Representing Attitude Structure: Issues and Evidence

ABSTRACT - The representation of attitude structure and associated causal issues has been a topic of considerable interest recently. This research examines whether variations in outcome evaluation produce corresponding variations in attitude, the discriminant validity between the cognitive and affective components of attitude, and the accuracy of a "cognition determines affect" causal ordering.

Citation:

Paul W. Miniard, Thomas J. Page, Jr., April Atwood, and Randall L. Rose (1986) ,"Representing Attitude Structure: Issues and Evidence", in NA - Advances in Consumer Research Volume 13, eds. Richard J. Lutz, Provo, UT : Association for Consumer Research, Pages: 72-76.

Advances in Consumer Research Volume 13, 1986      Pages 72-76

REPRESENTING ATTITUDE STRUCTURE: ISSUES AND EVIDENCE

Paul W. Miniard, The Ohio State University

Thomas J. Page, Jr., University of Wisconsin-Madison

April Atwood, University of Washington

Randall L. Rose, The Ohio State University

[The authors wish to thank anonymous reviewers for their useful comments and suggestions. Financial support for this research was provided to the senior author by the Marketing Faculty Research Fund, The Ohio State University.]

ABSTRACT -

The representation of attitude structure and associated causal issues has been a topic of considerable interest recently. This research examines whether variations in outcome evaluation produce corresponding variations in attitude, the discriminant validity between the cognitive and affective components of attitude, and the accuracy of a "cognition determines affect" causal ordering.

INTRODUCTION

In the past few years, the consumer research literature has witnessed a renewed interest in the Fishbein-Ajzen behavioral intention model (Ajzen and Fishbein 1980; Fishbein and Ajzen 1975). A primary thrust has been an examination of the model's causal properties (Bagozzi 1982, Burnkrant and Page 1982, Miniard and Page 1983, Ryan 1982, Shimp and Kavas 1984). This focus is quite understandable given the increased interest in structural equations as an analytical tool and the relatively strong theory specification offered by the Fishbein-Ajzen model.

One area of concern in these causal investigations involves the representation of attitude structure. A review of the literature reveals that researchers have rarely employed the same approach for representing the cognitive structure underlying attitude. Rather, several alternative approaches have been proposed which vary along conceptual and/or methodological dimensions. In the following section we discuss these alternative representational approaches and associated causal issues.

BACKGROUND

As a starting point, let us consider the relevant causal assumptions underlying the Fishbein-Ajzen model. It is postulated that the distinction between cognitive (Sb_ie_i) and affective (A_B) components of attitude is a meaningful one. Specifically, beliefs about behavioral outcomes weighted by the evaluation of these outcomes are viewed as the building blocks for overall evaluation of performing a behavior

Evidence concerning this proposition has yielded mixed support. Experimental research by Lutz (1975) revealed that a manipulation of beliefs produced changes in both the cognitive and affective components, but that a manipulation of outcome evaluation altered the cognitive component without influencing the affective component. Furthermore, when Lutz (1977) employed path analysis for these data, Sb_ie_i was a significant predictor of A_B in only the belief manipulation experiment.

Nonexperimental research has also reported challenging results. Burnkrant and Page's (1982) LISREL analysis of a survey focusing on charitable blood donation behavior found a lack of discriminant validity between the cognitive and affective components. That is, the affective and cognitive latent variables were so highly related that the parameter estimate between them did not significantly differ from one! Obviously, this result leads to very different conclusions about attitude relationships than those suggested by Lutz's (1977) findings. To the best of our knowledge, no other study of the Fishbein-Ajzen model has found the cognitive and affective dimensions to be so highly related that they would not achieve discriminant validity.

A further assumption underlying the Fishbein-Ajzen model is that the cognitive component is best viewed as a unidimensional construct in which different behavioral outcomes are aggregated into a single index of cognitive attitude. In accordance with this assumption, research has typically employed a unidimensional representation of cognitive attitude although such research has utilized a variety of approaches in doing so. Some studies (e.g., Miniard and Page 1983, Shimp and Kavas 1984), for example, have used the Eb_ie_i index as the sole indicator of cognitive attitude. Unfortunately, this approach is constrained by its inability to divest the influence of measurement error on parameter estimates.

Other research has avoided this limitation by employing multiple indicators of cognitive attitude. Ryan (1982), for instance, used each behavioral outcome as a separate indicator. This approach, however, may undermine the measurement model fit since each salient behavioral outcome may not yield, by itself, a valid estimate of cognitive attitude. Consider the person who perceives a product as too expensive but is favorably impressed with the product because of its performance on nonprice attributes. In this example, a behavioral outcome measure capturing pricing concerns, unlike nonpricing outcomes, would yield a less accurate estimate of cognitive attitude. Although this salient pricing outcome should not be deleted from the measurement model on theoretical grounds, its retention would have an adverse impact on model fit. Consequently, the requirements for the measurement model may be unfulfilled when salient behavioral outcomes are inconsistent (some are very positive while others are only slightly positive or even negative). However, when salient outcomes are consistent, then the problem essentially disappears.

Alternatively, some researchers have used different attitude scaling techniques as multiple indicators of cognitive attitude. Burnkrant and Page (1982) employed the Sb_ie_i index and Likert scaling as separate indicators, while Bagozzi and Burnkrant (1979) used Guttman, Likert, and Thurstone scaling as alternative assessments of cognitive attitude. Lacking, however, is research which has developed separate estimates of the Sb_ie_i index, an approach that would seem most appropriate for operationalizing Fishbein and Ajzen's conception of cognitive attitude.

RESEARCH OBJECTIVES AND HYPOTHESES

The following study provided additional evidence concerning the structural and discriminant relationships among cognitive and affective attitude constructs as conceptualized and operationalized within Fishbein and Ajzen's theory and model of intentions. [An important characteristic of Fishbein and Ajzen's formulation is that all outcomes, both personal and normative, are reflected by attitude. Other formulations, however, do not share this perspective (e.g., Miniard and Cohen 1983).] We examined the relationship between cognitive and affective attitudes in an attitude formation experiment (Carnegie Mellon Seminar 1978). Consequently, we could control the outcomes which determined subjects' attitude for this novel behavior, a feature very desirable in restricting measures of cognitive attitude to salient beliefs.

Our first hypothesis (H1) is that a manipulation of outcome evaluation will produce corresponding variations in both cognitive and affective attitude. This hypothesis is consistent with Fishbein and Ajzen's conceptions of attitude relationships, although Lutz (1975) did not substantiate this prediction in his experiment. We believe, however, that Lutz's null finding is attributable to methodological limitations which the present study overcomes (see Discussion section).

A second hypothesis (H2) which follows from Fishbein and Ajzen's model is that the affective and cognitive components of attitude are related but distinct constructs. We would expect this distinction to be more meaningful and vivid when the components are assessed in close temporal proximity to their creation. In this study, the components were measured only minutes after their formation.

It is also hypothesized (H3) that a "cognition determines affect" causal flow would occur in our study. As detailed below, subjects were required to process information about the outcomes of a novel behavior. Although there may be situations in which cognition does not precede affect (e.g., Zajonc and Markus 1982), the informational approach used in the present experimental procedures dictates a causal ordering of "cognition determines affect."

METHOD

[The current study employs essentially the same methodology reported in Miniard and Cohen (1983). The reader is referred to their article for additional details about the experimental procedures.]

Subjects

One hundred and twenty one male and female undergraduate marketing students participated in the experiment as partial completion of class requirements. The task was designed specifically for these subjects in order to heighten involvement and relevance. Accordingly, marketing students were placed in a "case analysis" type of setting involving a new product introduction. Subjects were assigned randomly to one of the two cells comprising the single factor design.

Procedure

Subjects were lead to believe that they were participating in a study of factors (e.g., communication patterns, group cohesiveness) that influence group decision making. Subjects were told that they would act as a marketing consulting group for a company that was introducing a new boiling bag food item. The group's task was to select one of two alternative brands (identified as Brands B and F) for market introduction. After reviewing with the experimenter a written copy of the experimental procedures, subjects were placed in separate rooms and read the materials describing the case situation. A critical section of this material was the product trial results consisting of target consumers' brand ratings on product attributes considered by these consumers in making their purchase decision. Brand B was rated as excellent in taste, good in nutrition and convenience, and average in price. Brand F was rated less favorably on taste (good) and nutrition (average), but more favorably on price (good) and convenience (excellent). These brand ratings were intended to make the two alternatives fairly equivalent in overall attractiveness. As a result, the influence of the manipulation (explained below) could more easily manifest itself. A configuration of attribute ratings which clearly favored a particular brand would have greatly lowered the manipulation's potential to impact on subjects' choice behavior.

After they had reviewed the case materials and attribute ratings, subjects received via the experimenter a written message from the group leader advocating brand F. This was followed by messages delivered sequentially from the remaining group members that supported the leader's choice. All messages were in fact fictitious and predetermined by the experimenter. Following these messages, subjects transmitted their communication, responded to a questionnaire, and indicated their choice on a brand recommendation form.

The experimental manipulation was operationalized in the form of a "group member award." This award was justified to subjects on the grounds that, due to the temporary nature of the group, it was unlikely that cohesiveness (claimed to be a group characteristic of interest) would be as prevalent in the present experimental group as it would be in more "natural" groups. Consequently, the award was to be given by the group through a voting system to the member demonstrating the highest level of "team spirit" in order to compensate for this study limitation. Subjects understood that this vote should be based "solely upon a group member's willingness to work in a harmonious fashion with the group and upon a member's loyalty to the group" and that the group could not vote for someone "unless he or she acts in a cohesive manner which, in this situation. is defined as recommending the same brand as the leader.

In one condition, the winner of the award was simply to be known as the best or most loyal group member. In the other condition, the award was to be accompanied by a 520 cash prize. Thus, the manipulation was intended to vary the importance subjects attached to one of the outcomes (i.e., winning the award) associated with their choice behavior. This manipulation of outcome evaluation, if successful, permits a further test of Lutz's (1975) challenging result where variations in outcome evaluation did not produce corresponding changes in affective attitude.

Questionnaire

The questionnaire began with one and a half pages of instructions on the appropriate usage of the 7-point bipolar response scales employed for measuring the various constructs. Although the questionnaire contained measures for all of the constructs within Fishbein and Ajzen's intention model, only measures relevant to the focus of this paper are discussed here. [The ultimate objective of this research is to test the entire causal motel proposed by Fishbein and Ajzen. In this first cut at the data, however, we restrict our focus to cognitive and affective attitude relationships.] The affective attitude toward recommending each brand (A_B) was measured by five evaluative semantic differential scales (good-bad, wise-foolish, beneficial-harmful, satisfying-dissatisfying, rewarding-punishing).

Cognitive attitude was assessed in accordance with the procedures for operationalizing the Sb_ie_i index. Beliefs (b_i) that recommending the brand would lead to a particular outcome were assessed by three probabilistic scales (likely-unlikely, possible-impossible, probable-improbable). Outcome evaluation (e_i) was measured by three evaluative scales (good-bad, beneficial-harmful, rewarding punishing). Beliefs and outcome evaluations are typically assessed with one. not three, measurement scales. The use of three scales, however, allowed us to construct three separate Sb_ie_i. indices by arbitrarily pairing belief and evaluation scales. [One might be concerned that the causal results may vary across different measurement pairings. However, a comparison of results using different pairings revealed virtually no difference.]

In order to identify the salient behavioral outcomes underlying subjects' choice behavior, a pretest was undertaken in which approximately 25 subjects participated in the experiment as described above, with one exception. Rather than completing a questionnaire concerning beliefs, attitudes, and intentions prior to their choice behavior, pretest subjects responded to a questionnaire consisting of several open-ended questions following their brand choice. These questions asked subjects to identify those considerations important to them in making their brand choice (e.g., the advantages/disadvantages of choosing brand B/F).

This pretest revealed that subjects' decision making largely focused around picking the best brand within the case context (a personal outcome). Winning the award (a normative outcome since it is under the control of others) was also considered by subjects, particularly by those in the $20 condition. Further evidence substantiating the salience of this outcome is presented in the Results section. Some subjects expressed concerns over attributions the group might make about them (e.g., "I chose F because I didn't want the group to think I was being defiant"), another normative outcome. These various concerns were represented in the form of three consequences which were used in the belief and outcome evaluation measures: meeting the needs of the target market, winning the group member award, and having the group act and/or think favorably about the subject.

Causal Model Analysis

Because measurements were collected for each brand individually,one analytical approach involved testing a causal model for each separate brand. However, Ajzen and Fishbein (1969) have demonstrated greater predictive accuracy when attitudes toward all behavioral alternatives are considered simultaneously (i.e., when difference scores are employed) than when the attitude toward only one of the possible alternatives is considered. Consequently, difference scores between the cognitive and affective attitudes for brands B and F were also used for deriving the correlation matrix employed in the LISREL VI analyses.

Figure 1 contains the basic causal model tested in this research. The experimental manipulation (Reward), treated as a dichotomous variable, should have a causal impact on cognitive attitude (assessed by three Sb_ie_i indices). Cognitive attitude, in turn, should both determine affective attitude (assessed by five evaluative scales) as well as completely mediate the influence of the manipulation on affective attitude (i.e., Reward should not have a direct effect on A_B).

FIGURE 1

STANDARD CAUSAL MODEL

RESULTS

Impact of the Manipulation

It would of course be inappropriate to test the first hypothesis if the manipulation failed to influence outcome evaluation. Consequently, it was necessary to establish that the manipulation of the group member award's monetary value was successful in altering subjects' evaluation of the "winning the award" outcome. As expected, subjects' responses to the three evaluation scales (e_i) pertaining to this outcome were significantly (p < .001) more favorable when the award carried a $20 cash prize than when it lacked a monetary value.

It is also useful to consider the manipulation's impact on subjects' choice behavior. In the condition where the award did not carry a financial incentive, nearly half (21 out of 47) of the subjects chose the same brand (F) as the group. In the condition where the award was accompanied by the $20 cash prize, two-thirds (49 out of 74) of the subjects selected brand F. This variation in choice behavior was significant (p < .02). In addition to substantiating the manipulation's effectiveness, these findings further support that winning the award was a salient consideration in subjects' decision making.

Causal Model Findings

Findings concerning various goodness-of-fit indices, the measurement model, and the structural model parameters are presented in Tables 1-3 respectively. Note that these tables contain the results based on difference scores as well as for the individual brand responses. As can be seen, the results are highly similar for the difference and individual brand models. The fact that the path estimate between Reward and Sb_ie_i is positive for the brand F model but negative in the other two models simply reflects the impact of how the dummy manipulation variable was coded (no $20 = 1, $20 = 2) as well as the manner in which difference scores were estimated (brand B response minus brand F response).

TABLE 1

CASUAL MODEL RESULTS FOR GOODNESS-OF-FIT INDICES

TABLE 2

MEASUREMENT MODEL RESULTS

TABLE 3

STANDARDIZED STRUCTURAL MODEL PARAKETERS

As was the case for all the causal models examined here, the chi-square test was highly significant (p < .001). However, there seems to be a growing consensus that this indicator of model fit is too conservative. Indeed, the results for alternative fit indices such as the root mean square residual (RMR) and Bentler and Bonett's (1980) rho statistic (r) suggest that the causal model depicted in Figure 1 provides a very good representation of data relationships.

The measurement model results are also quite strong. All factor loadings except one (Att 5 for brand B) are estimated to be greater than .80 and most exceed .90. In addition, the measured variables are excellent indicators of the latent factors as evidenced by the strong reliabilities reported in Table 2. However, several of the affective attitude measures shared considerable measurement error. Indeed, simply freeing three of the parameter estimates representing correlated measurement errors resulted in a model with a more "acceptable" chi-square. The results for the difference model, for example, were: x²(23) = 34.01, p = .065; RMR = .027; p = .988. Finally, as shown in Table 3, both causal paths manipulation to cognitive attitude and cognitive to affective attitude) are significant (p < .05) for all models.

Research Hypotheses

The preceding analyses have yielded several findings supportive of H1. It was shown that the evaluation of the "winning the award" outcome was affected by the manipulation. Accordingly, the manipulation had a significant impact on cognitive attitude, as reflected by the significant path estimate between Reward and Sb_ie_i in the prior causal model analysis. Similarly, the path estimate between cognitive and affective attitude was also significant.

These latter two results, while suggestive, do not directly show that the manipulation of outcome evaluation did ln fact influence affective attitude. Consequently, a direct test of this issue is necessary, particularly since it bears directly on Lutz's (1975) finding that such a manipulation did not influence affective attitude. An analysis of the difference model excluding cognitive attitude (i.e., on}y Reward and affective attitude were included) revealed a significant (p < .05) path between the manipulation and affective attitude (the same result also occurs in the individual brand models), the final piece of evidence necessary for supporting H1. Interestingly, this path estimate (g = -.169) was smaller than the estimate observed between the manipulation and cognitive attitude (g = -.310). These results suggest that the manipulation had a stronger impact on cognitive attitude and are consistent with the hypothesized causal flow of cognition determining affect (H3).

Examination of the discriminant validity between cognitive and affective attitude (E2) involved two analytical procedures. The first procedure involved testing whether the path estimate between cognitive and affective attitude differed from one. Although this path estimate in the difference model was substantial (b=.770, s.e. = .067), it did significantly (p < .05) differ from one. Similarly, the path estimates in the individual brand models also differed (p < .05) from one.

The second approach employed confirmatory factor analyses. A single factor model in which cognitive and affective indicators were loaded on one factor was compared to a two factor model consisting of separate but related cognitive and affective components. A comparison of the chi-square values for these two models revealed that the two factor model provided a significant (p < .01) improvement in representing data relationships for both the difference and individual brand models. The findings from both approaches therefore support the second hypothesis that cognitive and affective attitudes, while strongly related, are distinct components.

Evidence relevant to H3 was attained by analyses which reversed the ordering of cognitive and affective attitudes in the causal model. The results of these reverse order models are summarized in the bottom half of Table 1. Comparisons of the goodness-of-fit indices for models differing in their causal ordering of the attitude components reveals that the reverse order model results are only slightly worse than those found for the hypothesized order model.

In contrast, consideration of cognitive attitude's ability to mediate the influence of the manipulation on affective attitude (and vice versa) proved quite informative. A cognition before affect ordering suggests that the manipulation should influence affect only through cognition and should not have a direct effect on affect. Similarly, the affect before cognition ordering implies that the manipulation should influence cognition only through affect. Consistent with the hypothesized causal flow, the manipulation did not have a direct effect (p > .05) on affective attitude for either the difference or individual brand models. For the reverse order models, however, the manipulation did have a significant (p c .05) influence on cognitive attitude which was independent of affective attitude. We interpret these results as supporting H3.

DISCUSSION

This study provides additional support for the conceptualization offered by Fishbein and Ajzen concerning attitude structure. First, the results show that a manipulation of a salient behavioral outcome had a significant impact on both cognitive and affective attitude. This inconsistency with Lutz's (1975) findings (where the influence of the experimental manipulation on cognitive structure did not "carry over" to affective attitude) may be due to methodological differences. Lutz's study attempted to influence subjects' attitude toward an unknown brand of laundry detergent by varying information relevant to their evaluation of a detergent's sudsiness, a relatively minor attribute.

In contrast, the current study employed a powerful financial manipulation for altering outcome evaluation as well as a highly involving choice situation. The likelihood that affective attitude would also reflect influences impacting on cognitive structure would seem to be greater in the present study given subjects immediate need to make a choice between two alternatives which differ considerably in their outcomes (i.e., only one leads to the reward). Finally, Lutz's study did not enjoy the advantage of multiple measurements and structural equations analysis, and thus one can not eliminate the possibility of "biased" results due to measurement error.

Fishbein and Ajzen's position that cognitive and affective attitudes are distinct but highly related in that cognition is the basis for affect development is also supported by this research. Of course, other settings may generate different causal relationships than observed here. The informational approach employed in the experiment for affect formation and the assessment of the attitudinal constructs in close temporal proximity to their development would seem to be major determinants of the present findings.

REFERENCES

Ajzen, Icek and Martin Fishbein (1969), "The Prediction of Behavioral Intentions in a Choice Situation," Journal of Experimental Social Psychology, 5, 400-416.

Ajzen, Icek and Marting Fishbein (1980), Understanding Attitudes and Predicting Social Behavior, Englewood Cliffs, NJ: Prentice-Hall.

Bagozzi, Richard P. (1982), "A Field Investigation of Causal Relations Among Cognitions, Affect, Intentions, and Behavior," Journal of Marketing Research, 19 (November), 562-584.

Bagozzi, Richard P. and Robert E. Burnkrant (1979), "Attitude Organization and the Attitude-Behavior Relationship," Journal of Personality and Social Psychology, 37, 913-929.

Bentler, Peter M. and Douglas G. Bonett (1980), "Significance Tests and Goodness-of-Fit in the Analysis of Covariance," Psychological Bulletin, 88, 588-606.

Burnkrant, Robert E. and Thomas J. Page, Jr. (1982), "An Examination of the Convergent, Discriminant, and Predictive Validity of Fishbein's Behavioral Intention Model," Journal of Marketing Research, 19 (November), 550-561.

Carnegie-Mellon University Marketing Seminar (1978), "Attitude Change or Attitude Formation? An Unanswered Question," Journal of Consumer Research, 4 (March), 271-276.

Fishbein, Martin and Icek Ajzen (1975), Beliefs, Attitude, Intention, and Behavior, Reading, MA: Addison-Wesley.

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

Lutz, Richard J. (1977), "An Experimental Investigation of Causal Relations Among Cognitions, Affect, and Behavioral Intention," Journal of Consumer Research, 3 (March), 197-208.

Miniard, Paul W. and Joel B. Cohen (1983), "Modeling Personal and Normative Influences on Behavior," Journal of Consumer Research, 10 (September), 169-180.

Miniard, Paul W. and Thomas J. Page, Jr. (1984), "Causal Relationships in the Fishbein Behavioral Intention Motel," in Advances in Consumer Research, Vol. 11, ed. Thomas C. Kinnear, Ann Arbor, MI: Association for Consumer Research, 137-142.

Ryan, Michael J. (1982), "Behavioral Intention Formation: A Structural Equation Analysis of Attitudinal and Social Influence Interdependency," Journal of Consumer Research, 9 (December). 263-278.

Shimp, Terence A. and Alican Kavas (1984), "The Theory of Reasoned Action Applied to Coupon Usage," Journal of Consumer Research, 11 (December), 795-809.

Zajonc, Robert B. and Hazel Markus (1982), "Affective and Cognitive Factors in Preferences," Journal of Consumer Research, 9 (September), 123-131.

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