Attitude Structure and Search:&Nbsp; an Integrative Model of Importance-Directed Information Processing

Morris B. Holbrook, Columbia University
David I. Velez, Columbia University
Gerard J. Tabouret, Columbia University
ABSTRACT - In spite of extensive research on multiattribute attitude models and information-acquisition paradigms, the linkage between these two important facets of buyer behavior is not yet well understood. Theory and previous studies suggest an integrative model of importance-directed information processing. This model proposes (1) that attribute importance is positively related to the extent and order of search and (2) that the resulting set of acquired cues serves as the basis for attitude. The results suggest both the feasibility of a questionnaire-based search task and the desirability of a more integrative approach to the study of attitude formation.
[ to cite ]:
Morris B. Holbrook, David I. Velez, and Gerard J. Tabouret (1981) ,"Attitude Structure and Search:&Nbsp; an Integrative Model of Importance-Directed Information Processing", in NA - Advances in Consumer Research Volume 08, eds. Kent B. Monroe, Ann Abor, MI : Association for Consumer Research, Pages: 35-41.

Advances in Consumer Research Volume 8, 1981      Pages 35-41

ATTITUDE STRUCTURE AND SEARCH:  AN INTEGRATIVE MODEL OF IMPORTANCE-DIRECTED INFORMATION PROCESSING

Morris B. Holbrook, Columbia University

David I. Velez, Columbia University

Gerard J. Tabouret, Columbia University

ABSTRACT -

In spite of extensive research on multiattribute attitude models and information-acquisition paradigms, the linkage between these two important facets of buyer behavior is not yet well understood. Theory and previous studies suggest an integrative model of importance-directed information processing. This model proposes (1) that attribute importance is positively related to the extent and order of search and (2) that the resulting set of acquired cues serves as the basis for attitude. The results suggest both the feasibility of a questionnaire-based search task and the desirability of a more integrative approach to the study of attitude formation.

INTRODUCTION

Extensive progress has recently been mode by consumer researchers in understanding the structure of attitudes that shape brand choices and in tracing the process by which information is acquired as a basis for preference formation. Too often, however, studies of attitude structure and information acquisition have proceeded independently. As a result, the connection between these two crucial phenomena has not yet been fully investigated.

More specifically, multiattribute attitude research has tended to focus primarily on the static structure of attitude while neglecting the process by which new informational inputs are incorporated into the determination of affect (Holbrook 1978). By contrast, research on information acquisition has typically dealt in great detail with the search process without systematically examining its presumed impact on brand evaluation (Holbrook and Maier 1978). In short, each approach has, in its own way, been somewhat one-sided. As a result, the linkage between attitude structure and information acquisition remains a vitally important area of consumer research with neglected potential for integration. Accordingly, the main purpose of the present paper is to propose and test a simple integrative model of information processing.

A MODEL OF IMPORTANCE-DIRECTED INFORMATION PROCESSING

Importance and Search

The Principle of Parsimony (Haines 1974) argues that limited cognitive capacity forces problem solvers to adopt heuristics restricting their attention to the minimum amount of data necessary for satisfying decisions (Bettman 1974; Payne 1976; Wright 1973, 1974). Plausibly, one such heuristic might be to select cues bearing on those aspects of the decision regarded as most important. Indeed, such a search strategy would be consistent with the lexicographic decision model's assumption that information on various brand attributes will be processed in order of their perceived importance (Payne 1976; Wright 1975).

The precise meaning of "attribute importance" as a hypothetical construct is an issue that has been hotly debated, but not fully resolved by consumer researchers (Cohen, Fishbein, and Ahtola 1972; Fishbein 1976; Myers and Alpert 1977; Olson, Kanwar, and Muderrisoglu 1979; Ryan and Holbrook 1979). One clear consensus, however, is that perceived importance must be distinguished from such related constructs as salience, determinance, and value-satisfaction or evaluative aspect. For the present purposes, the term "attribute importance" will be defined, in the sense suggested by Myers and Alpert (1977), as an attribute's "consequence or significance in making choices among brands or in forming overall evaluations or rankings of products" (p. 106).

When attribute importance is thus defined, its hypothesized role in directing information acquisition receives indirect support from several theoretical perspectives. It is reasonable, for example, to assume that an attribute's perceived importance will depend upon the range in relative magnitudes of the payoffs associated with its different possible levels. When uncertainty characterizes such outcomes, Bayesian decision theory suggests that search in problem solving will depend in part upon this degree of discrepancy among conditional payoffs (Edwards 1965). Similarly, Berlyne's (1960) theory of cognitive motivation relates the receivers degree of exploratory behavior To the strength or "importance" (p. 36) of the competing response tendencies aroused by a cue. Such competing response tendencies (e.g., approach versus avoidance) engender conflict, which is relieved by specific exploration of the stimulus array. The stronger or more important the response tendencies, the more extensive the resulting specific exploration. An analogous concept in consumer research is the conflict that occurs when the consumer feels low self-confidence in choosing between alternatives with potentially serious consequences. The theory of perceived risk (Cox 1967) argues that, where such uncertain choices involve "important" consequences, increased search for information is one possible coping strategy for reducing the resulting level of risk back down to a tolerable level.

Such propositions from Bayesian, Berlynian, and perceived-risk theory have typically been applied by consumer researchers at the brand or product-class level. Nevertheless, they may reasonably be expected to pertain with equal force to the acquisition of information about a brand's attributes. Such an hypothesized relation between attribute importance and search priority was first tested by Tigert (1966) and has since received increased empirical support in several more recent studies (Heeler, Okechuku, and Reid 1979; Holbrook and Maier 1978; Sheluga, Jaccard, and Jacoby 1979; Quelch 1978). Indeed, decision-net researchers have sometimes actually defined attribute importance as the adjusted average rank of an attribute in the sequence of information seeking (Bettman 1974; Jacoby, Chestnut, Weigel, and Fisher 1976; Nakanishi and Bettman 1974).

Importance and Attitude Structure

Numerous studies have documented the failure of attribute-importance weights to improve upon the predictive performance of summative multiattribute attitude models (for reviews, see has and Wilkie 1973; Cohen, Fishbein, and Ahtola 1972; Lutz and Bettman 1977). A second implication of the Principle of Parsimony, however, is that attribute importance might plausibly play another kind of role in multiattribute models by serving as the basis for a parsimonious information-processing heuristic in which cues are combined in a relatively simple, energy-conserving way by restricting attention to a small set of most-important attributes. In this light, many studies have shown that a very simple model of attitude structure, summing only the scores on those attributes considered the few most important, predicts affect at least as well as more complex weighted forms of the multiattribute model (Bass and Wilkie 1973; Holbrook 1978; Holbrook and Hulbert 1978; Nakanishi and Bettman 1974; Wilkie, McCann, and Reibstein 1973). Attribute importance therefore appears to serve not so much as a weight in a complex multiattribute model, but rather as a factor directing the simplification of attitude structure.

An Integrative Model

The foregoing considerations may be summarized by the integrative model of importance-directed information pro-ceasing shown in Figure 1.

FIGURE 1

A MODEL OF IMPORTANCE-DIRECTED INFORMATION PROCESSING

This model emphasizes the proposed role of attribute importance in linking information acquisition with attitude structure. According to Arrow (1), when confronted with the need to evaluate unfamiliar versions of some established product, the decision maker should seek information on those attributes viewed as most important. Arrow (2) then indicates that the perceived desirability values of the limited set of cues thus acquired should serve as the basis for appraising competing versions. Thus, the much-publicized debate concerning the role of attribute importance (e.g., Bass and Wilkie 1973) is resolved by a dynamic model hypothesizing that importance guides the acquisition of information that, in turn, determines the structure of attitude.

As indicated earlier, some empirical support already exists for the hypothesized relationships represented by Arrows (1) and (2). However, with few exceptions (e.g., Holbrook and Maier 1978; Sheluga et al. 1979), such support appears to be largely piecemeal in nature--focusing on one relationship or the other, but not on both at the sale time. Thus, there is little firm evidence for the simultaneous operation of both arrows as part of a dynamic process in which importance guides the acquisition of cues that then compose the structure of attitude. Therefore, because this plausible, but inadequately tested integrative hypothesis appears to provide a useful way of linking research on information acquisition with that on attitude structure, the chief purpose of the present study was to provide a clearly-focused test of the importance-directed model of information processing.

Methods for Studying Information Acquisition

Methods for studying the acquisition of information have recently been the topic for extended reviews, comparisons, and evaluations (Arch, Bettman and Kakkar 1978; Bettman 1977; Chestnut and Jacoby 1978; Jacoby, Chestnut, Hoyer, Sheluga and Donahue 1978; Payne and Ragsdale 1978; Russo 1978). Briefly, the most frequently advocated approaches for investigating search processes involve the use of: (1) eye-fixation tracking, in which mechanical instrumentation monitors the eye movements of subjects scanning some visual field; (2) verbal protocols, in which the subject provides a running account of his conscious thought processes while performing a decision task; or (3) information-display boards, in which the subject selects cards containing information from a multi-celled brands-by-attributes matrix.

One seldom-mentioned limitation of most such approaches is their implicit restriction to the confines of a laboratory or other closely monitored setting in which individual sessions are conducted with what, for cost reasons, usually turns out to be a small convenience sample. Given the need for marketing research on large, representative samples of consumers (Ferber 1977a), the standard repertoire of techniques for studying information acquisition may prove inadequate to the tank of obtaining generalizable results. Toward this end, Holbrook and Maier (1978) introduced a modification of the information-display board intended to be usable in a mail questionnaire and, therefore, suitable for inclusion in studies employing large-scale survey techniques. Briefly, this procedure presents the respondent with a brands-by-attributes information-display sheet in which each cell is covered by a numbered sticker. The respondent acquires information by peeling stickers from the matrix and replaces them at the bottom of the page in a manner that provides a permanent record of the order in which they were removed, Holbrook and Maier (1978) found that this task was easily understood and performed by student subjects. A subsidiary purpose of the present study was to provide a further test of the usefulness of this questionnaire-based information-acquisition task.

Previous Findings Concerning the Importance-Directed Information-Processing Model

The Holbrook-Maier study.  Using the sticker-pulling task in a study simulating the evaluation of phonograph records, Holbrook and Maier (1978) found evidence consistent with the importance-directed information-processing model. Across six attributes, there were highly significant mean intra-individual correlations between rated attribute importance and (1) number of cues selected on that attribute (r = .59, p < .0001) and (2) average sequential order of selecting cues on that attribute (r = .63, p < .0001). Using desirability ratings of the acquired cues in multi-attribute models to predict preference produced equally good intra-individual rank-order correlations whether additive (r = .57 and .56, p < .0001) or averaging (r = .54 and .55, p < .0001) models were employed. This finding was interpreted as exploratory evidence bearing on the debate between adding versus averaging formulations (Anderson 1967, 1976; Bettman, Capon and Lutz 1975; Fishbein 1976; Fishbein and Ajzen 1978; Lutz 1976; Troutman and Shanteau 1976). Apparently, when selective infatuation acquisition is permitted rather than forcing exposure to factorially-designed stimuli, averaging models may not enjoy any predictive superiority over additive versions.

The Quelch study.  Quelch (1978) produced support for the first Holbrook-Maier finding by asking housewives to choose among six brands of cold cereal, after acquiring information on either four or five attributes. The mean intra-individual rank-order correlation between the order of information selection and subsequent measure of attribute importance was r = .71.

The Heeler study.  Heeler, Okechuku, and Reid (1979) contributed further support for the first Holbrook-Maier finding. Two groups of student subjects provided importance ratings or display-board search measures for 10 attributes of electric blenders. Rank-order correlation of aggregate scores across attributes found a moderately strong relationship between average rated importance and mean frequency of acquisition: r = .53 (p < .10).

The Sheluga study.  Finally, in a study performed concurrently with that reported in the present paper, Sheluga, Jaccard, and Jacoby (1979) extended Holbrook and Maier's (1978) "integrative approach" to include a comparison of several competing scaling methods. Though several of their findings concerning the evaluation of cameras are of considerable interest in their own right, the two most relevant to the issues addressed here are: (1) significant average Spearman correlations between retrospective "search importance" and extent (r = .80, p < .001) and order (r = .68, p < .001) of attribute-specific information acquisition; (2) relatively strong choice predictions (percentage correct) based upon acquired cues scored for utility by rating scales (641), conjoint measurement (701), and graded paired comparisons (100%).

Limitations in the Previous Findings

One limitation in Holbrook and Maier's (1978) design vas the fact that questionnaires were distributed by hand, thus raising questions concerning the usefulness of the sticker-pulling technique in mail-survey research. A subsidiary objective of the present study, therefore, was to establish the feasibility of this method for use in surveys that are distributed through the postal system.

A second weakness in the Holbrook-Maier study raised a possible alternative hypothesis to explain their finding concerning the effect of attribute importance on search direction. Specifically, attributes were inadvertently arrayed on the information-display sheet from left to right roughly in order of their overall average perceived importance. In accord with the well-established effect of task format on performance in information-acquisition strategies (Bettman and Kakkar 1977), it might therefore be argued that the relationship between attribute importance and cue selection could have resulted from respondents' tendencies to move from left to right in searching for information (Quelch 1978). Accordingly, another subsidiary objective of the present study was to rule out this alternative explanation by randomly rotating the order of the attribute columns in the information-display sheet.

A further limitation in the Holbrook-Maier design raised doubts concerning the apparent equality of predictive performance by additive and averaging attitude models. Since respondents selected information by self-exposure, it is possible that they tended to acquire the same number of cues for each stimulus object. Such a naturally-occurring lack of variation in the extent of search among objects could account for the absence of predictive differences between additive and averaging models since the two kinds of formulation give identical results when the number of cues processed on each stimulus is constant (Anderson 1967; Fishbein and Ajzen 1975). The present study therefore attempted, as a third subsidiary objective, to encourage variation in the number of cues chosen on various stimulus objects by constraining the respondent to choose exactly 24 cues in evaluating 8 objects. Thus, if a number of cues other than three were chosen on any one stimulus, some variation in cue utilization among objects would be insured.

METHOD

To maintain comparability with previous research using the sticker-pulling task, phonograph records were chosen as the product category to be tested. In accord with a suggestion by Ferber (1977b), the sample was composed of first year MBA students who demonstrated an interest in this product class by virtue of having purchased at least two pop or jazz recordings within the past six months. Questionnaires were mailed to 136 potential respondents, of whom 101 returned complete and usable sets of data for a response race of 74%.

Evaluation Task

Respondents were asked to evaluate eight vocal recordings that differed on six alphabetically listed attributes. Each attribute was defined by the following set of dichotomous characteristics: (1) Jacket--informative or visual; (2) Label--major or independent; (3) Singer's Style--traditional or contemporary; (4) Type of Music--pop or jazz; (5) Type of Production--studio or live; (6) Type of Songs--standards or originals and recent hits. These 6 attributes and 12 defining characteristics were described on the introductory page of the questionnaire by short paragraphs (omitted here for brevity) and were identical to those used by Holbrook and Maier (1978), except that Price in their study was replaced by Type of Music to avoid the somewhat ambiguous implications of price for preference formation.

Attribute-Specific Measures of Importance and Desirability

The six attributes were rated on 7-point check-mark scales for their degree of importance to the respondent's evaluation of a vocal recording from "not at all important" To "extremely important." In addition, the 12 defining characteristics were rated on comparable scales for their desirability in a vocal album from "extremely undesirable" to "extremely desirable." Importance ratings were scored from 1 to 7 while, in acknowledgement of the debate surrounding the proper coding of evaluative scores (Fishbein 1976; Holbrook 1977; Lutz 1976), two alternative codings of the desirability ratings were investigated: -3 to +3 and 1 to 7.

Test Objects

The dichotomous characteristics for each attribute were used to specify eight vocal albums according to the kind of fractional factorial design described by Green (1974) as suitable for conjoint measurement studies. The rows of the resulting 8 x 6 matrix were randomized and labeled from "Record A" to "Record H." To guard against the possibility of any position effects on the extent or order or cue acquisition (Bettman and Kakkar 1977; Quelch 1978), the columns of this basic information matrix were then arranged into 10 different random sequences. These 10 sequences were, in turn, assigned randomly to respondents.

The Information-Acquisition Task

As a basis for evaluating the vocal albums, the questionnaire provided an information-display sheet consisting of one of the 10 randomly-arrayed matrices with each cell covered by an appropriately labeled the 24 stickers that respondent was instructed to remove the 24 stickers that would best provide the information needed to rank the eight records in order of preference. Further instructions asked the respondent to replace each sticker, in its order of removal, into another matrix entitled "Order of Sticker Removal" with cells labeled from "1st" to "24th." Following this task, the respondent ranked the eight recordings in order of preference.

Operational Definitions

Extent and order of attribute-specific information acquisition.  The extent of attribute-specific information acquisition was defined operationally as the number of cues acquired across all eight records on that attribute. The order of acquisition was defined as the average rank of selecting information on that attribute across the eight records (with the rank order for non-acquired cues scored arbitrarily as 36.5--i.e., half way between 25th and 48th.)

Models of attitude structure.  The full set of competing attitude-structure models developed by Holbrook and Maier (1978) was tested in the present study. Specifically, the comparative predictive performance of additive and averaging models was explored, both with and without importance weights and with desirability scored both from -3 to +3 and from 1 to 7. In accord with the previous findings, however, no discernible differences arose from the inclusion of importance weights or from the alternative scorings of desirability. To save space, the present discussion will therefore be confined to the most conceptually appropriate attitude models--namely, those without importance weights and with desirability scored from -3 to +3 (Fishbein 1976; Fishbein and Ajzen 1975; cf. Rosen-berg 1956). With this restriction, the competing models investigated may be represented as follows:

Partial Additive Model  = EQUATION  (1)

Partial Averaging Model =  EQUATION  (2)

Full Additive Model =  EQUATION  (3)

Full Averaging Model =  EQUATION  (4)

where k refers to the 12 characteristics that s given record does (k = 1, ..., 6) and does not (k = 7, ..., 12) possess; Bk=1 if the record possesses characteristic k or Bk = -1 if it does not; Ek is the respondent's evaluation of the desirability of characteristic k; and ACQk is a zero-one dummy variable representing the acquisition of information on characteristic k.

More colloquially, the Partial Additive Nodal nay be described as the summed desirability scores of those characteristics known, after search, to be present in a given record album (where it is assumed that the respondent believes the information he has received to be true). The Partial Averaging Model is the mean desirability of those known characteristics. The Full Additive Model further subtracts the desirabilities of chose characteristics revealed to be absent from a given recording (on the assumption that the respondent makes inferences about what characteristics are missing from the record). And the Full Averaging Model computes the mean desirability of the present characteristics less that of those absent.

The full models assume that preference may be enhanced (reduced) by the absence of undesirable (desirable) characteristics, thereby incorporating logic analogous to that arguing for the bipolar coding of components in the Fish-been model (Fishbein 1976; Fishbein and Ajzen 1975; cf. Rosenberg 1956). Even these fuller versions of the attitude model do not, of course, account for the effects of additional beliefs that may be inferred on the basis of the information that has been acquired (Fishbein and Ajzen 1975; Lutz and Swasy 1977; Olson 1978). This, of course, is a limitation that is shared by the vast preponderance of attitude research.

In comparing the additive and averaging models, it is important to recognize that any differences between them in the present design result from the fact that respondents are free to select different numbers of cues for different vocal albums. Where the variation among records in the extent of acquisition is relatively large, the two types of formulation may make substantially different preference-rank predictions. The degree of such differences is an empirical issue, however, and not part of the research design controlled by the investigators themselves. Thus, this part of the study is correlational rather than experimental in nature.

Operational Hypotheses

Given the arguments and findings reviewed earlier, four specific hypotheses were tested:

H1- Extent and order of attribute-specific information acquisition are positively related to attribute importance on an intra-individual basis

H2 - Attitude-structure models based on the cues acquired predict intra-individual preference ranks

H3 - The partial attitude models perform at least as well as the full models in predicting preference ranks

H4 - The predictive performance of additive attitude models is at least as good as that of averaging models

RESULTS

In accord with the findings of Holbrook and Maier (1978), respondents appeared to experience few problems with the self-administration of the mail questionnaire. Only three questionnaires had to be discarded because of failures to follow the written instructions. The likely effect of such difficulties on the results obtained in an information-acquisition task therefore appears to be minimal.

H1. The first hypothesis was supported by highly significant mean frieze-individual correlations of attribute importance with (1) extent of information acquisition (r = .665, Zr = 21.08, p < .0001) and (2) order of information acquisition (r = .674, Zr = 22.81, p < .0001). Because of the presence of a small number of low-correlation outliers, the median intra-individual correlations were considerably stronger than the means: .775 and .782 respectively. These results compare favorably with the previously cited findings.

H2. The results for all four versions of the attitude-structure model supported the second hypothesis by parroting fairly well in predicting preference rank, with mean intra-individual rank-order correlations as follows: Partial Additive Model--r = .612 (Z = 19.63, p < .0001); Partial Averaging Model -- r = .602 (Z = 16.91, p < .0001); Full Additive Model--r = .598 (Z = 7.73, p < .0001); Full Averaging Model--r = .597 (Z = 16.86, p < .0001). Because of skewed distributions, the median correlations were again somewhat stronger than the means, ranging from .673 to .756. These relationships were, if anything, somewhat better than those found by Holbrook and Maier.

H3. In further accord with the Holbrook-Maier results, the third hypothesis was supported by the failure of either full attitude-structure model to perform better than its corresponding partial version. This finding contradicts the assertion that affect is determined, in part, by the evaluative aspects of those characteristics that are believed to be missing from some object.

H4. Finally, in line with previous findings and the fourth hypothesis, neither averaging model improved upon the predictive performance of its additive counterpart. It appears, however, that constraining respondents to the selection of 24 pieces of information was only moderately successful in encouraging variation in the number of attribute cues acquired across stimulus objects. Thus, 19 out of 101 respondents shoved no variation in the number of cues selected across recordings. As pointed out earlier, for such respondents, additive and averaging attitude models make identical preference predictions. The removal of these 49 respondents from the sample would only result, however, in an even greater relative superiority in the predictive performance of the Partial Additive Model versus the Partial Averaging Model (r = .622 versus .542). Such a procedure, therefore, would not change the basic conclusion that averaging models do not improve upon the predictive performance of additive versions. Accordingly, the more conservative comparison, based on the full sample of 101 respondents, should be used in summarizing the results for H3 and H4.

In this light, the findings concerning H3 and H4 may be summarized by comparing the predictive performance of the Partial Additive Model with that of the other three versions. In each case, the Partial Additive Model performed slightly, but significantly better in predicting preference: r = .642 versus .602 (tl00 = 3.09, p < .01); r = .642 versus .598 (t100 = 3.02, p < .01); and r = .642 versus .597 (t100 = 2.89, p < .01). It therefore appears that, in the kind of information-acquisition task investigated here, the Partial Additive Model is to be preferred over the full and/ or averaging versions--partly because of its marginal predictive superiority, but primarily because of its greater parsimony.

DISCUSSION

Limitations

Like most presently available research on attitude formation or information acquisition, the present study is characterized by certain limitations. The attitude model, for example, required the plausible, but untested assumption that the respondents believed the information to which they were exposed and, in effect, assigned each cue a subjective likelihood of 1.0. Also, the model could not account for the effects of any additional beliefs that might have been inferred on the basis of the cues acquired (Fishbein and Ajzen 1975; Lutz and Swasy 1977; Olson 1978). In these respects, however, the present approach shares assumptions that are fundamental to the vast preponderance of attitude research, including most of the work on conjoint analysis and information integration. Moreover, the importance of inferential beliefs in this study is called into question by the absence of any predictive improvement to be gained by including characteristics implicitly revealed to be absent from the recording {i.e., Ek= -1).

In common with other information-acquisition paradigm (eye-fixation tracking, display boards, protocols, etc.), the present search task was highly artificial in many respects. Restricting respondents to the use of 24 cues, for example, was one way of setting a reasonable limit on the amount of information acquired so as to permit some variance in cue utilization across attributes. Other methods for limiting cue selection--such as charging five cents per cue-- would have been less practical in a mall questionnaire. Moreover, it appears doubtful whether any one way of restricting information intake is inherently more realistic than another.

In contrast to many applications of the information-display board, the cues uncovered by respondents in the present study remained visible and did not, therefore, place any demands on memory capacity. The authors would argue that this procedure has the advantage of removing the confounding effects of memory overload, but some might feel that this constitutes a departure from realism.

Further, the recordings employed in the present study were hypothetical rather than real, though respondents were not informed of the fictitious nature. The advantage of using such artificial products is that it permits an otherwise unattainable degree of experimental control. Accordingly, this kind of artificiality appears in numerous other studies of information processing--including many of those on cognitive algebra (e.g., Bettman, Capon, and Lutz 1975), conjoint measurement (e.g., Green 1974), and attitude change (e.g., Lutz 1975).

Finally, it should be noted that--though the approach used here is of soma use in comparing the predictive efficacy of simple additive versus averaging models (Fishbein and Ajzen 1975)--it cannot discriminate between an additive model and the more elaborate weighted averaging model proposed by Anderson (1967) to take account of the "set size" effect. This latter formulation assumes that affect begins at an initial value that is then adjusted by the averaging-in of subsequent information. Such a model produces predictions that are quite similar to those generated by a simple additive representation. As pointed out by Fishbein and Ajzen (1975, p. 242), the inability to distinguish between a simple additive model and Anderson's (1967) differentially weighted averaging version is a limitation shared by most of the research in this area.

CONCLUSIONS

Subject to such limitations, the present study supports the proposed model of importance-directed information processing. Specifically, its results suggest (1) that attribute importance guides the acquisition of attribute-related cues and (2) that these cues are then incorporated into a relatively simple importance-shaped attitude structure so as to determine affect. Though both these effects have been supported independently by earlier studies, previous research has tended not to treat them together as parts of one dynamic importance-directed process. The present findings therefore suggest the usefulness of research attempting to integrate information-acquisition phenomena with the nature of multiattribute attitude structure.

In addition, the present study continues to support the adequacy of the relatively parsimonious Partial Additive Model when compared with more elaborate competing versions. To be sure, the research design left the respondent free to select among the six cues available for each object instead of being exposed to factorially-arranged combinations of cues so that the comparative tests of predictive performance are correlational rather than experimental in nature. Moreover, like most other research on consumer attitudes, the present study failed to account for the effects of additional beliefs that might be inferred from those explicitly manipulated and/or measured. Nevertheless, within this restricted framework, the findings may further reinforce the doubts of those questioning the often-claimed superiority of the averaging model (Fishbein 1976; Fishbein and Ajzen 1975; cf. Anderson 1967, 1976; Betimes, Capon, and Lutz 1975; Lutz 1976; Troutman and Shanteau 1976).

Finally, the study confirms the feasibility of the sticker-removal task as a device for incorporating the information-acquisition paradigm into mail-survey research designs. Though the respondents in the present study were students, the authors would anticipate no difficulties in applying the technique in mail questionnaires directed at larger scale, more representative samples. Obviously, the external validity of such an approach remains to be tested. But, on the basis of the evidence presented here, it appears that the label-pulling task may be useful in future attempts to generalize findings concerning information acquisition to the overall consumer population.

REFERENCES

Anderson, Norman H. (1967), "Averaging Model Analysis of Set-Size Effect in Impression Formation," Journal of Experimental Psychology, 75, 158-65.

Anderson, Norman H. (1976), "How Functional Measurement Can Yield Validated Interval Scales of Mental Quantities," Journal of Applied Psychology, 61, 677-92.

Arch, David C., Bettman, James R., and Kakkar, Pradeep (1978), "Subjects' Information Processing in Information Display Board Studies," Proceedings, Association for Consumer Research, Eighth Annual Conference, 555-60.

Bass, Frank M., and Wilkie, William L. (1973), "A Comparative Analysis of Attitudinal Predictions of Brand Preference," Journal of Marketing Research, 10, 162-9.

Berlyne, Daniel E. (1960), Conflict, Arousal, and Curiosity, New York: McGraw-Hill.

Bettman, James R. (1974), "Toward a Statistics for Consumer Decision Net Modeler" Journal of Consumer Research, 1, 71-80.

Bettman, James R. (1977), "Data Collection and Analysis Approaches for Studying Consumer information Processing," Proceedings, Association for Consumer Research, Seventh Annual Conference, 342-8.

Bettman, James R., Capon, Noel, and Lutz, Richard J. (1975), "Multiattribute Measurement Models and Multiattribute Attitude Theory: A Test of Construct Validity," Journal of Consumer Research, 1, 1-14.

Bettman, James R. and Kakkar, Pradeep (1977), "Effects of Information Presentation Format on Consumer Information Acquisition Strategies," Journal of Consumer Research, 3, 233-40.

Chestnut, Robert W., and Jacoby, Jacob (1978) "Methods and Concepts in Consumer Information Processing: Toward an Integrated Framework," Research Paper No. 141a, Columbia University.

Cohen, Joel B., Fishbein, Martin, and Ahtola, Olli T. (1972), "The Nature and Uses of Expectancy-Value Models in Consumer Attitude Research," Journal of Marketing Research, 9, 456-60.

Cox, Donald F., ed. (1967), Risk Taking and Information Handling in Consumer Behavior, Cambridge: Harvard University Graduate School of Business Administration.

Edwards, Ward (1965), "Optimal Strategies for Seeking Information: Models for Statistics, Choice Reaction Times, and Human Information Processing," Journal of Mathematical Psychology, 2, 312-29.

Father, Robert (1977a), "Research By Convenience," Journal of Consumer Research, 4, 57-8.

Father, Robert (1977b), 'What is the JCR Editorial Policy on Samples and Sample Requirements?" Newsletter, Association for Consumer Research, 7, No. 4, 16.

Fishbein, Martin (1976), "Extending the Extended Model," Proceedings, Association for Consumer Research, Sixth Annual Conference, 491-7.

Fishbein, Martin and Ajzen, Icek (1975), Belief, Attitude; Intention, and Behavior: An Introduction to Theory and Research, Reading, Mass.: Addison-Wesley.

Green, Paul E. (1974), "On the Design of Choice Experiments Involving Multifactor Alternatives," Journal of Consumer Research, 1, 61-8.

Haines, George H., Jr. (1974), "Process Models of Consumer Decision Making," in Buyer/Consumer Information Processing, ed. G. David Hughes and Michael L. Ray, Chapel Hill: University of North Carolina Press.

Heeler, Roger M., Okechuku, Chike, and Reid, Stan (1979), "Attribute Importance: Contrasting Measurements," Journal of Marketing Research, 16, 60-3.

Holbrook, Morris B. (1977), "Comparing Multiattribute Attitude Models By Optimal Scaling," Journal of Consumer Research, 4, 165-71.

Holbrook, Morris B. (1978), "Beyond Attitude Structure: Toward the Informational Determinants of Attitude," Journal of Marketing Research, 15, 545-56.

Holbrook, Morris B. and Hulbert, James M. (1975), "Multi-Attribute Attitude Models: A Comparative Analysis," Proceedings, Association for Consumer Research, Fifth Annual Conference, 375-88.

Holbrook, Morris B. and Meier, Earl A. (1978), "A Study of the Interface Between Attitude Structure and Information Acquisition Using a Questionnaire-Based Information-Display Sheet," Proceedings, Association for Consumer Research, Eighth Annual Conference, 93-8.

Jacoby, Jacob, Chestnut, Robert W., Hoyer, Wayne D., Sheluga, David A., and Donahue, Michael J. (1978), "Psychometric Characteristics of Behavioral Process Data: Preliminary Findings on Validity and Reliability," Proceedings, Association for Consumer Research, Eighth Annual Conference, 546-54.

Jacoby, Jacob, Chestnut, Robert W., Weigl, Karl C., and Fisher, William A. (1976), "Pre-Purchase Information Acquisition: Description of a Process Methodology, Research Paradigm, and Pilot Investigation," Proceedings, Association for Consumer Research, Sixth Annual Conference, 306-14.

Lutz, Richard J. (1975), "Changing Brand Attitudes Through Modification of Cognitive Structure," Journal of Consumer Research, 1, 49-59.

Lutz, Richard J. (1976), "Conceptual and Operational Issues in the Extended Fishbein Model," Proceedings, Association for Consumer Research, Sixth Annual Conference, 469-76.

Lutz, Richard J., and Bettman, James R. (1977), "Multi-attribute Models in Marketing: A Bicentennial Review," in Consumer and Industrial Buyer Behavior, ed. Arch C. Woodside, Jagdish N. Sheth, and Peter D. Bennett, New York: North-Holland.

Lutz, Richard J. and Swasy, John L. (1977), "Integrating cognitive structure and Cognitive Response Approaches to Monitoring Communications Effects," Proceedings, Association for Consumer Research, Seventh Annual Conference. 363-71.

Myers, James R., and Alpert, Mark I. (1977), "Semantic Confusion in Attitude Research: Salience Vs. Importance Vs. Determinants," Proceedings, Association for Consumer Research, Seventh Annual Conference, 106-10.

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

Olson, Jerry (1978), "Inferential Belief Formation in the Cue Utilization Process," Proceedings, Association for Consumer Research, Eighth Annual Conference, 706-13.

Olson, Jerry, Kanwar, Rajesh, and Muderrisoglu, Aydin (1979), "Clarifying the Confusion Regarding Salience, Importance, and Determinants Concepts in Multiattribute Attitude Research," Working Paper, The Pennsylvania State University.

Payne, John W. (1976), "Heuristic Search Processes in Decision Making," Proceedings, Association for Consumer Research, Sixth Annual Conference, 321-7.

Payne, John W. and Ragsdale, E.K. Easton (1978), "Verbal Protocols and Direct Observation of Supermarket Shopping Behavior: Some Findings and a Discussion of Methods," Proceedings, Association for Consumer Research, Eighth Annual Conference, 571-7.

Quelch, John A. (1978), "Behavioral Measurement of the Relative Importance of Product Attributes: Process Methodology and Pilot Application," Working Paper, University of Western Ontario.

Rosenberg, Milton J. (1956), "Cognitive Structure and Attitudinal Affect," Journal of Abnormal and Social Psychology, 53, 367-72.

Russo, J. Edward (1978), "Eye Fixations Can Save the World: A Critical Evaluation and a Comparison Between Eye Fixations and Other Information Processing Methodologies," Proceedings, Association for Consumer Research, Eighth Annual Conference, 561-70.

Ryan, ,Michael J., and Holbrook, Morris B. (1979), "On the Relationship of Importance, Elicitation Order, and Expectancy X Value," Working Paper, Columbia University.

Sheluga, David A., Jaccard, James, and Jacoby, Jacob (1979) "Preference, Search, and Choice: An Integrative Approach," Journal of Consumer Research, 6, 166-76.

Tigert, Douglas J. (1966), "An Investigation of Experimental Techniques to Determine the Relative Importance of Various Product Attributes in the Consumer Decision-Making Process," Working Paper, Krannert Graduate School of Industrial Administration, Purdue University.

Troutman, C. Michael, and Shanteau, James (1976), "Do Consumers Evaluate Products by Adding or Averaging Attribute Information?," Journal of Consumer Research, 3, 101-12.

Wilkie, William L., McCann, John M., and Reibstein, David J. (1973), "Halo Effects in Brand Belief Measurement: Implications for Attitude Model Development," Proceedings, Association for Consumer Research, Fourth Annual Conference, 280-90.

Wright, Peter L. (1973), "The Cognitive Processes Mediating Acceptance of Advertising," Journal of Marketing Research, 10, .53-62.

Wright, Peter L. (1974), "The Harassed Decision Maker: Time Pressures, Distractions, and the Use of Evidence," Journal of Applied Psychology, 59, 555-61.

Wright, Peter L. (1975), "Consumer Choice Strategies: Simplifying Vs. Optimizing," Journal of Marketing Research, 12, 60-7.

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