Issues Concerning Empirical Methods and Findings in Consumer Research

ABSTRACT - This is a summary of discussion of three papers which were presented at the "Empirical Methods and Results" session of the 1978 Conference of the Association for Consumer Research. One paper concerns empirical methods for the analysis of contingency data, and the logic behind these methods. Another is an empirical investigation of sex image related product perceptions. The third involves an empirical analysis of alternative predictors of credit card usage behavior. Each of the papers is briefly discussed.

Citation:

William D. Perreault, Jr. (1979) ,"Issues Concerning Empirical Methods and Findings in Consumer Research", in NA - Advances in Consumer Research Volume 06, eds. William L. Wilkie, Ann Abor, MI : Association for Consumer Research, Pages: 614-616.

Advances in Consumer Research Volume 6, 1979      Pages 614-616

ISSUES CONCERNING EMPIRICAL METHODS AND FINDINGS IN CONSUMER RESEARCH

William D. Perreault, Jr., University of North Carolina

ABSTRACT -

This is a summary of discussion of three papers which were presented at the "Empirical Methods and Results" session of the 1978 Conference of the Association for Consumer Research. One paper concerns empirical methods for the analysis of contingency data, and the logic behind these methods. Another is an empirical investigation of sex image related product perceptions. The third involves an empirical analysis of alternative predictors of credit card usage behavior. Each of the papers is briefly discussed.

INTRODUCTION

The "Empirical Methods and Findings" session of the 1978 Conference of the Association for Consumer Research focused on three interesting papers. One of these concerns a study of differences and similarities between consumers who prefer two alternative credit card systems: a bank card system and a retail store card system (Hirschman, Srivastava, and Alpert, 1978). A second paper concerns the sex image related product perceptions of a sample of students and how those perceptions vary according to the student's sex, sex role self concept, and several other factors (Golden, Allison, and Clee, 1978). The third concerns different models that have been proposed for analyzing consumer categorical responses to questions concerning the presence of an attribute for a stimulus product (Pessemier, 1978). These papers basically concern very different issues, and thus they will be discussed individually.

CREDIT CARD PREFERENCES

The study of credit card preferences by Hirschman, Srivastava, and Alpert (1978) appears to have been carefully done, and it is reported in an interesting way. Generally, the analysis suggests that consumers prefer to use a credit system that provides benefits which are most consistent with their life patterns and needs.

The results of the analysis are consistent with expectation. Given that the sample frame is limited to the customers of one relatively "upscale" department store, it is not a surprise that the discriminatory power of the demographic data is limited. In addition, when one considers the specific operationalization of the criterion variable, a discriminating benefit structure between the credit alternatives is logical. Specifically, to be included in the analysis a consumer must actually have had both types of credit cards and must also have previously used both types for purchases at the store involved. In considering the results of the study, it is important to keep this criterion selection process in mind, as we are dealing with a relatively select group contrasted with other definitions of card system preference which might yield different group characteristics.

The authors do an interesting job of comparing and contrasting the relative classification efficacy of the various predictors. This is a recurring methodological problem in applied research, yet there are no generally accepted tests for the comparison. The issue underlying such an analysis is the trade off between parsimony and better classification. From this comparative perspective, however, it is possible to directly test the marginal, multivariate between group discrimination of the more complex model (including both the demographic predictors and benefit variables) as contrasted with the simple (benefits only) model. Given the criterion variable used in this study, it seems that the strength of the benefit predictors is the most relevant theoretical issue. In addition, since the simple model is a nested subset of the total model, the partial F test may be used to determine if discrimination is improved by adding the demographic variables.

In summary, this is an interesting study. Although the generalizability of the specific results to other sample frames may be limited, the study provides a relatively clear empirical example of benefit segmentation. Benefit segmentation research has been approached from many different perspectives, yet frequently the reports in this area are primarily methodological in nature and do not document that the derived segments are "meaningful" in terms of differential market behaviors. Since this study approaches the segments after first identifying the different behaviors, it does provide that evidence in a sample of consumers in which most other potentially confounding factors are, in essence, held constant.

PRODUCT SEXUAL IMAGES

The study by Golden, Allison, and Clee (1978) documents that there are complex relationships between consumers' perceptions of products, and the consumer's sex, sex role self concept, and self esteem. A particularly interesting aspect of the descriptive analysis in this study concerns the empirical result that for the 125 products initially evaluated none were perceived (rated) as androgynous or neuter by the student respondents. Apparently most products have a relatively distinct sexual identity.

The study is ambitious in its objective of attempting to sort out various factors that may be related to sex-typing product perceptions. As is frequently the case with ambitious undertakings, however, the data are relatively complex. Specifically, from each student respondent repeated measures are taken concerning perceptions of both male and female sexual images of each of five different products; in addition, each of the 307 students is characterized as a member of one of 48 cells in a factorial design. The factorial design is based on the respondent's sex (yes, two levels), sex role self concept (4 levels), self esteem (3 levels), and product usage (2 levels, but the operational definition of these levels is not provided).

Perhaps it is because of this complexity that this reviewer had difficulty understanding the motivation for the research design developed or appreciating the importance of the basic issues which might be addressed by that design. Of course, the authors' relative allocation of emphasis (toward the reporting of data analysis) and away from the theoretical or practical issues relevant to their research) may simply have been a function of space constraints.

At a broad level, the paper stimulates the reader to think about a number of issues relevant to the sexual images of products. For example, the study indirectly raises the question "Can the sexual image of a product be altered by presenting it in alternative use situations?" A related question is whether or not such perceptual changes would in fact result in altered behavioral usage and consumption patterns. This should be an interesting area for future research. Or, alternatively, does use of a product alter the user's sexual image of that product? Does it alter it in the direction of the user's sexual self concept? Theories of cognitive consistency would predict such results. This would have interesting implications for advertisers of products who were attempting to re-pioneer their products by opening up new markets for old products by expanding the market segment across lines of gender. This is, of course, the strategy used by many manufacturers of hair spray a number of years ago. If the notion that products have a sexual identity is correct, then this type of positioning effort might be very effective for many companies. For example, if a pocket knife is a masculine product, and presumably the heavy user's are men, then why not a pocket-book knife for women. If one believes that over time advertising has a strong but subtle effect on social values, it might even be that the type of effect could result in some reduction in sex role stereotyping. This is clearly ideal conjecture, and not intended to suggest that these implications are drawn from the authors findings; rather, the intent is to illustrate the types of issues that are related, even if only indirectly to the topical area of this research.

The numerical results and conclusions reported by the authors are also interesting. However, there are methodological concerns relevant to this study which suggests that the authors conclusions and statistical results should be viewed with caution. For example, one must be very cautious in interpreting tests of main effects in the presence of interactions. This concern is especially relevant to nonorthogonal designs, and to designs with small cell size and highly disproportionate cell sizes.

Even ignoring the presence of both low and high order interactions, and the lack of independence of the different measures across the different products, tests of the main effects are likely to be correlated, and the probability levels associated with standard F tests in such a situation are generally not accurate.

These same difficulties may be a partial reason why a clearer pattern of effects does not emerge in the study. With so many cells, there are very few observations in some cells and therefore the response of a single individual may have a substantial impact on results, especially given the multiplicity of responses per subject. More complete descriptive statistics (and cell sizes) would allow the reader to better evaluate whether these concerns are unnecessary. The authors have faced a difficult trade off of reporting a less complete analysis versus a more complex one which is more difficult to interpret.

In summary, this study explores several thought-provoking issues relative to the sex-type image of products. Although the conclusions of the study must be considered with caution due to the complexity of the design and the resulting difficulties in partitioning sources of effects, the study does show that there are generalized differences in the sexual images of products across different respondents, and it stimulates the reader to consider a number of issues relevant to the sexual image of products.

UNDERSTANDING CONTINGENCY DATA

The paper concerning understanding and analysis of contingency data (Pessemier, 1978) is a very useful review of segments of the literature relevant to the measurement of consumer judgments. The source literature in this area is frequently and unnecessarily complex and disjointed. Thus, an effort to concisely integrate the major issues is a substantial challenge, and Pessemier's paper meets that challenge. He not only briefly reviews the major traditional response functions, however, but also provides several insights concerning the response functions. His thoughts concerning consumers' evoked sets, determination of relevant attribute sets, and data "encoding" processes are particularly creative. While other arguments might be advanced to explain some of the same phenomena, the issues Pessemier raises are pertinent not only to those concerned with contingency response data, but more broadly with consumer information processing, measurement theory, and composition models of consumer choice.

Pessemier touches on several interesting topics when he considers perceptual space models as alternatives to the traditional response models such as the law of comparative judgment or signal detection theory. Although references to appropriate sources are provided, this issue perhaps warrants more detail and emphasis than was provided. A serious limitation of the traditional response models is that they are not suited for individual level analysis, even though they are based on the supposition that there is a distribution of discriminal thresholds. By contrast, recent developments in individual differences multidimensional scaling (and in some forms of monotonic regression) are based on the theoretical basis that the "scale" underlying the individual's response categories may be recaptured, by the scaling process if not by the measurement process itself. These and other issues concerning the relationship between nonmetric multidimensional scaling and the traditional models seems to be very pertinent to potential users and thus to a retrospective synthesis of this sort.

An issue which Pessemier does not address directly, but which also seems pertinent to his synthesis, is the extent to which researchers can quantify the "true" interval scale values which underlie the basic categorical (yes, no, don't know ) responses of concern. For example, assume that it is possible to quantify the response categories so that each category (or even each individual within each category) is assigned a numerical value reflecting the relative nature of his or her response. In such a circumstance, the relationship between the response and other relevant behaviors or attitudes of the consumer might be evaluated with more traditional quantitative models. That is, the whole problem of contingency analysis (from the traditional perspective) would be bypassed. In fact, that is the logic motivating the development of a new set of numerical analysis procedures which rely on the principles of alternating least squares optimal scaling. Additional discussion concerning these approaches may be found elsewhere (see, for example, Young, de Leeuw, and Takane, in press; or Perreault and Young, 1978). It seems that these procedures hold much future promise for the type of research situation which Pessemier addresses. More generally, these models for quantifying non-metric data should be relevant to a broad realm of analysis situations, including applications of multivariate linear analysis to mixed measurement level (binary, nominal, ordinal, or interval) data and applications of individual differences metric and nonmetric multidimensional scaling.

In summary, the Pessemier paper simultaneously synthesizes a difficult area and make contributions concerning the relationship of the measurement models to different cognitive areas. It is important that applied researchers consider and understand the behavioral and cognitive (as well as statistical) assumptions underlying quantitative response models; the Pessemier paper, and others like it, are a means for researchers to efficiently gain that perspective.

CLOSING COMMENTS

In point of fact, the three papers presented in this paper are quite dissimilar in focus, both in terms of their substantive concerns and in terms of relevant methodological issues. They fit together logically, however, because they stimulate each of us to think about what we individually consider important in empirical research in consumer behavior. Specifically, implicit to each of these papers there appears to be a different philosophy by the authors concerning the roles of (1) theory, (2) research models and design, and (3) an applications orientation in empirical consumer research. As in any other applied discipline, the relationships among these issues is complex and involves trade offs; focus on one of these areas may result in less emphasis in another. The challenge to empirical researchers in consumer behavior, individually and collectively, is to find a proper balance among these trade offs.

REFERENCES

Linda L. Golden, Neil Allison, and Mona Clee, "The Role of Sex Role Self Concept in Masculine and Feminine Product Perceptions," in William L. Wilkie, (ed.), Advances in Consumer Research, this volume.

Elizabeth C. Hirschman, Rajendra K. Srivastava, and Mark I. Alpert, "An Empirical Examination of Alternative Models for Predicting Consumer Utilization of Two Credit Card Systems," in William L. Wilkie, (ed.), Advances in Consumer Research, this volume.

Edgar A. Pessemier, "Understanding and Analyzing Contingency Data," in William L. Wilkie, (ed.), Advances in Consumer Research, this volume.

William D. Perreault, Jr. and Forrest W. Young, "Alternating Least Square Optimal Scaling: Analysis of Nonmetric Data in Marketing Research," working paper, Graduate School of Business Administration, University of North Carolina at Chapel Hill, 1978.

Forrest W. Young, Jan de Leeuw, and Yoshio Takane, "Quantifying Qualitative Data," in H. Feger (ed.), Similarity and Choice (New York: Academic Press, in press).

----------------------------------------