Attribute Ratings As Predictors of Claimed and Actual Behaviors

ABSTRACT - Cash register receipts from grocery shopping trips and interview data about store attributes (price, location and quality) and behavior concerning grocery stores shopped at were collected from 80 respondents over five weeks. Results indicate a high correspondence between claimed and actual behavior and similar but not identical regression translations between attribute scores and claimed and actual behaviors.

Citation:

Penny Baron and Gerald Eskin (1976) ,"Attribute Ratings As Predictors of Claimed and Actual Behaviors", in NA - Advances in Consumer Research Volume 03, eds. Beverlee B. Anderson, Cincinnati, OH : Association for Consumer Research, Pages: 359-363.

Advances in Consumer Research Volume 3, 1976      Pages 359-363

ATTRIBUTE RATINGS AS PREDICTORS OF CLAIMED AND ACTUAL BEHAVIORS

Penny Baron, The University of Iowa

Gerald Eskin, The University of Iowa

ABSTRACT -

Cash register receipts from grocery shopping trips and interview data about store attributes (price, location and quality) and behavior concerning grocery stores shopped at were collected from 80 respondents over five weeks. Results indicate a high correspondence between claimed and actual behavior and similar but not identical regression translations between attribute scores and claimed and actual behaviors.

INTRODUCTION

Attribute models of consumer choice assume that goods consist of attributes or characteristics which satisfy consumer needs. Goods vary in the attributes they contain or are perceived to contain and consumers choose among them based on their preferences for the attributes. Consideration of such models leads one to inquire after the attributes that consumers want in particular choice situations, their preference for them, the attribute content of the various goods available in the situation and the relationship between these factors and choice. But how best to obtain these assessments is problematic. Empirically, use of marketplace data to inform multi-attribute models encounters serious problems. For example, variables of interest may not display adequate variation or relevant data may simply be unavailable. Observations are typically available only for aggregates of consumers. Thus, individual preference patterns are not revealed.

To directly assess individual consumers' preferences for attributes, most researchers turn to survey research or to laboratory experiments. In these settings, while data on individual attitudes and preferences can be obtained, marketplace behavior is not typically observable. Instead, one must settle for indicators of these behaviors, for example, overall attitude toward a product or respondents' claims about their behavior with respect to it.

Early applications of psychologists' work on attribute models focused on the tie between perceptions and evaluations of attributes and overall attitude or preference for the good in question. The presumption underlying this research was that such attitudes ultimately affected a consumer's marketplace choices. Subsequent work, however, has suggested that these relations, when they exist, can be expected to be quite weak. Overall attitude toward a good is only one of a number of factors affecting choices. Moreover, it was not quite clear in this early work how similar the relation between overall attitude and perceptions or evaluations of attributes was to the relation between these attribute responses and actual marketplace choices. Correspondence was often much less than perfect even under carefully controlled research situations in which respondent attitudes and behaviors were assessed under very similar conditions.

As a consequence, researchers have sought other variables which might tie beliefs and preferences concerning attributes more closely to actual behavior. Respondent claims about past and future behaviors constitute an important class of such variables. Recently, much research has been directed toward investigating the relation between various claimed behaviors with respect to past or future marketplace choices and beliefs or preferences with respect to relevant attributes.

This research has tended to follow the same paradigms as those used earlier to study the overall attitude variable. It attempts to assess relationships under conditions which are most likely to produce a strong association. For example, researchers try to obtain measures of self-reported behaviors and observations of actual behaviors within the same situational context; they try to keep the time interval between the two assessments to a minimum; and they seek to examine behaviors whose execution is entirely under the control of the consumer. For example, in this spirit Wilson, Mathews and Harvey (1975) obtained a rather elaborate series of paper and pencil measures designed to assess behavioral intentions to choose brands of toothpaste and various motivational and cognitive determinants of those intentions. Immediately upon completing the required questionnaire, each respondent was given an opportunity to choose a toothpaste from among those about which he had just been questioned. Needless to say, under these circumstances, the relation between stated intentions and actual behavior was high. About 85 percent of the respondents selected the brand with the "highest ranked intention to purchase." Similarly, when researchers control situational factors and other intervening events, they find that people do what they intend to do. (See, for example, Darrock, 1971.)

While results such as these are comforting because they suggest that there is a high correspondence between claimed and actual behavior, the generality of this result remains problematic- "Actual" behaviors under survey or laboratory conditions like those described above may not be the same as "marketplace" behaviors. In such controlled research settings, constraints are either not present or the researcher attempts to minimize their effects. But in the marketplace, behavior is constrained. Thus, the ties between attitudes or beliefs concerning attributes and behavioral choices in controlled research settings may not be the same as those which obtain for behaviors in the "marketplace."

A related issue concerns the type of observations used in measuring choice behavior. Typically, only one, or at best, a few choices are observed and the observations are all in terms of the same operational definitions, e.g., product or brand chosen. However, in the marketplace, numerous definitions of choice behavior are possible, e.g., average amount spent on a given occasion, frequency of purchasing a product in a given month, total expenditure for a product in a given year, number of units of a product purchased in a given period, etc. How attributes are perceived and evaluated with respect to these various behaviors is not necessarily the same. In short, perceptions and evaluations of attributes may vary depending not only on whether one is concerned about claimed purchase behavior, choices in a research setting or actual marketplace choices, but also on the particular marketplace behaviors observed. The constraints operating on different marketplace behaviors might vary.

Several questions arise, then, about these relationships:

1. Do claims about behavior forecast actual marketplace choices?

2. Do attribute perceptions that correlate with claimed behaviors also correlate with marketplace choices?

3. How do differences in behavior patterns relate to differences in the role played by various attributes?

4. How are different types of marketplace behaviors related to attribute perceptions?

The present study investigates these questions in the context of a field study of store selection. In this study we followed the grocery shopping behavior of 80 respondents over a five-week period. The research procedures involved both field interviews and observation of actual marketplace behavior. During the interviews several types of paper and pencil measures on attributes, store perceptions, preferences, attitudes and claimed behavior patterns were obtained. Objective records of actual shopping behaviors were obtained by regular collection of all sales receipts accumulated by each respondent during the five-week period. These receipts were treated as a source of direct measures of actual marketplace behavior.

Below we describe the store attributes, the response measures, and the procedures used to obtain them. Analysis procedures for assessing the relationships among the attribute ratings and the response variables are compatible with linear compensatory attribute models. Interpretations and methodological implications are discussed in terms of the questions posed above.

MEASUREMENT PROCEDURES

Sample

Respondents were selected using an area quota sampling procedure. Forty-three student interviewers were randomly assigned two 5-10 block sections of a small mid-western city. The sections spanned the entire city. Interviewers were instructed to obtain two eligible respondents using a standard door-to-door sampling procedure. An eligible respondent was a primary grocery shopper for a household which normally ate dinner at home.

Behavioral Measures

Out of the initial sample of 86, 6 withdrew from the study or failed to provide complete and usable data, e.g., they did not consistently complete records of their grocery shopping trips. Respondents who agreed to participate in the 5-week study filled out a record form for each grocery shopping trip they made and sealed all sales receipts from the trip in an envelope attached to the trip record form. These sales slips, together with corroborating information on the trip records about which stores were shopped at, constituted the source of measures on actual marketplace behavior.

The primary measures considered here are:

1. Store to which the most visits were made, the "most frequented" store.

2. Store at which the "most dollars" were spent.

Measures of Attributes and Perceptions

Interviews with respondents were conducted at the beginning of the study and at the end of the 1st, 3rd, and 5th week. Measures of respondent perceptions, and identification of relevant attributes were based on the responses obtained during these interviews.

1. Relevant Attributes. During the 2nd interview contact, respondents were asked which store they liked shopping at best and their reasons for liking it better than others. Similar questions were asked about the store liked "2nd best," "least," store "shopped at most frequently" and stores "not shopped at." These responses were content analyzed. Six attributes were identified as most important based on "frequency of mention" across the respondent population. They were:

Prices - How costly the store's products are considering both regular prices and specials.

Location - The amount of time required to travel to the store.

Efficiency - How quickly products can be found in. the store, checked-out and carried-away, for example, quick pick-up and easy parking.

Atmosphere - How pleasant it is to shop at the store, its appearance, cleanliness, and how friendly, helpful and courteous its personnel are.

Quality - Overall quality of the food sold.

Selection - The extent to which the store has the items wanted.

In the analyses reported here, only price, location and quality are considered. Price and location received much higher "frequency of mention" counts in the original content analysis than did the other 4 attributes. The smallest intercorrelations also obtained among price, location and the other 4 attributes. However, the intercorrelations among the remaining four attributes were relatively high. Of these, quality correlated least with the price and location attributes. Because it also seemed to be the most general of the four remaining attributes it was selected to represent them. Including the other three scales tends to make only marginal improvements in overall model fit. (Some comparisons of models specified in the three and six attributes are found in Section 1 of Appendix.) Factor analysis results were also consistent with this decision.

2. Perceived Attributes of Stores. Respondents were asked during the 3rd interview to rate each of the stores they had shopped at in terms of the above characteristics. They used a 5-point rating scale whose range of values went from "far above average" (5) to "far below average" (1). They were instructed to rate each store they were familiar with on each attribute in terms of their own impressions of the store. (See Section 2 of Appendix.)

3. Claimed Shopping Behaviors. On the 1st interview contact respondents were asked which store they shopped at most frequently and which they spent most of their food dollars at.

ANALYSIS PROCEDURES AND RESULTS

[We thank Stephanie Takazawa for her extensive contribution to the data processing and analysis phase of this work.]

Procedures

For each respondent, perceived attributes of stores and claimed and actual behaviors were coded separately by store. The actual behavioral measures were recorded from the trip record forms for the collection of stores actually shopped at during the 5 weeks. Measures on attributes and claimed behaviors were obtained during the interviews for the collection of stores with which the respondent claimed familiarity.

These procedures resulted in 3 groups of respondent-store records. The first group consists of "familiar only" respondent store records. It contains those stores which the respondents claimed they were familiar with but which they did not shop at during the research period. The second group, a "shopped familiar" group, consisted of records for stores that the respondent claimed to be familiar with and at which he actually did shop at least once during the research period. These two groups of data are sometimes combined into an "all familiar" set, i.e., all stores the respondents claimed to be familiar with, including those they shopped at and those they did not. This "all familiar" set of store records includes "0" observations on the behavior measures--is store most frequented and is store where most $ are spent?-for the stores where the respondent made no transactions. The third group of respondent-store records consists of stores which were shopped at during the study but with which respondents did not claim to be familiar. Thus no attribute ratings or attitude measures are available for this "shopped at only" group. Subsequent analysis did not include this set of data. (See Section 3 of Appendix for a discussion of the "shopped-at-only" respondent-store records.)

The principle analysis technique for the questions of interest here involved fitting multiple regression models which relate perceived store attributes to claimed and actual behavioral measures. The measure of perception of store attributes consists of respondents' ratings of the stores in terms of the price, location and quality attributes. Since the dependent variables can be treated as dichotomous (e.g., "most frequented store" and "all others") this approach is essentially similar to a discriminant analysis.

Implicit in this regression approach over the two previously defined data sets is the assumption that consumers possess identical and linear utility functions in the attributes considered but that they may differ in their perceptions of stores in terms of these attributes. It is clear in our case that these latter differences among consumers should exist. The best example is, of course, location. Differences in the relationship between the location of stores and of residence make it mandatory that attribute evaluations be considered a heterogeneous component. The homogeneity of the utility function across consumers may be questioned although previous empirical work has usually been unable to falsify it. (See, for example, Nakanishi and Bettman, 1974.) Here we maintain the assumption for simplicity. (In Section 4 of the Appendix evidence is given to support the assertion that this simplification does not substantially restrict the fit of the various models considered.)

Relation Between Claimed and Actual Behavior

Are the claims that people make about their consumption patterns consistent with their actual behaviors? An important assumption of most attribute model research based on questionnaire or interview data is that they are consistent. Violation of this assumption could affect interpretation of the role of attribute perceptions in determining marketplace choices. It may also affect interpretation of the role of consumers' attitudes toward their choice alternatives. Clearly, respondents' attitudes towards stores are predicated on their beliefs about what their actual shopping experiences have been. To the extent that what people do and what they claim to do are similar, variations between ties among marketplace behaviors and attributes and ties among claimed behaviors and attributes can be interpreted in terms of variations in the constraints surrounding each type of relation rather than in terms of respondent misperceptions of their past behavior or their inability to forecast future behavior.

The present study contains two sets of measures which enable a direct assessment of the relation between claimed and marketplace behavior. The measures are claimed vs. actual store which was "most frequented" and claimed vs. actual store at which "most food dollars'' were spent. As indicated in Table 1, seventy-two respondents reported that they had a single store that they frequented most often. Based on actual counts of store visits, 63 (87.5%) frequented this store more than any other during the 5-week test period. Five other respondents claimed that they tended to shop at a set of two stores most frequently. Of this group 4 out of the 5 actually shopped at one of these stores more than any other. Seventy-three respondents claimed that they spent more of their food dollars at one particular store than at any other. Of this group, there were 69 (94.5%) matches between the claimed and actual modal expenditure store. Again, 5 respondents claimed that there were 2 stores in their modal set and 4 out of the 5 had actual modal values that matched their claim.

TABLE 1

RELATION BETWEEN CLAIMED AND ACTUAL BEHAVIOR

These results suggest that there is reasonable consistency between these two types of measures even though the actual behaviors occurred subsequent to obtaining the claimed measures. Thus, these latter measures constituted a forecast of a future behavior pattern rather than a description of past behaviors. As such, their consistency is encouraging since the conditions surrounding the actual behaviors were those the respondents actually encountered in choosing stores to shop at during the five-week study period.

Relation Between Store Attributes and Claimed and Actual Behaviors

Here we consider the identification and measurement of the relationship between attribute ratings and the measures on store "most frequented" and store spent "most dollars" at. Table 2 compares regression coefficients and t statistics for models which relate the price, location and quality measures to the dichotomous frequency and expenditure measures. Results are shown separately for the claimed and actual measures of behavior and for the "shopped familiar" and "all familiar" sets of store records.

TABLE 2

RELATION OF STORE ATTRIBUTE RATINGS TO CLAIMED AND ACTUAL STORE "MOST FREQUENTED" AND STORE AT WHICH SPENT "MOST DOLLARS" (MULTIPLE REGRESSION COEFFICIENTS 8 AND T STATISTICS)

Comparison of the two measures of claimed behavior and the two measures of actual behaviors enables an assessment of the extent to which they are similarly related to perceived attributes of the stores. Comparisons are also possible with respect to the type of behavior at issue, frequency of visiting a store or amount spent at it.

With respect to the "most dollars" measure, good matches are obtained in both data sets.

Coefficient values are similar and the attribute rank orders are identical--price first, location second and quality last. The "t" statistics rank in a similar fashion. All tests on price and location are significant (p < .01) but those for quality are not. The only differences here between the results for actual and claimed behavior are that location is relatively more important than price in the translation to actual behavior than in the translation to claimed behavior.

For the "most frequented" measure, there is also some agreement across dependent measures but less than in the case of the dollar measure. Price and location again emerge as the terms with large and significant coefficients but their internal ranking is less similar. The results for claimed frequency indicate that price is the most important factor while the results for actual frequency indicate a stronger role for location. In both the "all familiar" and "shopped familiar" data sets, the regression coefficient of the price attribute is substantially lower for the actual behavior measure than for the claimed behavior measure. The t statistic parallels this shift. The coefficient for location is somewhat larger for actual than for claimed behavior and it becomes much more significant. Differences in results across samples appear slight for these data. Consideration of stores "not shopped at" does not seem to substantially alter the role of perceived attributes for any of the four measures.

One interpretation of these results is that consumers do not select a store for shopping trips on which they plan to spend a lot unless they perceive prices there to be good. For minor, convenience type trips which involve small expenditures, location increases in relative importance. This would explain the shift in coefficient size between the measures of actual "most frequented" and actual "most dollars." The higher coefficients of price for the "claimed" frequency measure may be attributed to consumers not making a clear separation between frequency and amount of expenditures when responding to the "claimed" question. It is also possible that they are unaware of their own choice patterns which appear to give more importance to convenient location and the related conservation of valuable time for minor shopping trips but to give them less weight when shopping trips involving large expenditures are being undertaken.

CONCLUSIONS

These data suggest that strong relationships do exist between what people claim they do and what they actually do. However, there is not a one-to-one correspondence, even for this relatively simple and important behavior, choice of a grocery store to shop at. While recall of such behavior is very good, and we suspect much better than for behaviors like "brand of chocolate bar last purchased" or "quantity of sugar consumed in a typical month," ambiguities do creep in.

The translation from attributes to claimed and actual behavior measures are similar but not identical. Differences in the consistency of results between claimed and actual store "most frequented" and claimed and actual store at which "most $ were spent," may be related to the differential clarity of the notion of "most frequented" as opposed to "most $." Asking about expenditures is a clearer question in the sense that it elicits consideration of shopping trips involving significant expenditures as opposed to "all trips" suggested by the "most frequented" measure.

In the absence of a clear understanding of how consumers process their perceptions of stores, all such conclusions must remain tentative.

APPENDIX

1. Columns 1 and 2, Table 3, compare the R2 coefficients for regression models specified to include all six attributes and the sub-set of 3 [prices, location and quality] considered in the body of the paper. Differences in the fit of these two models are small. However, the marginal contribution of the three additional attributes in reducing explained variance is statistically significant (p < .05) for most response measures. Thus it would be erroneous to conclude that only 3 attributes are operating here.

TABLE 3

R2 VALUES UNDER DIFFERENT MODEL SPECIFICATIONS FOR "ALL FAMILIAR" SAMPLE (N = 390)

2. Each respondent was also asked to rate each of these stores, using the same scale, in terms of his overall evaluation of it and to rate each store attribute in terms of its importance to him in choosing a grocery store to shop at. Respondents were also asked in the second interview to tell which store they "liked best" to shop at and which they liked to shop at "least." However, we do not treat these variables here.

3. About 50 percent of the respondents claimed to be familiar with each and every store that they actually shopped at. The remaining 50 percent failed to claim familiarity with about 40 percent of the stores they actually shopped at. Overall, then, for about 20 percent of the cases in which a respondent actually shopped at a store, he did not claim to be familiar with it. Visits to these stores accounted for about 13 percent of all visits made during the study period and for about 10 percent of the total dollar volume. These figures suggest that these stores are visited infrequently and that relatively few food dollars are spent at them.

Some portion of the respondents= failure to claim familiarity with one particular store may have been due to its closing during the research period. Some respondents visited this store only once. This was for its close-out sale.

4. With respect to differences among consumers in their utilities for the attributes, the question here is whether allowing for such differences improves the predictive value of the store attribute ratings. This was investigated by comparing 3 alternative procedures for weighting the relative contributions of each store-attribute rating. The first method employed standard regression techniques to produce the weights. The standard regression coefficients are estimates of the relative importance of the attributes, assuming that all consumers have similar utility functions. The 2^nd procedure involved weighting the store attribute ratings of each consumer by his self-reported importance ratings. In this version, store-attribute ratings were first transformed so that the scale values ranged from B2 to +2 rather than from 1 to 5, the average or "normal" level of an attribute now being coded "0" rather than "3." These attribute scores were then multiplied by their absolute importance ratings before subjecting them to regression analysis. The 3^rd method utilized relative importance scores to weight the transformed store-attribute ratings. Columns 3 and 4 of Table 3 present the R²=s obtained using each of these methods. Consistent with previous work on this issue, neither of the two procedures, which used weights based on individual consumer=s importance ratings of the attributes, resulted in more explained variation than the standard regression procedure. The magnitudes of the regression coefficients were also similar across the three procedures.

REFERENCES

R. I. Darroch, "Attitudinal Variables and Perceived Group Norms as Predictors of Behavioral Intentions and Behaviors in the Signing of Photographic Releases." Unpublished doctoral dissertation, University of Illinois, 1970.

M. Nakanish and J. R. Bettman, "Attitude Models Revisited: An Individual Level Analysis," Journal of Consumer Research, 1(1974), 16-21.

D. T. Wilson, H. L. Mathews and 3. W. Harvey, "An Empirical Test of the Fishbein Behavioral Intention Model," Journal of Consumer Research, 1(1975). 39-48.

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