frozen food product categories, proved to be a generally more effective predictor of the purchase of any given frozen food product category, as compared to the bivariate predictive strength of the demographic variables of this study. Comparing the results of the significance testing, via the F-test, product purchase variables used as predictors were notable as significant more frequently, and with a typically higher degree of statistical significance. For the product variables examined, significance levels of .000 were observed in 55 of the 66 separate product relationships tested, while Table 4.2 shows that only 32 of the 156 demographic-to-product relationships were able to meet the .01 significance level, and Table 4.3 shows that only 13 demographic relationships were found at the .000 level. 139 Again comparing the results of Tables 4.3 and 4.6, the Eta squared scores for the product variables as predictors are consistently higher than for the average demographic variables, and produced higher individual relationships, with three product-paired relationships over 10 percent, and with 17 relationships noted at the 5 percent predictive value defined as constituting a meaningful predictive relationship on a bivariate basis. By contrast, within the twelve product categories examined, relative to the thirteen demographic variables tested as predictors, the highest Eta squared predictive score was 7.7 percent, and only seven relationships were noted as giving a predictive value of 5 percent: or greater, from the 156 demographic-to-product relationships examined. Thus, while some demographic relationships did prove to be meaningful predictors at the 5 percent predictive level, and were sometimes more predictive for specific products than some of the product-predictive relationships, the general conclusion from the bivariate analysis of variance is that product-purchase variables usually were more effective as predictors of specific purchases of frozen food products than were the demographic variables tested in this study. Thus, the hypothesis, that demographic variables would prove to be equal to or more effective than the use of product-purchase variables as predictors of frozen food purchases, must generally be considered as not true, based upon the results of this study. 140 Discriminant Analysis: Multivariate Ability to Predict and Classify Buyers, Based Upon Product- Purchase Variables The next step in examining the predictive value of the product-purchase variables as independent variables is to apply the linear, multivariate analysis provided by discriminant analysis, as provided by subprogram "Discriminant" of the Statistical Package for the Social Sciences,5 in the same way that this method was applied to the multivariate use of the demographic variables, earlier in this chapter. All product variables, except for the specific product category being examined, are used to jointly predict and classify individual purchasers, relative to their purchase of a specific frozen food product category. The results of this discriminant analysis are shown in Tables 4.7 and 4.8. Table 4.7 shows the percentage of purchasers who were correctly classified by means of the product-purchase variables. The first column on this table presents the total percentage of both buyers and non-buyers of a specific frozen food product category who were correctly classified. The second column gives the percentage of buyers who were correctly classified as buyers of a specific frozen food product category, based upon knowledge of other frozen food categories purchased. Table 4.8 presents the canonical correlation of the combined, multivariate use of the product-purchase predictive variables, a value which can be generally interpreted as a multivariate R correlation. 141 Table 4.7 Total Classification Efficiency of the Discriminant Analysis, Utilizing the Product-Purchase Variables Percentage of Percentage of All Consumers Buyers Frozen Food Category Correctly Classified Correctly Classified :7. z Breakfast-baked goods 74.4 66.3 Dessert-baked goods 72.7 65.3 Seafoods 70.0 68.7 Uncooked poultry 76.6 54.9 Cooked poultry 82.7 63.5 Potatoes 68.4 67.9 Vegetables 67.8 69.0 Fruit juices 66.4 67.4 Fruit or berries 73.2 60.5 Pizza 69.9 63.3 Complete dinners 71.4 64.0 Main courses 72.0 64.6 Average of all products 72.1 65.5 142 Table 4.8 Statistical Strength and Distribution of the Discriminant Analysis Using the Product-Purchase Variables Percentage of Non-Buyers Percentage Canonical Incorrectly of Buyers Frozen Food Category Correlation Classified in Sample :72 2. Breakfast-baked goods .474 23.0 24.8 Dessert-baked goods .481 24.5 27.7 Seafoods .441 28.9 44.0 Uncooked poultry .395 19.0 17.0 Cooked poultry .453 16.3 9.5 Potatoes .414 31.2 45.1 Vegetables .382 35.0 70.2 Fruit juices .386 34.6 64.2 Fruit or berries .398 24.4 16.6 Pizza .404 27.7 26.2 Complete dinners .428 24.5 27.9 Main courses .449 25.2 27.2 Average of all products .425 26.2 33.3 143 This value, when squared, presents the total, linear predictive value of the product-purchase variables, relative to the observed variation in the purchase of the specific frozen food product categories. The second column presented in Table 4.8 is the percentage of non-buyers who were incorrectly predicted and classified as being buyers. Finally, the third column presents again the actual percentage of the survey sample who actually did report purchasing at least one package of the specific frozen food product category during the 30 days prior to the survey. Classification of the discriminant analysis, using the product- purchase variables as predictors. As defined in Chapter I, the standard _ for determining the existence of a meaningful multivariate predictive relationship was to demonstrate at least a 75 percent ability to cor- rectly classify the buyers and non-buyers of the specific frozen food product categories examined in this study. This section of the study will examine the ability of the knowledge about consumer purchasers in eleven frozen food categories to predict the purchase of frozen products in another specific purchase category. As shown in Table 4.7, product-purchase variables were able to correctly classify buyers and non-buyers into the correct categories, at the 75 percent level of success, for two of the twelve product categories examined, uncooked poultry and cooked poultry, with correct classifications of 76.6 percent and 82.7 percent, respectively. In addition, breakfast-baked goods was almost able to meet the 75 percent criteria level, with a total classification success rate of 74.4 percent. 144 The average classification success, for both buyers and non-buyers was 72.1 percent for the twelve product categories, with the high of 82.7 percent for cooked poultry, and a low of 66.4 percent for fruit juices. The ability to correctly classify the buyers of the specific frozen food categories was somewhat less effective, with an average classification success of only 65.5 percent, with the greatest success of 73.5 percent for cooked poultry, and the poorest success for uncooked poultry, with a 54.9 percent correct classification rate. In all twelve product categories, the knowledge of the other product purchases of the respondent was not able to provide correct classification of the buyers at the rate of 75 percent, as was hypothesized in this study. However, despite the general failure to provide correct classification at the 75 percent rate, knowledge of other frozen food purchases did seem to provide classification success for specific frozen food product categories that could be g§gf§l_to the marketer of frozen food products. The implication of this observed classification success will be discussed in Chapter V. Statistical predictive ability of the product1purchase variables. As shown in Table 4.8, the multivariate ability of the product-purchase variables to explain the variation in specific product category buying was substantially above any of the bivariate predictive ability of any one product-purchase variable, as shown in Table 4.6. The highest bivariate relationship shown in Table 4.6 was 13.9 percent, with only three bivariate relationships proven to provide predictive value greater than 10.0 percent. However, the average canonical 145 correlation for the twelve product categories was .425, which when squared, indicates a total linear, statistical predictive value of 18.06 percent. The highest canonical correlation was noted for the product category of dessert-baked goods, at .481, indicating an explanation of the observed variance in purchases of 23.14 percent. The poorest canonical correlation was noted for frozen vegetables, with a score of .382, representing a 14.60 percent explanatory power from the knowledge of the purchases in the other eleven frozen food product categories. Comparison of demographic versuslproduct variables as predictive - variables in the multivariate discriminant analysis. Based upon the results of the multivariate discriminant analysis, as presented in Tables 4.4 and 4.5 for the demographic variables, and in Tables 4.7 and 4.8 for the product-purchase variables, the product-purchase vari- ables proved to be generally more effective in the classification and prediction of purchases of the specific frozen food products categories. Comparing the results of the classifications, product-purchase variables were able to classify buyers and non-buyers at an average rate of 72.1 percent, and correctly classified product buyers at an average rate of 65.5 percent. Non-buyers were incorrectly identified as buyers at an average rate of 26.2 percent. An average, multivariate canonical correlation of .425 was obtained by the use of product- purchase variables, representing a statistical predictive power of 18.06 percent. 146 Alternatively, the demographic variables were able to correctly classify both buyers and non-buyers at an average rate of 60.5 percent, and were able to correctly classify the buyers with an average success of 61.3 percent. An average canonical correlation was obtained of only .220 from the demographic variables, giving a net predictive statistical efficiency of only 4.84 percent, with the highest reported predictive value of only 10.05 percent, well below the average 18.06 percent predictive value of the product-purchase variables. Non-buyers were incorrectly classified as buyers at an average rate of 40.9 percent by the use of demographic variables, compared to a 26.2 percent rate by the product-purchase variables. Thus, on an average basis, the hypothesis, that demographic variables would provide a multi-variate ability to classify and predict that was equal to or greater than the ability shown by product-purchase variables, must be regarded as not true, based upon the results of this study. However, it may be noted that demographic variables did exhibit better results for specific product categories than did the product- purchase variables. For example, demographic variables correctly classified the buyers of uncooked poultry at a rate of 57.4 percent, compared to 54.9 percent by means of product-purchase variables. Pizza buyers were correctly classified by demographics at a rate of 66.5 percent, while the product-purchase variables were correct only 63.3 percent. Seafoods, potatoes, vegetables, and pizza buyers were correctly classified at rates of over 66 percent by means of 147 demographics, success rates higher than the average showed by the product-purchase variables, and very close to the classification success demonstrated for these product categories by the product- purchase variables. Moreover, it should be noted that the relatively close classification efficiency of buyers, 61.3 percent by demographic variables, on an average, compared to 65.5 percent by means of the product-purchase variables, must raise certain questions and cautions before coming to a firm conclusion as to which set of predictive variables is the best or most useful basis for the practical segmen- tation of the buyers and non-buyers of frozen food products in actual marketing practice. Additionally, beyond the substantial issue of the availability of demographic data compared to specific product-purchase data in the marketplace, substantial questions must be raised about the relatively modest sample size which this study is based upon, especially when the distribution of buyers and non-buyers, as low as 9.5 percent for one product category, or 68 respondents, limit the statistical certainty of the differences noted in the classification efficiencies. Moreover, as noted previously, demographic variables often exhibited non-linear effects, thus limiting the ability to classify by means of linear, multivariate methods. These issues will be discussed, relative to implications and conclusions from this study, in Chapter V. 148 The Value of Total Food Purchases as a Predictor of Specific Frozen Food Product Categories It was hypothesized that both demographic variables and specific frozen food purchase variables would prove to be more effective than the total food purchases by the consumer respondents, relative to the prediction of the purchase of specific frozen food categories. As shown in Table 4.9, total food purchases, as a predictor variable, tended to be a relatively modest predictor of the purchases of the specific frozen food categories of this study. Within Table 4.9, the first value indicated for each product category is the result of the F-test of significance, via the same analysis of variance program that was used to examine the effect of the demographic and the product-purchase variables on a bivariate basis. Significance levels are reported up to the level of .05, but the defined level of significance, for purposes of this study, to require a statistically significant relationship at the .01 level. Within Table 4.9 it can be noted that only four of the twelve product categories exhibited a relationship that was significant at the .01 level, with two more categories showing weaker relationships within the .05 significance level. The second value presented for each product category, if a minimum of a .05 significance level was exhibited, is the Eta squared value for the relationship, representing the total, non-linear pre- dictive value of knowing total food purchases, relative to the ability to predict or explain the variation in the purchases of the specific frozen food categories. 149 Table 4.9 The Predictive Effect of Total Food Purchases and Home Freezer Ownership on Frozen Food Purchases Predictive Variables Dependent Variables Total Food Ownership of a Frozen Food Category Purchases Separate Freezer Breakfast-baked goods .000 NS .056 .031 Dessert-baked goods .037 NS .028 .011 Seafoods NS NS Uncooked poultry NS NS Cooked poultry NS NS Potatoes .019 NS .031 .023 Vegetables .001 NS .044 .017 Fruit juices .001 NS .042 .022 Fruit or berries NS NS Pizza .001 NS .044 .027 Complete dinners NS NS Main courses NS NS 150 As Table 4.9 shows, the highest Eta squared value found was in relationship to the purchase of breakfast-baked goods, with a value of 5.7 percent. Yet, dessert-baked goods did not exhibit an .01 level of significance, and only showed a 2.8 percent Eta squared value, with a significance of .037. Frozen vegetables and pizza showed a 4.4 per- cent predictive ability, as measured by the Eta squared score, and a score of 4.2 percent was exhibited, relative to the purchase of fruit juices. Frozen potatoes also failed to meet the .01 significance test, with a reported .019 significance level, but showed a 3.1 percent predictive value of Eta squared. No other relationships were able to meet even the less significant F-test level of .05. The third value reported for significant purchase relationships to total food purchases is the R squared value, which measures the linear predictive value of knowing the consumer's food purchases. As can be seen in Table 4.9, there was substantial variation, and loss of predictive power, by the use of the linear R squared measurement compared to the non-linear Eta squared measurement. Among the four relationships significant at the .01 level, an average non-linear predictive score of .046 was noted, but only a .024 average linear score, an approximately 48 percent loss in predictive value. Thus, among the relationships that were significant, the relationship was relatively non-linear. By comparison with Tables 4.3 and 4.6, the value of total food purchases as a predictor of specific frozen food purchases, compared to the use of demographic variables, or the use of specific product- purchase variables, would not seem to be very effective. 151 Specific demographic variables often exhibited stronger predictive relationships to specific frozen food product categories. Moreover, total food purchases exhibited relationships at the .01 level to only four frozen food categories, and only two more at the .05 level. By comparison, demographic variables exhibited .Ol relationships to nine of the twelve frozen food product categories, and a total of eleven categories showed at least a .05 level of significance to at least one of the demographic variables tested. The product-purchase variables exhibited an average of 9.25 percent relationships at the .000 level, between a specific frozen food category and the other eleven categories used as predictive variables. An average of over ten categories was significant at the .01 level, with most relationships providing predictive values that were higher than the few reported predictive relationships based upon total food purchases. Thus, total food purchases did not seem to be as effective as either demographic variables or specific product-purchase variables in predicting the purchase of specific frozen food product categories, thus supporting the hypothesis about the relative effectiveness of total food purchases as a predictor of frozen food purchases. The Value of Freezer Ownership as a Predictor of the Purchase of Specific Frozen Food Product Categories Although every reported study of frozen food purchases has asked about the existence of a separate freezer unit in the household, none of these studies has reported any test of the significance or the 152 predictive value of this relationship, if any relationship exists. In fact, since this question always seems to be part of past frozen food studies, the assumption by past researchers seems to be that a relationship is present between freezer ownership and the purchase or consumption of frozen foods. The hypothesis of this study was that freezer ownership would not provide a statistically significant or meaningful predictive rela- tionship to the specific frozen food product categories of this study. As shown in Table 4.9, the ownership of a separate freezer unit did not exhibit a single statistically significant relationship, at either the .01 or the .05 level, to any of the specific frozen food categories,» thus conclusively proving the hypothesis of this study about the predictive value of freezer ownership; there was none. CHAPTER IV--FOOTNOTES 1Norman H. Nie, C. Hull, J. Jenkins, K. Steinbrenner, and 0. Brent, Statistical Package for the Social Sciences (New York: McGraw-Hill Book Co., 1975), pp. 249-275. 2Ibid. 3Ibid., pp. 434-467. “Ibid., pp. 249-275. 5Ibid., pp. 434-467. 153 CHAPTER V CONCLUSIONS AND IMPLICATIONS Introduction The purpose of this chapter is to present the conclusions and the implications of this study. The first section reports the overall conclusions about the research hypotheses of the study, based upon the empirical research findings presented in the previous chapter. The second section discusses the research implications of the study for further use of demographic variables as a basis for market segmentation. Next, the third section presents the management implications of the findings. Finally, the fourth section offers some suggestions for further research. Conclusions The Use of Demographic Variables as Purchase Predictors The major hypothesis of this study was that demographic variables can be used to successfully predict and segment the purchasing behavior for consumer products. Specifically, the research of this study focused upon the ability of demographic variables to predict and segment the purchasing behavior for twelve specific categories of frozen food products, as a means of testing this major hypothesis. 154 155 During the course of testing this major hypothesis, four subhypotheses were considered, relative to the degree that demographic variables could be tested and proven effective in predicting product purchases. These four subhypotheses were as follows: a. There will be a statistically significant, bivariate relationship between the individual demographic variables and the purchases of specific frozen food product categories at the .01 significance level. b. There will be a bivariate predictive relationship between individual demographic variables and the purchases of specific frozen food product categories with a total non—linear predictive power of 5 percent or more. c. There will be a linear predictive relationship for demographic variables that are significantly related to the purchase of specific frozen food products on a bivariate basis. d. There will be a multivariate predictive relationship between the total set of demographic variables and the purchases of specific frozen food product categories with the resultant ability to correctly classify 75 percent or more of the product purchasers and non-purchasers. Conclusions from the empirical testing of each of these four subhypotheses will be presented in order, followed by an overall conclusion about the ability of demographic variables to successfully predict and segment the purchases of the frozen food categories of this study. Statistical significance of the bivariate relationships. As noted above, the first subhypothesis of the major research hypothesis of this study considers the statistical significance of the bivariate relationships between the individual demographic variables and the specific frozen food categories of this study, to determine if 156 relationships exist beyond what could reasonably be attributed to chance variation in sampling. A total of 156 bivariate relationships were examined (thirteen demographic variables times twelve frozen food categories) in an analysis of variance, via the F-test. The basic conclusion from this series of statistical tests is that significant relationships gg.gxi§t_between §gmg_individual demographic variables and the purchase of specific frozen food product categories, thus generally proving this subhypothesis of the study. Altogether, 32 relationships were found to be significant at the .01 level, and a total of 56 relationships were noted at a more liberal .05 significance level. However, it should be noted that the existence of significant relationships varied substantially between specific product categories, both in the number of demographic variables which exhibited a significant relationship, and also in terms of which particular demographic variables were shown to be related. No single demographic factor proved to be significantly related at either the .01 or the .05 level to all of the twelve tested product categories. In fact, the maximum number of products significantly related to a single demographic variable was only five, exhibited by both the variables of age and the number of persons in the household. Moreover, the demographic factors were more likely to show statistical relationship to certain product categories, rather than others. Thus, at the .05 level, ten of the thirteen demographic variables exhibited significant relationships to the purchase of seafoods, while not one demographic variable exhibited this level of significance in relation to the purchase of frozen fruits or berries. 157 At the hypothesized .Ol significance level, demographic variables exhibited at least one significant relationship to nine of the twelve tested frozen product categories. At the more liberal .05 significance level, demographic variables proved to have at least one significant relationship to eleven of the twelve product categories. Predictive power of the bivariate relationships. The second subhypothesis regarding the use of demographic variables as purchase indicators was that there would be a bivariate predictive relationship between individual demographic variables and the purchase of specific frozen food product categories, with a total, non-linear predictive power of 5 percent or more. All relationships were examined that exhibited at least an .05 significance level. Of the 156 bivariate relationships that were originally examined by this study, only six instances were found where a demographic variable was able to provide at least a 5 percent predictive Eta squared value for the purchase of a given category of frozen foods. Thus, the general conclusion from this study was that on a bivariate basis, most demographic relationships that were tested, even though statistically significant, did not provide meaningful statistical predictive strength, even at the seemingly modest standard of providing a 5 percent or greater predictive power. In fact, study findings show that relationships which failed to meet the hypothesized significance standard of .01 on the F-test rarely provided predictive strength as high as 1 percent, relative to the observed variance in purchases of the tested product categories, 158 thus providing empirical justification for adopting the relatively strict .Ol significance standard. Thus, while a few instances of meaningful demographic, bivariate relationships wgrg_found, this subhypothesis must generally be regarded as having failed. However, sufficient bivariate predictive power was found to provide some potential for the multivariate prediction of product purchases, to follow in this study. Linearity of the bivariate relationships. Most prior studies of the relationship of demographic variables to product purchases have made the basic assumption that the predictive value of such relation- ships was substantially linear, both for the sake of simplicity in research, and also to allow the use of many statistical techniques that require the existence (or assumption) of a linear relationship. In fact, the multivariate method to be used in this study requires such an assumption, since it considers only the linear predictive powers of the independent variables in calculating multivariate predictive efficiency. The third subhypothesis of this study, regarding the use of demographic variables as product purchase predictors, was that the predictive power of the demographic variables was indeed linear, thus testing this common, but typically unverified research assumption. Therefore, the linear R squared predictive value of each significant relationship was calculated and compared to the total non-linear Eta squared value for the relationship. To the degree that R squared is less than Eta squared, predictive power is lost from the assumption of a linear relationship. 'l 159 The general conclusion of this test was that many of the relationships between demographic variables and the purchase of the tested frozen food categories were substantially non-linear. The result was that half or more of the total, predictive value of a demographic variable was often lost by use of the linear R squared measurement, compared to the non-linear Eta squared measurement of predictive value. For example, the number of persons in the household was shown to have a 4.9 percent predictive Eta squared value relative to the purchase of seafoods, and a 5.0 percent value relative to the purchase of frozen fruit juices. Yet, the R squared values were only 2.3 percent and 2.0 percent, respectively, thus indicating a net pre- dictive loss of about 60 percent of the non-linear value. Similarly, income gave a 4.3 percent non-linear value for the purchase of seafoods, but showed only a 1.2 percent linear value. A basic conclusion, then, is that the assumption of linearity is often a significant cause of loss of predictive power for most of the tested bivariate relationships in this study that exhibited a significant statistical relationship, thus generally disproving this subhypothesis of the study. A resultant conclusion would be that non- linear statistical methods would increase the net predictive power of demographic variables, and should be employed when possible. Unfortunately, the availability of appropriate non-linear multivariate techniques is very limited, and often requires much larger sample sizes to be statistically appropriate, limitations that both apply to this study. The issue of linearity will be discussed later 160 in this section, as a limitation affecting the multivariate portion of this study, and also will be discussed as an implication of this study, later in this chapter. The immediate solution for this study was to recode some of the response categories for the demographic variables which exhibited the most drastic loss of predictive power by the assumption of linear- ity, by combining categories with similar purchase means. The result is loss of detail, as well as some loss of predictive efficiency, with the remaining predictive values continuing to be less than true, non- linear predictive values. Thus, unfortunately, the linear, multivariate discriminant analysis to follow in this study must be considered flawed. by the relative failure to include the total bivariate predictive values of the demographic variables. Multivariate classification oflproduct purchasers. The fourth subhypothesis about the use of demographic variables as product purchase predictors was that the multivariate use of the demographic variables would provide a multivariate ability to predict product purchases of the specific frozen food categories of this study, with the resultant ability to correctly classify 75 percent or more of the product purchasers and non-purchasers. The results of this test showed that the total set of demographic variables was not able to classify the purchasers and the non-purchasers for any_of the twelve frozen food product cate- gories with a 75 percent rate of success. Thus, by the standards of the subhypothesis, the empirical results of the discriminant analysis failed to support the subhypothesis. 161 However, while demographic variables failed to provide the defined, highly meaningful classification ability of 75 percent, the research findings did show that eight of the twelve product categories wgrg_correctly classified, as to the purchasers of the category, at rates of 60 percent or greater, with four instances of a 66 percent ability to correctly distinguish category purchasers. Buyers were correctly identified at an average rate of 61.3 percent, and all respondents were correctly identified as buyers or non-buyers at an average rate of 60.5 percent. Thus, while demographics failed to provide the 75 percent or greater rate of classification accuracy that was hypothesized and hoped . for in this study, the classification rates that were obtained seems to provide at least a useful ability to distinguish product purchasers from non-purchasers, based upon knowledge of the purchasers' demographic characteristics. Another conclusion that might be inferred from the demonstrated classification rates, combined with the relative failure of the sub- hypothesis relating to linearity of the demographic predictors, would be that the ability to include the total, non-linear predictive values of the demographic variables might allow total classification efficiency that could reasonably be projected to at least come closer to the hypothesized rate of classification, based upon the proportional differences between the linear and the non-linear bivariate predictive values. 162 The square of the canonical correlation can be used to indicate a predictive efficiency that is equivalent to a multivariate R squared. By this measurement, the predictive value from the use of the demo- graphic characteristics remained low, averaging 4.84 percent predictive ability for the twelve tested product purchase categories, with a high predictive score of 10.05 percent relative to the multivariate, linear predictive value for the purchases of seafoods. However, even with admitted problems of linearity, the multivariate linear value for each of the product categories was_higher than any one demographic variable could offer per product on a total non-linear basis. Thus, even though the lack of linearity limits the total accuracy of the linear discriminant analysis technique, it still provided better predictive totals than the non-linear measurement on a bivariate basis. Moreover, the 61.3 percent average ability to identify the purchasers of specific product categories, when the average proportion of purchasers was only 33 percent of the survey sample, indicates that the multivariate classification functions derived from the demographic characteristics of the survey sample gave a resultant ability to identify and classify product purchasers that was somewhat higher than the 4.84 percent statistical predictive value might otherwise suggest. Summary--The use of demographic variables as purchase predictors. Based upon the findings and conclusions from the testing of the four sub-hypotheses relating to the use of demographic variables as purchase predictors, the overall conclusion would be that demographic 163 variables gan_be used to successfully predict and segment the purchasing behavior for consumer products. A number of statistically significant relationships were found, but it should be noted that such relationships varied greatly between product categories, thus indicating that the broad category of total frozen food purchases, as used by several previous studies of the frozen food purchaser, may be too heterogeneous to be regarded as a distinct product category. No single demographic factor was significantly related to more than five of the twelve frozen food categories tested as dependent variables of this study. Demographic variables generally failed to provide a meaningful predictive power of 5 percent or greater on a bivariate basis, but despite the relative lack of linearity, were still able to provide sufficient classification power in a linear discriminant analysis to correctly identify and classify the purchasers within the twelve frozen food categories at an average success rate of 61.3 percent. While this classification rate was less than hypothesized or hoped for, it would still appear to be evidence of a useful ability to identify the purchasers of these product categories. Finally, it was concluded that the use of non-linear methods would be likely to increase the success of classification. 164 A Comparison of Demographic Variables to Product-Purchase Variables as Predictors of Frozen Food Product Purchases The second hypothesis of this study was that demographic variables would prove to be equal to or more effective than the use of product-purchase variables, in the ability to predict the purchases of frozen food product categories, and to classify them into segments of purchasers and non-purchasers of each specific frozen food category. For purposes of testing this hypothesis, statistical tests were performed that were parallel to the tests used to examine the use of demographic variables as predictors, with the results compared on both a bivariate and a multivariate basis. Purchases for each specific product category were employed as independent, predictive variables, relative to the purchase of the other product categories used in this study, testing a total of 66 product-to-product relationships. The overall conclusion is that this hypothesis failed. Product-purchase variables were generally found to be better predictors in all tests, although the predictive advantages were not found to be so great that the use of demographic segmentation techniques should be abandoned as totally inferior. Comparative statistical significance of product-purchase variables. In the tests of statistical significance, the average product-purchase variable was found to be significant at the .01 level with ten of the other eleven product categories, with an average of 9.25 percent of these relationships at the .000 level. Thus, an apparent conclusion is that buyers of one frozen food product tend 165 to buy other frozen food products. Moreover, the proportion of bivariate relationships based upon product-purchase predictors that were found statistically significant at the .01 level, was substantially higher than shown by the demographic predictor variables. Comparative bivariate predictive strength of product-purchase variables. Using the standard of a 5 percent or greater Eta squared value as an indication of meaningful predictive strength, 17 product- to-product relationships gave a meaningful predictive value, compared to only six demographic relationships. The highest Eta squared score by a product purchase variable was 13.9 percent, compared to the highest demographic predictive relationship of 7.7 percent. However, it may be noted that no single product-predictor gave a meaningful predictive value to all the other product categories, thus requiring that specific product-to-product relationships be known, rather than assuming that the knowledge of any frozen product purchase can be used to provide a power of predictive ability that is meaningful to any other product. In addition, specific demographic to product relationships were frequently equivalent or higher than given product- to-product relationships for the same specific frozen food category. The coding form and structure of the product-to-product relationships resulted in an Eta squared measurement that could only give a linear relationship, thus the assumption of linearity was not an issue for the linear, multivariate discriminant analysis using the product-purchase variables. 166 Comparative classification efficiency of productipurchase variables. Product-purchase variables were usually superior to the use of demographic variables for the classification of product pur- chasers and non-purchasers of the specific frozen food product categories of this study, although specific instances were found where demographics proved to be basically equivalent or slightly better for classification purposes. The average classification success rate, for both purchasers and non-purchasers was 72.1 percent for the twelve studied categories, compared to a 60.5 percent average by means of demographic variables. However, in the identification of the actual purchasers, probably the most important factor to an actual marketer of frozen foods, the product-purchase variables gave only a 65.5 percent rate of success in classification of the purchasers, compared to a 61.3 percent rate by means of demographics. This difference could have been caused by basic sampling and/or coding error, as well as by the previously noted loss of predictive power by the demographics, caused by the assumption of linearity for the discriminant analysis. Thus, based upon the sample and the statistical method used by this study, product-purchase variables were found to be generally more effective in multivariate classification of the product purchasers and non-purchasers. However, the caution is made that a different sample of respondents, a larger sample, or the use of non-linear discriminant methods, might all be expected to reduce much of the observed differences in the classification abilities of demographic variables compared to the use of product-purchase variables. 167 Another factor that should be noted in this comparison is that the canonical correlation value of the product-purchase variables was substantially higher than for the demographics. The average multivariate linear predictive value of the product-purchase variables was 18.06 percent, compared to the average of only 4.84 percent by means of demographic variables, thus indicating a clear predictive superiority for the product-purchase variables. Again, it should be noted that the assumption of linearity can be expected to reduce the reported predictive value of the demographic variables, due to sub- stantial non-linearity of the bivariate demographic relationships to the product purchases. The Value of Total Food Purchases as a Predictor of Specific Frozen Food Purchases The third hypothesis of this study was that both the demographic variables and the specific frozen food product purchase variables would prove to be more effective than the total food purchases by the consumer- respondents, relative to the prediction of the purchase of specific frozen food categories. This hypothesis was tested by means of the same bivariate tests of significance and predictive power that were used to examine the use of demographic and product-purchase variables. This hypothesis was basically proven to be true. At the .01 level of significance, total food purchases was significantly related to four frozen food product categories, a frequency of relationship that was matched or battered by several demographic variables, and by all the product-purchase variables. Even at the .05 level, total food purchases exhibited a relationship to only six of the twelve categories, 168 compared to an average product-purchase variable that exhibited an .01 relationship to ten of the other eleven product categories. In terms of bivariate predictive power, the highest strength of relationship was only 5.7 percent, a value that was matched or exceeded by several demographic and product-purchase variables. The average predictive value among the six relationships at the .05 level was 4.1 percent, but it may be noted that the linear portion of this explanatory value declined to 2.2 percent, a net loss of 46 percent of predictive power from the non-linear value. Thus, the use of total food purchases did not seem to offer any clear advantage to the use of either demographic or product- purchase variables, and in most cases, would provide less net predictive value than either of these tested alternatives. The Value of Freezer Ownership as a Predictor of Specific Frozen Food Purchases The fourth and final hypothesis was that freezer ownership would not provide either a statistically significant or meaningful predictive relationship to the purchases of the specific frozen food categories of this study. This hypothesis was also tested by means of the tests of significance and bivariate predictive power, as was used in the prior three hypotheses. The conclusion of this study is that no significant relationship exists between freezer ownership and any of the frozen food product categories at the .01 or the .05 levels of significance, thus, the hypothesis must be regarded as being rather conclusively proven. 169 Since no statistical significance was found to exist, no predictive power was calculated. Research Implications of the Study The Use of Demographic Variables as Segmentation Criteria The major implication of this study is that demographic variables may have the potential, if not the demonstrated ability, to segment product purchasers and non-purchasers, with a useful degree of accuracy. As noted in the conclusions, the use of a linear, multivariate discriminant analysis was handicapped by the fact that much of the predictive power of the demographic variables used in this study was found to be non-linear. Yet, the use of demographic characteristics of the survey sample still resulted in an average ability to distinguish product purchasers, of the specific product categories, that was correct 61.3 percent of the time. This degree of classification success would seem to provide at least a gsgfgl_tool for the marketing researcher or practitioner, even if a higher predictive ability is desirable, for purposes of basic sales projections, as well as media selections relative to the expected demographic characteristics of predicted purchasers. Moreover, as cautioned previously in Chapter I, relative to expected limitations of this study, the set of demographic variables employed in this study was limited, and did not include other factors such as regionality, ethnic background, religion, age, or number of 170 children, or the stage in the family's life cycle; such additional demographic factors might be expected to provide additional variance in product purchases, and additional classification accuracies, by means of demographic variables. Although the demographic factors employed did not prove to provide the same degree of predictive power as exhibited by the use of product-purchase variables, this should not imply that product- purchase variables should be used by future practitioners or researchers as a more desirable or more practical basis for market segmentation, based upon the findings of this study. Rather, as noted by Arndt,1 the evaluation of the total use- fulness of predictor variables for segmentation purposes must be considered relative to the combined operational requirements of identifiability, accessibility, and predictive efficacy. Thus, although the product-purchase variables exhibited a higher degree of predictive ability, expected problems of identifiability and accessibility may severely limit or even preclude the use of the product-purchase variables as a practical, Operational method of market segmentation. The limited ability to identify wh9_does buy the other products, without surveying individuals, rather than representative samples, may leave the marketer with the problem noted previously by Twedt;2 volume buyers may only be identifiable as being volume buyers, thus limiting the ability to operationalize segmentation efforts toward consumers who are defined as being product buyers. Similarly, even if buyers can be identified, accessibility, 171 via media alternatives, can be expected to raise additional problems in reaching the desired segments of the market without virtual blanketed, promotional efforts. The result may well be that a product-purchase based segmentation effort would need to be translated to demographic terms, in order to solve the operational needs of identifiability and accessibility, as well as applying predictive efficacy. Thus, the average ability of the product-purchase based classification to distinguish buyers at an average 65.5 percent rate, compared to a 61.3 percent rate by the demographics, may not prove to be sufficient predictive advantage relative to the use of a demographic based classification of product buyers, when the additional operational . advantages of the demographic variables are considered. The relative closeness of the classification ability of the demographic variables, compared to the use of product-purchase variables, would seem to provide a significant implication, that demographic variables may be the most useful, operational method of distinguishing and marketing to market segments. Moreover, if the use of additional demographic predictor variables can be integrated with the use of non-linear, multivariate techniques, the demographic variables may prove to exhibit predictive efficacy that is very comparable to product-purchase based segmentation methods. Thus, the use of demographic variables should certainly not be ignored or abandoned as potential segmentation criteria, despite several studies, including this one, that report a higher predictive efficacy 172 via the use of product-purchase variables. If the predictive efficacy of demographic variables can achieve values that are reasonably close to the multivariate classification of buyers and non-buyers by means of product-purchase variables, as was done in this study, the impli- cation must be that demographic variables may provide an operationally more useful method of segmentation. Specific Versus General Segmentation A second implication of this study's findings would be that segmentation efforts directed toward specific products are likely to be more accurate than segmentation of general product markets. Using either the demographic or product-purchase variables, it was obvious that associated predictive relationships varied sub- stantially, both in frequency of significant bivariate relationships, as well as in the predictive values. Moreover, the classification success of the multivariate discriminant analysis varied even more dramatically, between the tested specific product categories of this study. Thus, this implication seems to echo the conclusion by Weisenberger,3 who suggested that the more micro-approach of specific product segmentation is more likely to offer an accurate definition of the consumer, compared to attempts at general segmentation of a broad, product class. Therefore, segmentation efforts toward specific product categories, such as frozen seafoods, are more likely to be successful than attempted segmentation of more general categories of products, 173 such as total frozen foods. Similarly, segmentation relationships that are established for a product category such as seafoods, may not provide practical segmentation efforts toward frozen fruit and berries, a product category for which ng_statistically significant demographic relationship was found, among the demographic variables used in this study. Thus, while it may be tempting for a multiproduct, multibrand manufacturer or industry to conduct a single study, trying to establish a single consumer profile for a broad line of products, the results are likely to continue to be less than satisfactory. General categories of measured purchase response can only be expected to achieve, at best, general predictive ability. Relative failure of demographic variables to predict some product categories as well as others might suggest that some categories are too heterogeneous in the range of included products, such that specific relationships for specific products are lost within a relatively diverse category-group of products. For example, cheap TV dinners are likely to appeal to a different demographic profile than Weight Watcher diet dinners. Thus, the limited ability of demographic variables to accurately predict the purchases of frozen complete dinners may relate to a growing diversity of product price, function, prestige, and universality of product use within this combined category of frozen food products. Thus, the moderate degree of classification success of purchasers does not imply that higher levels of classification 174 efficiencies do not exist wilhin_the product categories tested in this study. Similarly, the diversity of both bivariate and multivariate relationships toward the purchase of the specific product categories of this study, on either a demographic or product-purchase variable basis, suggest that segmentation of a typical "frozen food" purchaser would only be accomplished in more general and less accurate terms. Therefore, a broader implication of this study may be that the observed failure of demographic variables in other, past segmentation studies to accurately predict product purchases may be due not to the lack of predictive ability for demographic variables pgr_sg, but rather to the generality of the defined product markets that were utilized as the dependent variable. Statistical Significance Versus Predictive Strength A third basic implication of this study's findings is that many prior studies of demographic variables, as well as those who accept their conclusions, may have fallen into a "semantic trap," by incor- rectly perceiving that a finding of statistical significance can be translated to mean an important or meaningful relationship has been determined to exist, that therefore would be useful to the marketer in explaining the purchase behavior of the consumer. The term, statistical significance, does not directly imply that meaningful predictive strength is present in the relationship, but only that the relationship was shown to exhibit a tie that was beyond normal expectations of random variation that could be expected 175 to result from chance in the sampling process. Thus, even at the .05 level of significance, probably the most popular level of significance for academic research, the observed relationship could have been expected to occur, due to chance, five times in one hundred. Although 56 of the 156 bivariate relationships between demographic variables and specific frozen food categories were shown to be statistically significant at the .05 level, only six of these relationships proved to provide the hypothesized predictive strength of 5 percent or greater, that was regarded as a meaningful strength of relationship. Similarly, the study findings showed that relation- ships which exhibited a significance level higher than .01 rarely had a predictive strength that reached as high as 1 percent. Therefore, the implication of this finding is a simple warning that statistical significance is generally only a first step in the research process, to determine if a relationship is likely to exist beyond expected variations due to chance. Additional statistical analysis is required before assuming that a statistically significant relationship is an important or useful relationship. However, if a statistically significant relationship is ngt_found to be present, the presence of an important, predictive relationship is highly unlikely. Thus, reported levels of statistical significance might be more properly regarded as indicating that some relationships probably do not exist, rather than a direct implication that relationships that meet the established test of statistical significance should be regarded 176 as important for explaining variation in the tested, dependent variable. Linearity of the Demographic Relationships_ The fourth basic implication of this study flows from the finding that a substantial part of the predictive power of the rela- tionship between the demographic variables and product purchases was non-linear. Thus, the use of a linear, statistical measurement of the predictive strength of the demographic variables in this study was shown to reduce the predictive strength of these tested relationships, com- pared to the total, non-linear statistical measurement of these relationships. Beyond the obvious limitations upon the use and interpretation of the linear, multivariate discriminant analysis used in this study, the finding of significant non-linearity also has important implications for past and future studies of the relationship of demographic variables to product purchases. Most reported studies have simply made the assumption that the relationships were linear, without specific testing of the actual rela- tionship linearity. To the extent that the predictive powers of the demographic variables are non-linear, research conclusions based upon linear techniques, such as R squared, linear discriminant analysis, and linear regression, must be considered somewhat suspect. Reported con- clusions about predictive strength or ability must be perceived as conclusions about linear strength of the relationship, not necessarily 177 the actual, total strength that might be derived from a non-linear measurement of the relationship. However, the major implication, from a finding of significant loss in actual predictive strength via linear statistical methods, would be that non-linear methods should be used, when available to the researcher, and when consistent with the size and format of the research data base. This implication will be discussed further, as a suggestion for further research, later in this chapter. Management Implications of the Research Findings Demographics as an Operational Management Technique Implications of this study for marketing management generally parallel the research implications, as discussed in the previous sec- tion. However, it seems appropriate to bring attention to some specific implications of the research findings, relative to suggested application and meaning for the industry level marketing practitioner, relative to the use of demographic segmentation. The basic implication is that demographic segmentation seems to have a very real potential for prediction of product purchases. Even if other methods of segmentation delineation seem to offer some pre- dictive advantages over the use of demographic techniques, demographics are likely to offer significant advantages in the identifiability of the consumer in existing, or even projected markets, by readily available demographic profiles for most metropolitan areas. Moreover, the dimen- sion of accessibility to targeted consumer segments is usually another 178 major advantage of demographic techniques, since most media that offer consumer data about their audiences have limited their data to demo- graphic characteristics of their audience-market.“ Thus, as long as reasonable predictive efficacy can be obtained via demographic segmen- tation criteria, demographics may be the most practical operational method of predicting consumer response, as well as targeting marketing efforts. The demographic variables that were researched in this study did provide a useful degree of predictive efficacy for most of the frozen food categories that were utilized in the empirical research of this study, with the results leading to related implications about methodologies that could be expected to increase the level of predictive efficacy that was demonstrated in this study. The results, combined with the comparative analysis of prior industry studies in the frozen food industry, also resulted in some strong implications for the mar- keting manager, to avoid some of the mistakes that have limited the past value of demographic segmentation attempts. The Need for Basic Statistical Testing One of the strongest implications of this study is that greater acceptance, use, and understanding of statistical techniques is a necessity for the practical, effective use of demographic segmentation. A sentiment expressed by many industry researchers and managers is that industry practitioners do not care about statistics, they only want results, as a rationale for ignoring or refusing to use even basic statistical techniques. 179 This common but unfortunate practice in market research is satirically discussed by Robert Ferber, in "How Not to Do Research-- An Editorial."s In this discussion, Ferber facetiously suggests to market researchers: In analyzing the data, don't bother to make significance tests. If people will not accept your word on the reli- ability of the data, too bad. Any results are meaningful in sgmg_sense, so with a little rationalization you can have a truly significant piece of work.6 Unfortunately, published "results," reported conclusions, or management decisions that are based upon statistically invalid and/or untested relationships, cannot be reasonably expected to effectively contribute to the practical knowledge or management expertise of the marketing practitioner. Published or company level studies that purport to present findings or conclusions about demographic relationships to product purchases, without any_reported process of statistical testing, should be regarded with a high degree of scepticism. The marketing practitioner need not be a statistical expert to at least recognize the necessity of basic statistical testing methods, as well as to appreciate the potential of more elaborate, multivariate techniques. The Need for Multivariate and Non-Linear Techniques Another research implication for marketing managers is that multivariate statistical methods will probably be a necessity for accurate use of demographic segmentation methods. As shown in the research of this study, no one demographic variable was significantly related to all of the product categories. Moreover, the bivariate 180 demographic predictive powers were generally too low, individually, to give practical predictive value. Yet, the multivariate use of techniques such as discriminant analysis may be able to combine several bivariate relationships of limited predictive power to provide a multi- variate basis for predicting purchases and classifying buyers that may be very useful to marketing management, with an increase in net predictive results for industry operationalization. Moreover, the empirical research of this study showed that much of the predictive power for the demographic variables was lost by means of linear measurement techniques. Thus, the implication that non-linear techniques should be used, when possible, seems quite clear, since greater predictive power would be expected to result from the same set of demographic variables, as well as delineating possible relationships that may have been hidden by linear statistical analysis. Modern computer facilities are available to most marketing firms today, even on a time-sharing basis, along with a growing support base of statistical programming, such that the suggestion of non-linear and multivariate techniques should not be unreasonably difficult or costly to obtain, particularly in comparison to the cost of collecting a modern market research survey, or in relationship to the possible value to the firm of being able to more accurately segment the market for its product or to better predict product sales. 181 Some Practical Management Uses for Demggraphic Segmentation Beyond the value of delineation of market segments as targets for active marketing efforts, demographics may prove to be especially effective in the prediction of future sales, based upon the data pro- vided by regional surveys or test markets relating product sales to demographic traits of the purchasers. Thus, as a new product is expanded beyond a test market, the information relating to who purchased the product could be used to select media alternatives, adjust image, or alter other marketing efforts to better appeal to those persons who should be the target of the firm's marketing efforts, ij_a definite buyer segment can be defined within the general population. Similarly, ' knowledge of the demographic characteristics of who purchased the product in one region or test market could allow better prediction of expected sales in a new region, based upon the demographic purchasing characteristics of the data base, extrapolated to the demographic characteristics of the new regional target market. Thus, sales territories could be more accurately evaluated for the marketing potentials of sales territories, for purposes of expansion or contraction of marketing efforts or product offerings, by means of the demographic profiles of the consumers known to live in the territories under analysis. For example, changes in price, taste, image, or distribution could be tested in a reasonably rep- resentative area, related to demographic variables of the purchasers, and then used to project results of such changes much more accurately than for simply the "average" consumer. Demographic adjustment of test 182 market results should also make it much easier to utilize test markets, since demographic profiles of the test market need not be as repre- sentative of the total market, as long as the relationships between demographic variables and purchases can be assumed to be consistent between the test market and the market area for which a marketing projection is sought. Demographic projections of consumer purchases could also be very useful to marketing management in the prediction of future sales trends for a given product market, by means of the very accurate pro- jections of demographic changes that are expected in the market which are readily available to the market researcher through such sources as government planning agencies, or the academic research in the field of Demographics. However, it should be noted that longitudinal studies, of demographic purchasing relationships over time, are noticeably vacant in the marketing literature, not only for demographic segmentation, but for all segmentation techniques.’ Failure to Find the "Northwest Passage" The failure to find a demographic-based "Northwest Passage" to market segmentation should not be perceived as a failure of this research. Rather, the road to effective use of demographic segmentation criteria is likely to come from the path of investigating specific product purchasing relationships, rather than trying to establish a general pattern for a broad family of products. While there is a con- stant temptation to cut corners in the cost of market research and strategic segmentation efforts, by seeking a single segmentation 183 pattern that can be applied to an entire group or family of products, the findings of this study suggest that efforts in segmentation are more likely to be successful, when directed toward specific products rather than a general product line. The Need for New Research Data Collection Methods A final, and resultant, implication to marketing management is that the combined need for specific product purchase data, non-linear and multivariate statistical techniques, and the use of an expanded set of demographic variables, all point to the requirement of larger sample sizes for market research data inputs than have typically been collected for past segmentation studies. This apparent need for larger sample sizes, to develop a more accurate data base for the effective use of demographic segmentation as well as the desire for information that is reflective of current market conditions, both point to the need for new techniques in collecting and processing of market research data inputs. This need becomes even more pronounced when the rising costs of traditional research techniques, such as the use of in-home questionnaires, is also recognized. Thus, just as the computer has become the principal method of analyzing large data sets, the computer may also become the principal method of data collection, with a resultant solution to current market- ing research problems of obtaining the large, current data base that becomes necessary in the use of the suggested multivariate techniques, for the use of demographic variables as segmentation criteria. It is 184 suggested that the use of passive data collection methods, such as by the use of electronic computer-based cash registers, may provide such a data source, if consumer purchases can be linked to demographic characteristics of the purchaser via credit card, electronic funds transfer cards, check cashing identification cards, or even an in- store survey of consumer demographic data. Technologically, there is nothing in the current state of computer development which would prevent such a system from being implemented at a projected moderate cost. Some computerized cash register systems are already being used to record product movement for inventory control purposes. It would seem to be a relatively modest task to add the additional dimension of recording individual consumer purchases, linked to the demographic characteristics of the purchaser, thus eliminating the usual research problems of interview cost, key punching and coding of questionnaires, and similar barriers to large, quick, and economical data bases for consumer purchasing research. Suggestions for Future Research Based upon the empirical findings, conclusions, and the derived implications of this study, this section of the study provides some sug- gestions for future marketing research involving the use of demographic variables as predictors of consumer purchasing behavior. The first suggestion is that research be conducted relative to the purchase of other specific product categories, and for specific products, if possible. Much of the past research relating to the use 185 of demographic variables seems to have involved rather broad categories of measured response, such as total food, total frozen foods, automobile brand, or other similar categories that may not allow measurement of specific differences in purchase response between different demographic categories. In fact, many product categories or product lines are strategically marketed to appeal to a broad range of consumer tastes, via substantial variety of specific products, brands, prices, features, etc. Thus, failure of demographic variables to disclose substantial differences between demographic purchases within a relatively broad category should not be perceived as a failure of the demographic variables, but rather, an expected response to current marketing practice. The findings of this study showed substantial difference in the relationships of the tested demographic variables to each of the twelve specific frozen food categories of this study. Further research within the frozen food industry seems to be suggested, for purchases within other frozen food categories, such as frozen meats, frozen desserts, ice cream, and other categories not surveyed in this study. Moreover, even the categories of this study, although much more specific than simply a measurement of total frozen foods, were still relatively broad in the number of products, and the inclusive variety of product prices, sizes, functions, and expected consumer appeal. Greater accuracy for prediction of purchases would seem to demand more specific measurement of what the consumer actually does buy. 186 Moreover, as research efforts seek to delineate the purchases of more specific products, it must be expected that larger samples must be used to properly use multivariate techniques. If a product-purchase category is so defined that 60 percent or more of the population buys it, specific definition of a market segment seems less likely. Simi- larly, if a product does appeal to a limited segment within the popu- lation, as in the case of cooked poultry in this study, a relatively large data sample is required to effectively define the segment. For example, only 68 families, out of 718 families in the entire survey, bought at least one package of cooked chicken during a month's reported purchases. It seems obvious that 68 respondents can give only a limited- statistical ability to determine what characteristics about this group are significantly different from non-buyers of this product category. It must also be recognized that this study can only give a sample of the relationships that might be derived from the demographic characteristics of the consumer, since only a limited number of char- acteristics were actually examined in this study. Thus, additional demographic variables are suggested for inclusion in future research, such as geographic region, religion, ethnic background, age of children, stage in the family life cycle, the number of children in the household within certain age brackets, or the medical characteristics and needs of the consumer. It is suggested that increasing use of computer technology to record specific product movement may allow an increased ability to acquire data for specific product purchases, for large numbers of 187 consumer respondents. Thus, electronic cash registers now can list specific items on the register receipt, as well as maintaining constantly up-to-date data on product movement. To the degree that consumer demographic characteristics can be surveyed at the point of sale, via identification cards, or electronic funds transfer cards, access to fast, accurate, and economical market research data is gradually becoming a reality, with an intriguing potential for future marketing research in the area of consumer demographics. Finally, the finding of relatively non-linear relationships for several of the demographic variables, suggests that the linearity of relationships should at least be tested in future research, rather than making the convenient, but probably incorrect, assumption that linear relationships are present. Thus, it is suggested that non-linear statistical methods should be applied to the research of demographic variables, a process which has been hindered by the lack of non-linear computer programs, combined with the need for larger sample sizes that are typically needed for non-linear multivariate methods of statistical analysis, two limitations that prohibited the use of a non-linear discriminant analysis in this study. However, non-linear techniques are becoming more available, such as the non-linear discriminant analysis which can now be obtained via the Institute for Social Research at the University of Michigan.8 This program, titled Multivariate Nominal Scale Analysis, is designed to handle research problems where the independent variables may be 188 measured at any level of measurement, including the nominal level. Thus, non-linear demographic categories can be measured without recoding, restructuring, or making the assumption of linearity, thereby avoiding the predictive loss that often follows such methods. Although this program does need a substantially larger sample size to realisti- cally analyze strongly segmented product purchases, this particular program, or other similar computer programs for multivariate analysis, may provide substantial potential for future marketing research. CHAPTER V--FOOTNOTES 1Johan Arndt, Market Segmentation: Theoretical and Empirical Dimensions (Bergen, Norway: Universitetsforlaget, 1974), pp. 24-25. 2Dik Warren Twedt, "How Important to Marketing Strategy Is the Heavy User," Journal of Marketing 28 (January 1964): 71-72. 3Terry Mathew Weisenberger, "Generalized Market Segments: A Study Using Selected Convenience Goods in Vigo County, Indiana" (Ph.D. dissertation, Michigan State University, 1977), pp. 123-124. “Henry Assael and Hugh Cannon, "Do Demographics Help in Media Selection," Journal of Advertising Research 19 (December 1979): 7-8. 5Robert Ferber, "How Not To Do Research--An Editorial," Journal‘ of Marketing Research 5 (February 1968): 104. 6Ibid. 7Yoram Wind, "Issues and Advances in Segmentation Research," Journal of Marketing Research 5 (February 1968): 104. 8Frank M. Andrews and Robert C. Messenger, Multivariate Nominal Scale Analysis (Ann Arbor: Survey Research Center, The University of Michigan, 1973). 189 APPENDIX A RELEVANT SECTIONS OF DATA BASE QUESTIONNAIRE FROM 1976 AKRON BEACON JOURNAL SURVEY APPENDIX A RELEVANT SECTIONS OF DATA BASE QUESTIONNAIRE FROM 1976 AKRON BEACON JOURNAL SURVEY aééfinaaép.rchéaéaéaxg9éa “£55237 8577/114/76 CONSUMER PROFILE ANALYSIS :5 i Akron, Ohio (3-6) 1976 (7) 1 ZONE: TELEPHONE PREFIX: COUNTY WHERE INTERVIEW COMPLETED: l - Summit 2 - Medina 3 - Portage MONTH OF INTERVIEW: O - January 4 - May 8 - September 1 - February 5 - June 9 - October 2 - March 6 - July X - November 3 - April 7 — August Y — December 190 191 "Hello, I'm interviewing on a market research survey. I would like to speak to the person who is the homemaker, that is, the one who buys most of the food and other household products for this home." Day of Week Date Time M T w TH F ST SN Remarks 1 2 3 4 Replacement for "hot at home on 3rd call" 5 Replacement for refusal Respondent's Name Respondent's Address City INTERVIEWER CERTIFICATION: I hereby certify that this interview was actually made at the address shown, that it is a true record of the information given me, and that the cost of this interview may be deducted from my pay if it is fbund to be improperly or inaccurately completed. Signature of Interviewer Date verified by: Date 192 65. "How much money do you estimate all members of this family spent last week on groceries, including home delivered groceries?" Under $15.00 4 $35.00 - $39.99 $15.00 - $24.99 5 $40.00 - $44.99 6 7 $55.00 - $59.99 $60.00 - $69.99 $70.00 & over Don't Know $25.00 - $29.99 $45.00 - $49.99 $30.00 - $34.99 $50.00 - $54.99 gator-‘0 l I I r<><©m I 69. "Thinking back over the last 30 days, about how many packages have you bought of (Record in appropriate column, below) l 6 11 16 21 26 31 36 41 46 to to to to to to to to to or None 5 10 15 20 25 30 35 40 45 More D.K. a. Frozen breakfast baked goods & breakfast pastries 0 1 2 3 4 5 6 7 8 9 X Y including frozen pancake/waffle batter & donuts b. Frozen dessert baked goods & 0 1 2 3 4 5 6 7 8 9 X Y pastries c. Frozen fish, shrimp, 0 or other seafood d. Uncooked frozen poultry, whole 0 l 2 3 4 5 6 7 8 9 X Y or in parts e. Cooked ('heat & serve') frozen 0 l 2 3 4 5 6 7 8 9 X Y poultry f. Frozen potatoes 0 1 2 3 4 5 6 7 g. Frozen vegetables 0 1 2 3 4 5 6 7 8 9 X h. Frozen concentrated X Y fruit juices 0 1 2 3 4 5 6 7 8 9 1. Frozen fruit or berries 0 1 2 3 4 5 6 7 8 9 X Y j. Frozen pizza 0 1 2 3 4 5 6 7 8 9 X Y k. Frozen complete dinners 0 1 2 3 4 5 6 7 8 9 X Y 1. Frozen main courses, including macaroni & cheese, chop suey, & meat or chicken pies, 0 1 2 3 4 5 6 7 8 9 X Y not including poultry 683. 193 ll Do you have a freezer in which to store f rozen foods that is a se arate ' part of the refrigerator?" P unlt, nOt Yes 0 - No 79. (INTERVIEWHR: Observe Race of'Respondent) l - White 2 - Black 3 - Latin-American 4 - Other 80. ACE OF HEAD OF HOUSEHOLD l - Less than 18 5 - (45 - 49) 9 - (65 or older 2 - (18 - 24) 6 - (50 - 54) Y - Don't Know/ ) 3 - (25 - 34) 7 - (55 - 61) Refused 4 - (35 - 44) 8 - (62 - 64) 81a. SPECIFIC OCCUPATION OF HEAD OF HOUSEHOLD (Industry) (Specific Job) 81b. SEX OF HEAD OF HOUSEHOLD . l - Male 2 - Female 81c. SEX OF HOMEMAKER RESPONDENT l - Male 2 - Female 82. (Interviewer: Hand Income Card to respondent) "Will you please select the number which represents your total household income before taxes last year. Please in- clude income for all household members, including dividends, rentals, etc." 0 - Under $2,500 4 - $8,000 - $8,999 8 - $15,000 - $17,999 1 - $2,500 - $4,999 5 - $9,000 - $9,999 9 - $18,000 - $19,999 2 - $5,000 - $6,999 6 - $10,000 - $11,999 X - 520.000 or more 3 - $7,000 - $7,999 7 - $12,000 - $14,999 Y - Don't Know/Refused 83. "What was the last grade of school completed by the head of the household?" College, incomplete College, graduate Some post graduate Don't Know/Refused l — Grammar School (grades 1 thru 6) 2 - Grammar School (7th or 8th grades) 3 - High School, incomplete (9th - 11th grades) 4 - High School, complete «:uoxm I 194 84. "Do you own this home or rent it?" T D- Rent E1 - Own (Skip to Q.86) ‘ (If "Rent", ask:) 85. "Which of these groups includes the monthly rental you pay?" (Hand list to respondent) l - Under $40 5 — $100 - $149 9 - $300 and over 2 - $40 - $59 6 - $150 - $199 Y - Don't Know/Refused 3 - $60 - $79 7 - $200 - $249 4 - $80 - $99 8 - $250 - $299 (Ask of'all:) 86, "How long have you lived in this home?" 1 - Less than one year 3 - Five to ten years 2 - At least a year, but 4 - Ten years or more less than five years Y - Don't Know 87. "Do you plan to move during the next three years?" <:--- 1 - Yes 2 - No (Skip to Q.90) 3 - Undecided (Skip to Q.90) X 88. "Do you plan to stay in the Akron area?" 1 - Yes 2 - No 3 - Undecided 89. "Do you plan to rent, buy or build?" 1 - Rent 2 - Buy 3 - Build 4 — Undecided 90- Household is in: l - Single dwelling unit structure 3 - Three or four unit structure 2 - Duplex structure 4 - Structure containing 5 or more units 91- "When did the head of the household first move to the Akron area?" 1 - Less than one year 4 - Ten — twenty years 2 - At least a year, but 5 - Twenty years or more less than five years Y - Don't Know 3 - Five - ten years 195 92. "How many persons are there in this household?" 1 - One 5 — Five 9 - Nine 2 - Two 6 - Six X - Ten or more 3 - Three 7 - Seven 4 - Four 8 - Eight 93. "How many members of this household are now employed?" 0 - None 2 - Two 4 — Four or more 1 - One 3 - Three Y - Don't Know/Refused 94. "Is there a telephone in this household?" (If "Yes", record telephone number) 1 - Yes 2 - No "THAT IS THE END OF THE INTERVIEW. THANK YOU." BIBLIOGRAPHY BIBLIOGRAPHY Alderson, Wroe. Marketing Behavior and Executive Action. Homewood, 111.: Richard D. Irwin, Inc., 1957. American Accounting Association, "Report by the Committee on Cost and Profitability Analysis for Marketing." Supplement to the Accounting Review, 1972. Cited in Crissy, Fischer, and Mossman, pp. 45-46. Andrews, Frank M., and Messenger, Robert C. Multivariate Nominal Scale Analysis. Ann Arbor: Survey Research Center, The University of Michigan, 1973. Appel, David L. "An Analysis of Consumer Market Segmentation in Response to an Institutional Innovation in the Food Industry." Ph.D. dissertation, Michigan State University, 1968. Assael, Henry, and Cannon, Hugh. "Do Demographics Help in Media Selegtion?" Journal of Advertising Research 19 (December 1979 : 7-11. Barton, S. G. "The Life Cycle and Buying Patterns." In Consumer Behavior. Vol. II, pp. 53-57. Edited by Lincoln Clark. New York: New York University Press, 1955. Bass, Frank M.; Tigert, Douglas J.; and Lonsdale, Ronald T. "Market Segmentation: Group Versus Individual Behavior." Journal of Marketing Research 5 (August 1968): 264-270. Bell, Martin L. Marketing Concepts and Strategy. Boston: Houghton Mifflin Co., 1979. Berridge, William A.; Winslow, Emma A.; and Flinn, Richard A. Purchasing Power of the Consumer: A Statistical Index New York: A. W. Shaw 8: Co., 1925. "Better Homes and Gardens Surveys the Frozen Food Markets." Frozen Food Factbook, 1981. Hershey, Pa.: National Frozen Food Association, 1981, pp. 40-62. Blattberg, Robert C., and Sen, Subrata K. "Market Segmentation Using Models of Multidimensional Purchasing Behavior." Journal of Marketing 38 (October 1974): 17-28. 196 197 Blattberg, Robert C., and Sen, Subrata K. "Market Segments and Stochastic Brand Choice Models." Journal of Marketing Research 13 (February 1976): 34-45. Boyd, Harper W., Jr. "The JMR's Editorial Objectives." Journal of Marketing Research 13 (February 1976): 1-2. Brandt, S. C. "Dissecting the Segmentation Syndrome." Journal of Marketing 30 (October 1966): 22-27. Brogowicz, Andrew A. "Race as a Basis for Market Segmentation: An Explanatory Analysis." Ph.D. dissertation, Michigan State University, 1977. Calantone, Roger J. "Research Methodologies for Benefit Segmentation Analysis." Ph.D. dissertation, University of Massachusetts, 1976. Abstract listed in "Marketing Doctoral Dissertation Abstracts." Bibliography Series No. 29. Chicago: American Marketing Association, 1977, pp. 26-27. Calantone, Roger J., and Sawyer, Alan G. "The Stability of Benefit Segments." Journal of Marketing_Research 15 (August 1978): 395-404. Chamberlin, E. H. Theory of Monopolistic Competition. Cambridge: Harvard University Press, 1933. Clark, Lincoln H. Consumer Behavior. Vol. 2. New York: New York University Press, 1955. Crissey, W. J. E.; Fischer, Paul; and Mossman, Frank H. "Segmental Analysis: Key to Marketing Profitability." M.S.U. Business Topics 21 (Spring 1973): 42-49. Dean, Joel. Managerial Economics. Englewood Cliffs, N.J.: Prentice- Hall, Inc., 1951. Dhalla, Nariman K., and Mahatoo, Winston H. "Expanding the Scope of Segmentation Research." Journal of Marketing 40 (April 1976): 34-41. Duesenberry, J. S. Income, Saving and the Theory of Consumer Behavior. Cambridge: Harvard University Press, 1952. Evans, Franklin B. "Psychological and Objective Factors in the Prediction of Brand Choice." Journal of Business 32 (October 1959): 340-369. Ferber, Robert. "How Not To Do Research--An Editorial." Journal of Marketing Research 5 (February 1968): 104. 198 Fisher, Janet. "Income, Spending and Saving Patterns of Consumer Units in Different Age Groups." In National Bureau of Economic Research Studies in Income and Wealth. Vol. 15, pp. 75-102. Princeton: Princeton University Press, 1952. . "Family Life Cycle Analysis in Research in Consumer Behavior." In Consumer Behavior. Vol. 2, pp. 28-36. Edited by Lincoln Clark. New York: New York University Press, 1955. Frank, Ronald E. "Market Segmentation Research: Findings and Implications." In Application of the Sciences in Marketing Management. Edited by Frank M. Bass et al. New York: John Wiley & Sons, Inc., 1967. Frank, Ronald E.; Massy, William F.; and Boyd, Harper W. "Correlates of Grocery Product Consumption Rates." Journal of Marketing Research 4 (May 1967): 184-190. Frank, Ronald E.; Massy, William F.; and Wind, Yoram. Market Sege mentation. Englewood Cliffs, N.J.: Prentice-Hall, Inc., 1972. "Frozen Food Study, 1971-72." Akron Beacon Journal, Akron, Ohio, 1972. "Frozen Food Study, 1977." Akron Beacon Journal, Akron, Ohio, 1977. Haley, Russel I. "Benefit Segmentation: A Decision-Oriented Research Tool." Journal of Marketing 32 (July 1968): 30-35. . "Implications of Market Segmentation." Conference Board Record, March 1969, pp. 33-37. . "Beyond Benefit Segmentation." Journal of Advertising Research 11 (August 1971): 3-8. Hanan, Mack. Market Segmentation: The Basis for New Product Innovation and Old Product Renovation. American Marketing Association Bulletin No. 109. Chicago: American Marketing Association, 1968. Herrmann, Robert 0. "Household Socio-Economic and Demographic Characteristics as Determinants of Food Expenditure Behavior." Ph.D. dissertation, Michigan State University, 1964. Kerlinger, Fred N. Foundations of Behavioral Research. New York: Holt, Rinehart, and Winston, Inc., 1973. Keynes, J. M. The General Theory of Employment Interest and Money. New York: Harcourt, Brace & Co., Inc., 1936. 199 Kotler, Phillip. "Behavioral Models for Analyzing Buyers." Journal of Marketing 29 (October 1965): 37-45. . Marketing for Nonprofit Organizations. Englewood Cliffs, N.J.: Prentice-Hall, Inc., 1975. Kotler, Phillip, and Levy, Sidney J. "Broadening the Concept of Marketing." Journal of Marketing 33 (January 1969): lO-l5. LaLonde, Bernard J. "A Profile of the Frozen Food Consumer." Frozen Food Factbook, 1973. Hershey, Pa.: National Frozen Food Association, 1973, pp. 61-91. LaLonde, Bernard J.; Hansen, Robert A.; Scott, Carol A. "Profile 1975: The Consumer and the Frozen Food Industry." Frozen Food Factbook, 1975. Hershey, Pa.: National Frozen Food Association, 1975. Lansing, John B., and Morgan, James N. "Consumer Finance Over the Life Cycle." In Consumer Behavior. Vol. 2, pp. 36-53. Edited by Lincoln Clark. New York: New York University Press, 1955. Leone, Robert P., and Schultz, Randall L. "A Study of Marketing Generalizations." Journal of Marketing 44 (Winter 1980): lO-lB. Levit, Theodore. "Marketing Myopia." Harvard Business Review, July- August 1960. Reprinted in The Great Writings in Marketing, pp. 36-57. Edited by Howard A. Thompson. Plymouth, Mich.: The Commerce Press, 1976. Mahajan, Vijay, and Jain, Arun K. "An Approach to Normative Segmen- tation." Journal of MarketinglResearch 15 (August 1978): 338-345. Marshall, Alfred. Principles of Economics. London: The Macmillan Co., 1890. Martineau, Pierre. "Social Classes and Spending Behavior." Journal of Marketing 23 (October 1958): 212-230. Massachusetts Bureau of Statistics and Labor. Sixth Annual Report of the Bureau of Statistics and Labor. Public Document 31. Boston: Wright & Potter, State Printers, 1875, p. 438. Mayer, Kurt. "Diminishing Class Differentials in the United States." In Marketing and the Behavioral Sciences, p. 203. Edited by Perry Bliss. Boston: Allyn and Bacon, Inc., 1963. Michman, Ronald 0. "Market Segmentation Strategies: Pitfalls and Potentials." American Marketing Association Proceedings. Chicago: American MarketingTASSociation, 1971, p. 322. 200 Miller, Donald. "The Life Cycle and the Impact on Advertising." In Consumer Behavior. Vol. 2, pp. 57-59. Edited by Lincoln Clark. New York: New York University Press, 1955. Mossman, Frank H.; Crissy, W. J. E.; and Fischer, Paul M. Financial Dimensions of Marketing Management. New York: John Wiley & Sons, 1978. National Industrial Conference Board. Expenditure Patterns of the American Family, New York: National Industrial Conference Board, 1963. Market Profiles of Consumer Products. New York: National Industrial Conference Board, 1965. Nickels, William G. Marketing Principles. Englewood Cliffs, N.J.: Prentice-Hall, Inc., 1978. Nie, Norman H.; Hull, C.; Jenkins, J.; Steinbrenner, K.; and Bent, 0. Statistical Package for the Social Sciences. New York: McGraw-Hill Book Co., 1975. "Orange Juice, Dinners, Vegetables, and Pizza Head List of Fastest Moving Frozen Food Items." Supermarketing, 1 April 1975, p. 42. Phillips, C. L., and Leininger, R. D. "The Influence of Selected Socio- Economic Variables on Supermarket Purchase Patterns." Food Marketing Paper No. 8. Michigan State University, East Lansing, Michigan, 1968. Plummer, Joseph T. "The Concept and Application of Life Style Segmentation." Journal of Marketing 35 (April 1971): 35. Reynolds, William H. "More Sense About Market Segmentation." Harvard Business Review, September-October 1965, pp. 107-114. Robinson, Joan. The Economics of Imperfect Competition. London: Macmillan, 1933. . The Economics of Imperfect Cgmpetition. London: Macmillan, 1948, pp. 179-188. Scott, Raymond C. "Frozen Food Purchases in Syracuse, New York." Department of Agricultural Economics, New York State College of Agriculture, Cornell University, Ithaca, New York, February 1948. Schwartz, Kenneth. "Fragmentation of the Mass Market." Dun's Review and Modern Industry, July 1962. 201 Schyberger, BO W. Theory and Methods of Market Segmentation. Research Report No. 64. Stockholm: Department of Business Administration, Stockholm University, 1971. Shaffer, Carol Warwick. "Income and Expenditure Patterns Related to the Life Cycle." Ph.D. dissertation, Michigan State University, 1964. Shaw, Arch W. Some Problems in Market Distribution. Cambridge: Harvard University Press, 1915. Sibley, Stanley 0. "The Identification of Behavioral, Knowledgeable, and Demographic Market Segments of Purchasers of Household Durables." Ph.D. dissertation, Michigan State University, 1972. Smith, Wendell R. "Product Differentiation and Market Segmentation as Alternative Marketing Strategies." Journal of Marketing 21 (July 1956): 3-8. Springer, Jack. "1975: Bad Year for New Products, Good Year for Segmentation." Advertising Age, 10 February 1975, pp. 38-40. Toffler, Alvin. Future Shock. New York: Bantam Books, 1971. Tollefson, John 0., and Lessig, V. Parker. "Aggregation Criteria in Normative Market Segmentation Theory." Journal of Marketing Research 15 (August 1978): 346-355. Trout, Jack, and Reis, Al. "Positioning Cuts Through Chaos in the Marketplace." Advertising Age, May 1972. Reprinted in Marketing Classics, 4th edition, pp. 422-429. Edited by Ben M. Enis and Keith K. Cox. Boston: Allyn & Bacon, Inc., 1981. Twedt, Dik Warren. "How Important to Marketing Strategy Is the Heavy User." Journal of Marketing 28 (January 1964): pp. 71-72. U.S., Department of Agriculture. Family Incomes and Expenditures. Miscellaneous Pub. No. 339-489 (193941941). Veblen, Thorstein. The Theory of the Leisure Class. New York: The MacMillan Co., 1899. Waldo, C. N. "What's Bothering Marketing Chiefs Most? Segmenting.‘I Advertising Age, 4 June 1973, p. 77. Warner, W. Lloyd., and Lunt, Paul. The Social Life of a Modern Community. New Haven: Yale University Press, 1941. Warner, W. Lloyd; Meeker, Marchia; and Ellis, Kenneth. Social Class in America. Chicago: Science Research Associatesj'1949. 202 Weisenberger, Terry Mathew. "Generalized Market Segments: A Study Using Selected Convenience Goods in Vigo County, Indiana." Ph.D. dissertation, Michigan State University, 1977. Wheatley, John J.; Chiu, John S. Y.; and Stevens, Andrea C. "Demographics to Predict Consumption." Journal of Advertising Research 20 (December 1980): 31-38. Whyte, William H., Jr. "The Web of Word of Mouth." Fortune 50 (November 1954): 140. Wilemon, David L. "An Analysis of the Age-Occupation Matrix as a Criterion for Vertical and Horizontal Market Delineation." Ph.D. dissertation, Michigan State University, 1967. Wind, Yoram. "Issues and Advances in Segmentation Research." Journal of Marketing Research 15 (August 1978): 317-337. Wind, Yoram, and Green, P. E. "Some Conceptual, Measurement, and Analytical Problems in Life Style Research." In Life Style A and Psychographics, pp. 99-126. Edited by R. Haley. Chicago: American Marketing Association, 1974. Winter, Frederick W. "A Cost-Benefit Approach to Market Segmentation.‘ Journal of Marketing 43 (Fall 1979): 103-111. Yankelovich, Daniel. "New Criteria for Market Segmentation." Harvard Business Review 42 (March-April 1964): 83-90. Young, Shirley; Ott, Leland; and Feigin, Barbara. "Some Practical Considerations in Market Segmentation." Journal of Marketing Research 15 (August 1978): 405-411. Zimmerman, Carle C. Consumption and Standards of Living. New York: 0. Van Nostrand Co., Inc., 1936.