An Empirical Examination of Alternative Models For Predicting Consumer Utilization of Two Credit Card Systems

Elizabeth C. Hirschman, New York University
Rajendra K. Srivastava, University of Texas at Austin
Mark I. Alpert, University of Texas at Austin
ABSTRACT - Credit-facilitated purchase behavior is an increasingly important phenomenon in society. Within this context little empirical research has investigated predictive models of credit system preferences and utilization. This paper reports three models developed in a hierarchical fashion from demographic, mobility and benefit variables applied to users of two alternative credit card systems.
[ to cite ]:
Elizabeth C. Hirschman, Rajendra K. Srivastava, and Mark I. Alpert (1979) ,"An Empirical Examination of Alternative Models For Predicting Consumer Utilization of Two Credit Card Systems", in NA - Advances in Consumer Research Volume 06, eds. William L. Wilkie, Ann Abor, MI : Association for Consumer Research, Pages: 592-598.

Advances in Consumer Research Volume 6, 1979      Pages 592-598

AN EMPIRICAL EXAMINATION OF ALTERNATIVE MODELS FOR PREDICTING CONSUMER UTILIZATION OF TWO CREDIT CARD SYSTEMS

Elizabeth C. Hirschman, New York University

Rajendra K. Srivastava, University of Texas at Austin

Mark I. Alpert, University of Texas at Austin

ABSTRACT -

Credit-facilitated purchase behavior is an increasingly important phenomenon in society. Within this context little empirical research has investigated predictive models of credit system preferences and utilization. This paper reports three models developed in a hierarchical fashion from demographic, mobility and benefit variables applied to users of two alternative credit card systems.

INTRODUCTION

Credit cards appear to serve three general functions in society. First, they may effectively ease purchasing on credit by reducing transaction time. Second, they may permit more immediate gratification of consumer needs and wants. Third, they have been said to serve as an economic stimulus and thus raise living standards. Credit cards have also been characterized as having some negative social effects. Among these are first, the stimulation of materialistic values and hedonism; and second, the creation of indebtedness among consumers which may lead to anxiety and bankruptcy.

Despite the normative debate concerning their relative utility, credit cards are an innovation with a longstanding history in the United States. Here credit cards have been introduced consecutively in three major forms: retail store-issued credit cards, "travel and entertainment" (Diners Club, Carte Blanche, etc.), "bank" cards (VISA, Master Charge). Each possesses different combinations of common and unique attributes. It would appear that by virtue of coming last the bank card is the most advanced of the credit cards. However, at the present time it shows no signs of eliminating its predecessors. That different consumer need-fulfillments are provided by each of the three forms of credit card is evidenced by the fact that many people carry two and sometimes all three of the different forms of cards, as well as multiple types within a particular form [Shay and Dunkelberg, 1975].

The credit card in all three forms may be usefully conceptualized as a service system for facilitating value exchanges. That is, the card, itself, is simply a symbol which permits the carrier to effect monetary transactions. The three basic credit card systems available at the present time, however, are all constrained with respect to the types and value of the transactions which may be conducted. For example, a credit card issued by a department store may generally only be used to conduct transactions at units of that store. A credit card issued by a bank, while usually enjoying a wider geographic and inter-retailer acceptance than a typical retail store credit card, is still not accepted by all retail outlets, particularly those retail chains having their own nationwide credit systems [i.e. Penneys, Sears]. Further, both of these credit systems are limited in their utility to the user by issuer-imposed "ceilings" or "credit limits" on the absolute monetary value of the exchanges that may be made with them.

Despite the similarity of their constraints, the retail store credit card system and the bank credit card system would appear to provide the user with a dissimilar set of benefits. For example, the fact that a retail store issues its own credit card may provide the user with a perceived enhanced ability to negotiate with the store in exchanging or returning merchandise and in rectifying incorrect billing, [Shay and Dunkelberg, 1975].

Conversely, the bank issued credit card system places an intermediary [i.e. the bank] between the purchaser and the retailer, thereby reducing the direct negotiating power of the buyer. However, as mentioned earlier, this potential disadvantage of the bank card may be offset for some users by the fact that the bank card is accepted as a valid means of exchange at a much wider variety of retail outlets, as well as having almost global recognition and acceptance.

It would appear both theoretically and pragmatically useful to determine if this hypothesized benefit structure is so perceived by the consumer. Of perhaps even greater utility would be the determination of whether or not an individual's credit-card-facilitated exchange transactions were influenced by the way in which the benefits attributed to a credit card were evaluated. In other words, the theoretical research question becomes one of ascertaining whether persons who place differential importance on certain credit card attributes also favor and utilize different types of credit card systems for facilitating their exchange processes.

A high correlation between perceptions of credit card system features evaluations and credit card choice, of course, paves the way for a benefit segmentation approach to both the understanding and promoting of a given credit system. However, in the interest of parsimony, it would be desirable if such a benefit-structure segmentation proved superior to segmentation derived from a simpler and more easily applied basis such as demographics. Demographic variables have already been demonstrated effective in describing and predicting users of bank-issued credit cards [Adcock, Hirschman, Goldstucker 1976; Goldstucker, Hirschman 1977; Hirschman, Blumenfeld, Tabor, 1977; Matthews and Slocum, 1969; Plummer 1971; Slocum and Matthews 1975; Wiley and Richard, 1975], as well as retail store-is-sued credit cards [Russell, 1975; Mateer, 1969; Shay and Dunkelberg, 1975]. Therefore, to be considered most useful in a pragmatic sense, benefit segmentation should be shown to serve either as a replacement for or complement to the traditional segmentation base of demographics.

To enhance the potential utility of such a segmentation approach, one should investigate reported purchase behavior, [rather than reported preferences] and a situation in which the individual is confronted with a genuine choice between alternative credit card systems. One shortcoming of much of the previous research has been that the descriptive profiles developed have been based upon an individual's "usual" mode of payment for credit purchases or, even more often, simply upon possession of a certain type of card. Few attempts have been made to measure the correlation between possession and usage. Second, in many instances no control has been made for different situational contexts. For example, an individual may "usually" use a bank-issued credit card, but in department store purchase situations may prefer (or be forced) to use a retail store issued card. These possible limitations in some of the prior research on credit cards may cast some doubt on the validity of the findings reported.

Thus, it would appear that to address adequately the problem of measuring the validity of a demographic-based credit card system choice model versus one based solely or incrementally on perceived benefits two conditions must be present-the individual's purchased behavior must be studied while controlling for the usage-context, and the individual must be confronted with a genuine choice as to credit card payment mode.

The purpose of this research is therefore three-fold. First, an investigation will be made to determine if persons who primarily use bank-issued credit cards when they have a choice of these versus retail store-issued cards can be effectively discriminated from those who prefer to use the latter, using standard demographic variables. Second, we shall investigate whether more effective discrimination and prediction can be obtained by augmenting basic demographics with particular demographics deemed relevant to choices of credit systems, such as mobility. Third, we shall investigate what, if any, additional explanatory and discriminatory power may be gained by utilizing information concerning the relative importance of various credit system benefits to those individuals choosing alternative systems. The addition of non-demographic variables should be advocated only to the extent that they provide significant explanatory and predictive strength to the process of understanding consumer choices and their implications for (in this case) marketing and credit management.

METHODOLOGY

Data Gathering. Data for this research were drawn from a survey of department store customers who were intercepted on a random basis by professional interviewers and administered a structured questionnaire. The questionnaire dealt with purchases made at the store, mode of payment used, attitudes concerning various credit instruments, store patronage, and a detailed set of demographic characteristics. The survey was conducted in two major southwestern cities from March 21 to April 3, 1977. One of the survey cities has a population of slightly under 2,000,000; the other has approximately 500,000 inhabitants.

In the first and larger city 3,000 interviews were conducted, while 1,225 were gathered in the second city. A department store chain operating stores in both cities cooperated in the research by allowing interviews to be conducted with exiting customers. All interviews were conducted on a random intercept basis at the major store entrances during all hours of normal store operations. Interviews were conducted at five branch stores in the first city and one branch store in the second city.

Because certain quotas of purchasers/non-purchasers and bank card users/non-users had to be met and respondent screening used to meet these quotas, the demographic data are not representative of this department store chain. Given the focus on credit system choices, cash-users are underrepresented here. This qualification, however, does not inhibit the utility of the data for the purposes intended here.

Questionnaire

The questionnaire used to generate the data for this research consisted of an extensive set of structured questions designed to investigate several aspects of shopping and credit-related activities. One of the portions of the questionnaire relevant to this research consisted of two questions designed to qualify the respondent as a person who usually used either the bank card or the retailer-issued credit card for making purchases at the store, and who had prior experience in making credit purchases at the store. These questions are given in Exhibit 1.

EXHIBIT ONE

VARIABLE MEASUREMENT

A second portion of the questionnaire relevant to this research was concerned with the respondent's demographic characteristics. These demographics were divided during the model-building part of the study into two groups. The first group consisted of the "basic" demographic traits of age, education, income, sex, race and life cycle stage, and are shown under the heading "Demographic Predictors" in Exhibit One. A second group of demographic variables--length of residency in the market and resident/non-resident status were held out of the basic demographic model, as it was anticipated that they would possess significant incremental predictive ability, even though they are infrequently used in demographic segmentation studies. These are shown under "Mobility Predictors" in Exhibit one.

Third, a set of eleven attributes relevant to the selection of alternative credit card systems was included in the questionnaire. The respondent was asked to tell if each attribute was of very much importance, moderate importance or little importance in using a credit card. Responses were scored in reverse order, 1, 2 or 3, respectively. These are shown in Exhibit One, also, under the heading "Benefit Predictors."

This set of benefits was arrived at by conducting a series of ten focus group sessions in the two markets prior to the in-store interviewing phase of the research. The topic at each of the sessions was the use of credit and credit cards. The benefits [i.e. relevant attributes] identified by the various focus group sessions were largely consistent from group to group regardless of group composition. Thus, the benefit list developed is believed to possess some degree of generalizability.

Hypotheses

Three hypotheses were generated at the outset of the research. First, it was anticipated that since there are few, if any, differences in the demographic qualifications imposed by issuers of bank cards and retail store credit cards on persons applying for these cards, it follows that there should be few demographic differences between the groups using these cards. In fact, since persons belonging to either the bank or store card user groups possessed both types of cards, demographic variables should be of little value in differentiating between them. Further, since the retail store surveyed appeals largely to higher income customers, little co-variation between card choice in this context and demographic characteristics would be expected.

Second, it was anticipated that a shorter residency period [implying greater mobility] and visitor status should be useful discriminators between bank card users and store-issued card users. The reasoning behind this hypothesis was that persons who were from out of town or recently arrived would prefer a credit card that could be used across a wider geographic area, and also have less store loyalty than longer-term residents; both these factors should contribute to a heightened likelihood that such persons would use a bank card as their usual purchase mode.

Third, it was hypothesized for the reasons set forth in the introduction to this paper that the benefits of ability to return merchandise and straighten out incorrect billing would be more highly valued by persons using store-issued credit cards; while the benefit of ability to use at a wide variety of stores, ability to use all over the country and ability to consolidate bills would be more highly valued by persons using bank credit cards.

These three research hypotheses are stated below:

H1: There will be no demographic differences between persons who prefer to use bank cards, ["bank card users"], and those who prefer to use store-issued credit cards, ["store card users" ].

H2: There will be discriminating mobility differences between bank card users and store card users,

H3: There will be discriminating benefit importance differences between bank card users and store card users.

For statistical analysis, hypotheses were restated in the 'null' form.

Analysis and Findings

To test the predictive and explanatory validity of the demographic, demographic plus mobility and demographic-mobility-benefit models, a group of respondents who possessed a bank card and a card issued by the department store whose customers were surveyed was selected. Since the store's acceptance of bank cards was a recent event, we narrowed the analysis to include only those persons who had used both types of credit cards to charge purchases at this store chain. This procedure would assure first, that persons who chose to use the retail card did not do so simply because of ignorance of the bank card's acceptability at the store rather than a preference for retail credit cards, and second, that the person was not constrained by lack of possession of one of the two relevant credit cards. Bank card users were defined as those who had used both types of cards but indicated that they usually used bank cards to pay for their purchase at the store. Retail card users were similarly defined as persons who had used both types but usually paid for their purchases with their retail card.

Having obtained a group of respondents who were qualified to choose between using their store-issued card or their bank card, we could be sure that the choice involved a conscious and unconstrained decision. A series of multiple discriminant analyses were run to test the discriminating ability of hierarchical combinations of demographics [Model II, mobility [Model II], and benefits sought [Model III]. By adding mobility to demographics, we could see whether additional explanatory power could be obtained by considering the impact of recent movement to the city, with other demographic variables "held constant" by the discriminant function. By adding benefits sought to this enriched model, we could see whether persons with similar demographic and mobility profiles might have differential probabilities of choosing to usually use a store-issued versus a bank card, when considering benefits sought from credit systems. To the extent that benefits are correlated with demographics and/or mobility, the incremental contribution of these variables would be lessened, once the predictive ability of the enriched demographic/mobility model had been obtained. However, if additional predictive validity could be obtained by considering attitudes not already "explained" by the earlier variables, some additional conceptual and pragmatic strength could be claimed.

Table 1 presents a summary of the results of testing the three models in the form of linear discriminant analysis, using the BMDP7M stepwise discriminant analysis with the "jackknife" method of cross-validation. There were 242 respondents who had used both types of credit systems at the intercept stores, and had supplied complete answers to the questions measuring the variables that were to be used to attempt to classify them. Persons who had not used both types of cards, or who had missing data were eliminated to prevent making specious comparisons. Of these 242 respondents, 119 indicated that they normally used a bank card at the intercept store, and 123 indicated a loyalty to the retail card of that store. While these figures provide approximately equal group sizes for the discriminant analysis, it is important to note that the percentage actually using bank cards in these stores would be substantially lower than 50%, since this sample was trimmed to include those who had actually used both types of cards.

TABLE 1

COMPARISON OF MODELS - DISCRIMINANT ANALYSIS

[Unlike the "change in Rao's V" test for the change in the statistical discriminatory abilities of models which are formed by adding variables to existing models, there exists no straightforward test for the significance of incremental practical significance, or percentages correctly classified by two models. However, it is possible to construct such a test by recognizing that the cross-validated predictions of various models formed incrementally may be treated as "repeated measures" or correlated observations in an Analysis-of-variance design. The dependent variable becomes dichotomous ("0" for misclassified, "1" for correctly classified), but simulations have shown that ANOVA is appropriate for dichotomous as well as continuous measures, provided the samples are reasonably large (upwards of 30 to 50 subjects per group, depending on the mean percentages scored as "1"). Table 1 thus summarizes the ANOVA results for these comparisons of changes in "success-rates" for pairs of correlated observations of the success/failures of the four models.]

Table 1 indicates that the demographic variables alone were able to discriminate between "bankcard-loyal" users, as the Rao's V statistic for Model 1 was significant at beyond the .05 level. While statistical significance for demographics was found, the practical significance was marginal, as indicated by two factors. Using a regression approach to discriminant analysis [Green & Tull, 1975], one can view the discriminant problem as regression with a binary dependent variable. For this 2-group problem, it can be shown that "R2" = 1 - Wilks' Lambda. Accordingly, one measure of the marginal practical significance of the discriminant function utilizing demographics is its Wilks' Lambda of .9175 or "R2'' = .0825.

A more traditional interpretation of practical significance is the percentage of group members who could be correctly classified by the predictor variables. To avoid the upward bias that would occur by using the same data to generate and then validate the discriminant function coefficients, it is necessary to cross-validate the function(s) by holding out data from the parameter-generation phase and then classifying these cases using parameters from the estimation sample. Rather than "waste" considerable data by running a conventional split-sample procedure, we used the "jackknife" option of the BMD Program. Under this option, one case is held-out while parameters are estimated from the N-1 data points. These parameters are used to classify this case, and a "hit" or a "miss" is scored. The case is then pooled with the others, while another one is extracted for validation and parameters are again estimated with a "new" N-1 cases. Repeating the procedure N times provides a complete test of the classification ability of the variables without ever using any to calculate their "own" function [Crask and Perreault, 1977]. In addition the weights for each estimate can be averaged to provide the same kind of stability that would have been sought by using the entire sample without "throwing away" data for cross-validation.

Our interest at this point is in describing the predictive validity of the demographic model, while the discriminant weights will later be used to examine explanatory power using Table 2, which describes the discriminating variables. The demographic variables alone correctly classified 56.6% of the users in the "jackknife" validations, while 50.0% would have been expected according to the proportional-chance criterion [Morrison, 1969]. While this represents a statistically significant percentage correctly classified, the large proportion of misclassification errors parallels the low R2 estimate and indicates that demographics alone would not provide very effective classification variables for credit card system choices [Hypothesis 1].

On the other hand, adding two quasi-demographic variables that measure consumer mobility significantly improved the discriminatory power of the set of demographic variables [Hypothesis 2]. Model 2 improved the "R2'' to .2514, and the change in Rao's V was significant at beyond the .01 level. Accordingly, the correct classification percentage improved 12.0% to 68.6% a change which was also significant at the .01 level. Explanatory power of the relationships was also improved, as will be noted below.

Model 3 augmented the two previous models by incorporating the stated importance which respondents attached to benefits provided by alternative credit card systems. Adding these benefits sought from credit cards substantially improved the ability to discriminate retail from bank card users, [Hypothesis 3]. The change in Rao's V was significant beyond the .01 level, and the "R2'' improved to .4422. Further, the percentage correctly classified in the jack-knife cross-validation jumped another 9.5% [significant beyond the .01 level], giving the combined model the ability to classify correctly 78.1% of the sample. The explanatory power of the three models will be discussed next.

TABLE 2

DESCRIPTION OF MODELS - DISCRIMINANT ANALYSIS

Table 2 provides data on the abilities of particular variables within each model to discriminate group membership, in a univariate sense, as well as their individual contributions to the overall Model 3 in a multivariate sense. Because of the presence of intercorrelations among some of the predictor variables, one must use caution in interpreting the relative importance of variables based on the traditional criterion of significance of the discriminant weights.

However, combining this information with the univariate 2-group comparisons between mean scores for variables (column 3) may provide some explanation for the credit card choices and marketing implications of this behavior.

For example, according to the ratio of discriminant weight divided by its standard error, the only demographic variable that provides significant discriminatory power, holding constant other demographics, mobility, and benefits sought, is household income. All things being equal, the higher the income, the more likely a respondent was to choose the retail card. Other data in the survey also show that retail card users were more frequent shoppers at this "top-line" department store than were bank card loyal users, which correlates with the importance of income as a discriminator. In fact, when frequency of shopping at the store is controlled for (by adding it to the discriminant function), the effect of income is no longer statistically significant. [For brevity, the analysis controlling for frequency of shopping is not reported here.]

In a managerial sense it is probably worth knowing that those preferring to use the retail card have higher income levels and are more frequent shoppers at the store. Heavy users of the retail credit cards "make good customers," especially when it is known that their income is high enough to lead to low credit losses. At the conceptual or theoretical level, however, the fact that bank card users who are equally frequent customers of the store have similar incomes to the retail card users suggests that there is nothing inherently discriminating about retail vs. bank cards' appeals to various income classes, when purchase rates are "controlled." Nevertheless, multicollinearity inhibits making these kinds of theoretical statements from these data, as the stability of the coefficients and the ability to "control" other variables is open to question.

For example, the significantly negative discriminant weight for the importance of reputation of prestige of the card indicates that, ceteris paribus, the lower the score on that variable, the more likely one is a bank card user. Given the coding of importance, a low score implies high importance for the prestige variable among bank card users. Whether this is in compensation for their lower income or, holding income constant, a greater need for the prestige of bank cards, or an artifact of unstable (due to multicollinearity) discriminant weights remains to be established in future research.

For the two mobility variables added to demographics to make Model 2, the univariate and multivariate inferences are more consistent. Retail card users as a group are longer-time residents in the survey city, as indicated both by the comparison of group means as well as the significantly negative discriminant weight (implying a lower score for the bank card users). A separate analysis showed that 31% of the bank card users were visiting the city (which also correlates with their lower frequency of store shopping) vs. 3% of the retail card loyal users. The discriminant coefficient is marginally significant in the same direction (1-tailed a = .06)

As is shown in Table 2 several of the benefits sought from credit card systems do possess a significant relationship to which type of credit card was used. As put forward in Hypothesis Three it was anticipated that persons using bank cards would place more importance on the attributes of ability to use a card at a wide variety of stores, the ability to use a card all over the country, and the consolidation of billing. It was also expected that persons who used the store-issued card as their usual mode of payment would place more importance on the ability to easily return merchandise and to straighten out incorrect billing. These expectations were entirely confirmed by the univariate t-tests with significance levels mixed between p # .05 and p # .01.

In addition to these expected differences, one which was not anticipated also appeared. This was the significantly greater emphasis users of store-issued cards placed upon the ability to utilize additional credit plans, [p # .01, 2-tailed test]. Given the presence of credit limits on bank cards, this finding is intuitively appealing.

When the benefits sought were added to the discriminant model containing the demographic and mobility-related variables, many of these univariate differences disappeared. This would appear to indicate that some benefits sought are closely associated with demographic or mobility-related characteristics of the individual, as we suggested in Hypothesis 2. Specifically, differences between bank card users and store-issued card users on evaluations of the ability to use a card at a wide variety of stores or all over the country were no longer significant once residency period and resident/ non-resident status had been controlled for [which effectively "removed" mobility as a variable]. A significance level of between p = .10 and .05 was still found for the ability to straighten-out bills, and differences between evaluations of ability to easily return merchandise, utilization credit plans, and consolidated billing were still significant in the same directions at the p # .01, and .01 levels, respectively.

SUMMARY AND CONCLUSIONS

This research was designed to investigate alternative models for predicting usage of retail store issued and bank issued credit cards. To develop these models a subsample was selected from the major sample on the basis of the following criteria: (1) that individuals possessed both types of cards, and (2) that individuals had prior experience in using both types of cards. The second restriction (made feasible by question 2, Exhibit 1) is especially relevant because a respondent may possess both types of cards, yet not be aware that bank credit cards could be used at the store under investigation. Of the three research hypotheses proposed:

H1: Was rejected at the p = .05 level. That is, -- there were statistically significant differences in demographic variables between users of bank cards and store-issued cards.

H2: Was confirmed at the p = .01 level. That is, there were discriminating residency differences between users of bank cards and store issued cards.

H3: Was confirmed at the p = .01 level. That is, there were discriminating benefit importance differences between users of bank cards and users of store-issued cards.

Examination of Tables 1 and 2 reveals that while the demographic variables are statistically significant, little practical significance (in terms of predictive power) can be gained from Model I. The addition of the two mobility variables in Model II provided both statistically and practically significant improvements in prediction. Similar improvements were also noted when variables reflecting the importance of benefits sought from credit systems, Model III, were added.

The individual variables that were significant in a multivariate sense (that is, by statistically controlling for the other variables) were: household income (demographic); length of residence in area (mobility); ease of merchandise return, utilization of additional credit plans, reputation/prestige of card, and consolidated billing (benefits sought). As discussed, the direction of the differences in group means were in agreement with the expectations cited earlier in the paper.

The results provided by this study may possess both theoretical and practical utility, as they illustrate the pattern of benefits sought with respect to credit systems by two diverse types of credit card users in a retail purchase setting. This may pave the way for a benefit segmentation approach which will be generally applicable to credit-facilitated exchange processes.

On a more theoretical level, the findings from the study suggest a need to explore the linkage between credit card system benefits sought and demographic and life style related variables. Such research may serve to develop a more substantial theoretical base for understanding credit system choices and credit-facilitated consumer purchasing behavior.

REFERENCES

William O. Adcock, Elizabeth C. Hirschman and Jac. L. Goldstucker, "Bank card Users: An Updated Profile," paper presented at the Association for Consumer Research Annual Conference, 1976.

R. V. Awh and D. Waters, "A Discriminant Analysis of Economic, Demographic, and Attitudinal Characteristics of Bank Charge-Card Holders: A Case Study," Journal of Finance, 29 (June, 1974), pp. 973-980.

Melvin R. Crask and William D. Perreault, Jr., "Validation of Discriminant Analysis in Marketing Research," Journal of Marketing Research, 14 (February, 1977), 60-8.

Jac L. Goldstucker and Elizabeth C. Hirschman, "Bank Credit Card Users: A New Market Segment for Regional Retailers," MSU Business Topics, 25, (Summer 1977), pp. 5-11.

Paul E. Green and Donald S. Tull, Research for Marketing Decisions, (Englewood Cliffs, NJ: Prentice-Hall, Inc. 1975), PP. 453-56.

Elizabeth C. Hirschman, Warren S. Blumenfeld, and Dwight Tabor, "An Attempt to Use Respondent Sex as a Moderator for Profiling Bank Card Users," Proceedings, Ninth Annual Conference, American Institute for Decision Sciences, Chicago, 1977, pp. 227-229.

William H. Mateer, The Checkless Society, MSU Business Studies, Bureau of Business and Economic Research, Michigan State University, East Lansing, Michigan, 1969.

H. Lee Matthews and John W. Slocum, Jr., "Social Class and Commercial Bank Card Usage," Journal of Marketing, 33 (January, 1969), pp. 71-78.

H. Lee Matthews, "Rejoinder to 'Social Class or Income?'" Journal of Marketing, 36 (January, 1972), pp. 69-70.

Donald G. Morrison, "On the Interpretation of Discriminant Analysis," Journal of Marketing Research, 6 (May, 1969), 156-63.

Joseph T. Plummer, "Life Style Patterns and Commercial Bank Credit Card Usage," Journal of Marketing, 35 (April, 1971), pp. 35-41.

Thomas Russell, The Economics of Bank Credit Cards, New York: Praeger Publishers, 1975).

Robert P. Shay and William C. Dunkelberg, Retail Store Credit Card Use in New York, Studies in Consumer Credit, No. 4, Graduate School of Business, Columbia University Press, New York, 1975.

John W. Slocum, Jr., and H. Lee Matthews, "Social Class and Income as Indicators of Consumer Credit Behavior," Journal of Marketing, 35 (April, 1975), pp. 59-74.

James B. Wiley and Lawrence M. Richard, "Application of Discriminant Analysis in Formulating Promotional Strategy for Bank Credit Cards," in Advances in Consumer Research, Vol. II, Mary Jane Schlinger, Editor, Association for Consumer Research, 1975, pp. 535-544.

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