A Discussion of Attitude Research and Behavioral Intentions

Paul R. Warshaw, McGill University
ABSTRACT - Each of the three papers in this session focus upon issues related to the determinants of behavioral intentions, Holbrook et al, focus upon attitude formation. Specifically, the role of attribute importance in attitude structure is addressed. Brinberg looks more at the dimensionality of attitudes and norms. Employing Fishbein's and Triandis' models he shows both attitude and normative constructs may be bi-dimensional. Miniard, on the other hand, questions whether measured attitudes and norm are really independent factors of intent.
[ to cite ]:
Paul R. Warshaw (1981) ,"A Discussion of Attitude Research and Behavioral Intentions", in NA - Advances in Consumer Research Volume 08, eds. Kent B. Monroe, Ann Abor, MI : Association for Consumer Research, Pages: 53-56.

Advances in Consumer Research Volume 8, 1981      Pages 53-56

A DISCUSSION OF ATTITUDE RESEARCH AND BEHAVIORAL INTENTIONS

Paul R. Warshaw, McGill University

ABSTRACT -

Each of the three papers in this session focus upon issues related to the determinants of behavioral intentions, Holbrook et al, focus upon attitude formation. Specifically, the role of attribute importance in attitude structure is addressed. Brinberg looks more at the dimensionality of attitudes and norms. Employing Fishbein's and Triandis' models he shows both attitude and normative constructs may be bi-dimensional. Miniard, on the other hand, questions whether measured attitudes and norm are really independent factors of intent.

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

The Holbrook, Valet and Tabouret paper examines how attribute importance is related to attitude structure. While the study essentially replicates (with some methodological improvements) an earlier piece by Holbrook and Maier (1978), the findings are quite interesting and highlight some theoretical and methodological issues that consumer researchers should attend closely. The primary thesis of Holbrook et. al, is that product attributes very in their importance to consumers. That is, some attributes are perceptually more important than others in determining product-related attitudes. The principal is certainly not novel, since attitude researchers have long acknowledged this fact. The concept is an outgrowth of Fishbein's proposition that attitudes are formulated from a small set of salient beliefs. However, the Fishbein approach assumes that once an attribute is incorporated within the salient set, its only contribution to attitude comes from the subjective probability of it being associated with the target object or action and its goodness or badness. Other researchers, quite naturally, became intrigued with the notion that there must be some hierarchical importance structure among those attributes in the evoked set. After all, since the evoked set contains approximately 5-9 attributes, the remaining attributes are tautologically non-important. If so, doesn't it logically follow that the salient attributes are similarly disectable into most important and least important classes? In concert with this line of reasoning, attempts have been made to secure attribute importance ratings from subjects and to use these ratings as weights in expectancy value or other type attitude models. The problem is that importance weight inclusion does not seem to enhance predictions of affect. The Holbrook et. al, paper proposes and tests a conceptual framework for resolving this seemingly perplexing issue. They claim that attribute importance has a major impact, but not within the model structure per se. Rather, importance relates more directly to information search processes. That is, the more important an attribute, thc more information is gathered about it, with attitudes formed thereafter on the basis of cues acquired about these important attributes. The study was fairly well done and seemingly verifies the authors' model. More significant, it implies an issue that consumer researchers have too often ignored. A common practice in marketing and related disciplines is for researchers to haphazardly decide that some variable (e.g., a personality trait like self-confidence) is related to some aspect of consumer behavior (e.g., purchase of name-brand goods). Typically, there is no systematic, careful development of theoretical linkages between the variables studied (cf. Kassarjian 1971). Rather, in knee jerk fashion, the variable is tossed into some equation modal in an additive or multiplicative fashion (without theoretical rationale} and run through a regression to see if significant betas pop out. Alternatively, some type of discriminant, factor or canonical analysis is conducted to delineate 'underlying factors or relationships'. The results are typically meaningless and not at all generalizable. This scenario is typical of marketing studies pertaining to attitude theory issues. The so-called importance weight debate is no exception. Hence, in my opinion, the greatest contribution of the Holbrook et. al, manuscript is its orientation toward building theoretical linkages in systematic, logical fashion. This tight approach is certainly more difficult and time consuming, and requires broader literature reviews than the alternative approach which is based mainly on personal opinion, scant surveys of existent research and frantic applications of the ' latest', 'hottest' data analysis techniques.

The paper has some other features I'd also like to highlight:

1. An interesting methodological technique is presented that might usefully be employed by other researchers. Namely, use of a sticker-pulling procedure in a mail-survey questionnaire was attempted (apparently quite successfully). This raises the possibility of conducting studies by mail that heretofore required laboratory settings. A major feature of social science research is that most studies employ relatively small, student samples. This creates many problem (e.g., nongeneralizable results, sensitized subjects, lack of statistical validity, etc.). Everyone would like to use large, representative samples but the dollar costs are typically prohibitive (especially where laboratory type formats seem warranted). Experimentation by mail certainly allows for better and larger sampling. Thc current sticker-pulling technique and further developments along theme lines could help bring the laboratory to the people in an inexpensive, simple manner.

2. Part strength and part weakness of Holbrook et. al, was that the authors were statedly aware of certain key assumptions inherent in their methodology, For example, "it is assumed that the respondent believes the information he has received to be true...". It would have been better methodology to verify this assumption during debriefing. Many consumer researchers are unaware that this and other key assumptions are integral components of their research design. This is particularly true when treatment manipulations entail differential communication across subject groups. Without appropriate manipulation check analysis (which includes believability questions), one may not justifiably conclude that the findings resulted from the stated treatments. Rather, all sorts of demand characteristic and other explanations are equally possible, thereby threatening the experiment's internal validity. While Holbrook et. al, recognized the problem, they did not attempt to ameliorate it. Similarly, the authors correctly acknowledge that they are assuming subjects assigned a subjective probability of 1.0 to each cue, and that inferential beliefs may have been generated but were unassessed by the research instrument. It is meritorious that Holbrook et. al, were aware of these assumptions, but it would have been much better to have dealt directly with these issues or, at the least, to explore them via manipulation checks.

A COMPARISON OF TWO BEHAVIORAL INTENTION MODELS

Brinberg's paper compares the Fishbein and Triandis intention models. Results purportedly support Triandis' notion that effective and evaluative components of attitude should be measured separately. Further, contradicting both models, Brinberg finds two dimensions of social influence (whereas both Fishbein and Triandis postulate alternate unidimensional constructs). However, when Brinberg attempts To predict intent from the revealed attitudinal (two) and social influence (two) dimensions, only small predictive improvements are realized.

This is certainly an interesting paper for consumer researchers, but not because the study itself evidences exceptional methodology or creative insight. Rather, its greatest contribution comes from the fact that it brings the Triandis model to the attention of the ACR audience. One major blind spot among marketing researchers concerns their acknowledgement and study of existing non-Fishbeinean type intention models. Unfortunately, most attitude researchers in management departments are totally unaware of the existence of these competing paradigms. Typically at best, only the Fishbein model is cited, and then frequently in erroneous fashion. Hence, Brinberg's conference paper performs the valuable function of publicizing one of the important alternate conceptualizations of intent. Let me take this opportunity to point out other models that consumer researchers should attend: Wicker's paradigm incorporating judged influence of extraneous events (e.g., Wicker 1971); Schwartz's model, which discusses moral norms and ascription of responsibility, and Snyder's use of self-monitoring as a moderator variable (cf. Zuckerman and Reis 1978); Sheth's model (Ragu, Bhagat and Sheth 1975); and Warshaw's behavioral intention paradigm (1980). While some of these approaches are on stronger conceptual footing than are others, each adds something to our understanding of attitudes and intentions. Hence, consumer research in this area would be enriched considerably by conscientious attention to all competing theoretical tracks.

Concerning Brinberg's research per se, a few limitations seem evident:

1. Dissecting and separately measuring both effective and evaluative dimensions of attitude may be necessary only for a restricted class of behaviors. As defined by Triandis and herein applied, affect concerns strong emotional states associated with some criterion act (e.g., disgust, nausea}. While these feelings may be germane and present within certain behavioral concerts (e.g., blood donation, hospital volunteer work, purchase of pornographic material), such strong affective responses are unlikely to be associated with most product-related settings. Hence, the author's findings may have limited applicability for the bulk of consumer research. Nevertheless, this is primarily an intuitive hypothesis on my part and is not intended to deprecate Brinberg's results. Certainly, further work is merited on the affect versus evaluation issue.

2. I have strong reservations concerning some of Brinberg's questionnaire items. For example, Fishbein's Aact was measured by the following item: "Donating blood at the blood bank this semester is good-bad, etc.". However, the criterion act concerns subjects' own intentions to donate blood.' Clearly, a person might claim that 'donating blood' is good but ascribe non-intent to personally donate blood. The problem here is that the target of Aact and BI differ. While Aact concerns the general concept of blood donation, BI focuses upon the target person per se. Ajzen and Fishbein (1977) clearly point out the necessity of employing the same target, object, time and context for both Aact and BI.

Similarly, Brinberg employs an imprecise intent measure. That is, he assesses intent by the following item: "I intend to donate blood at the blood bank this semester: likely-unlikely". This measures the likelihood that the subject will intend to donate blood, not the likelihood that he/she will donate blood. Clearly, these are not equivalent, and accurately measuring subjective probability of behavior requires the latter 'will donate' format. To clarify this distinction, think of a person who is self-professedly likely to intend to stop smoking, but admits that the probability of his stopping smoking is meager. Part of the problem here is that Triandis' definition of intent is not directly comparable to Fishbein's parallel measure. Brinberg uses the Triandis measure but then relates both Triandis' and Fishbein's independent variables to this singular construct. Fishbein's subjective probability of behavioral performance is more akin to the objective, derived probability of behavior (Pa) variable in Triandis model (where Pa equals the sum of habits (i.e., the number of times the behavior has been performed in the past) and intent multiplied by a facilitating conditions measure). Hence, to be fair, Brinberg should have employed two alternate items to assess 'intent'.

Finally, I an unconvinced that appropriate scales were used to measure affect (gut feeling}. The adjective sets 'enjoyable-disgusting' and 'exciting-nauseating' do not constitute bipolar adjective pairs. 'Disgusting' is not the opposite of 'enjoyable' (in fact, some acts may be enjoyable because they are disgusting). Similarly, exciting and nauseating may be quite unrelated. A better pair, for example, would be exciting-dull. Hence, interpretation of revealed factors is subject to debate.

Hence, the present study has sufficient instrumentation problems to render suspect the thrust of its conclusions. Nevertheless, the issues raised are important and justify additional research employing multiple types of behavior.

EXAMINING THE DIAGNOSTIC UTILITY OF THE FISHBEIN BEHAVIORAL INTENTION MODEL

Miniard's piece contains a rather lengthy review of evidence concerning the validity of Fishbein's behavioral intention model. While far from exhaustive, this literature review is certainly the strongest aspect of the paper. Miniard's experiments, on the other hand, are very troublesome. Their stated purpose is to ascertain whether the social norm (SM) measure receives a significant regression weight when social influences 'do affect' intentions but is nonsignificant when social influences don't impact upon intent. However, methodological deficiencies, questionable logic and inattention to recent published research negate the usefulness of these studies.

Regarding Study 1, the following problems seem evident:

1. In the 'others expectations unknown' treatment, Miniard claims the situation is devoid of information concerning the social environment. Without sufficient manipulation checks (which apparently weren't done), this is an unwarranted claim. The role playing scenario format makes this a projective technique. As in all such artificial, 'pretend' tasks, a wide range of inferences, projections from past experiences and other situation-structuring cognitions are triggered. Hence, it is unreasonable to assume that subjects perceived a blank, featureless social environmental context. If one wants to make this rather counterintuitive claim, manipulation check verification is imperative.

2. Similarly, there were no manipulation checks for the "others expectations 'positive' and 'negative'" treatments. Given the nature of the research hypothesis, it is necessary to verify whether subjects believed and accepted these statements as factual truths. If subjects did, in fact, believe these statements, then the subsequent disclaimer, "of course, what others feel you personally should do here is of no concern to you" is highly unlikely to negate fully the normative manipulation. If so, this is not a 'no normative influence' experiment (as Miniard claims) and SN should relate to BI. Since this did not occur, I suspect that subjects were not properly manipulated into perceiving differential 'others expectations'. If so, the treatments are invalid. Nevertheless, this still may serve as a no-influence experiment. Lack of a significant beta weight on SN and n.s. cor (SN,BI) reinforce this conclusion. However, the logic behind the experimental format seems highly questionable. Why bother to manipulate normative influences here? If the sole purpose was to demonstrate that SN is nonsignificant in contexts devoid of social influence, a simple, more tightly controlled design would surely have sufficed.

Essentially, my primary criticism here is that the author makes a lot of claims about what people were thinking and how they approached the task without any sort of verification via debriefing procedures. Without this structure, numerous explanations for the findings are possible.

Study II constitutes a reanalysis of previously published data. Miniard claims that the experimental format implies the existence of social influence. Hence, SN should correlate with BI and carry a significant regression weight. Once again, there are methodological problems and strained logic:

1. Since the role-playing scenarios depicted some hypothetical female buying a dress, why were both male and female subjects employed? This task is so external to the experience of most males that their responses would clearly represent projections of stereotypic 'female' behavior rather than their personal cognitive processes. Females, on the other hand, are likely quite familiar and experienced with the scenario settings. This familiarity probably generated projections from similar experiences in memory. Hence, their responses might be based on factors external to the treatment manipulations. Given that the task was so different for males than females, either only female subjects should have been used (with excessive manipulation checks), or else data should be analyzed separately for roles and females.

2. Miniard only manipulated situation-specific information concerning the neighbor. However, NB and MC were additionally measured for the hypothetical woman's husband and parents. This creates several problems. First, SN is supposed to concern subject's perceptions of his/her 'most important others'. Miniard is assuming that neighbor, husband and parents are the 'important set' for each subject. This claim is only justified if verified by manipulation checks, since there is good cause to believe that some or all of these 'referents' may be irrelevant for dress buying. Further, what about the woman's 'best friends'? They might be expected to be highly important others.

Further, Miniard had 57 additional subjects read the scenarios and estimate the impact of the social influences in the scenario on their BIs. Finding differences between NMI and MI, Miniard claims social influence was important in the scenario settings. However, social .influence in the trenchant paragraphs involved only the neighbor, while the measuring instrument assessed attributions concerning husband, parents and most important others as well as the neighbor. To say the neighbor exerted social influence is not equivalent to saying that a measure which incorporates the neighbor and several other referents is related significantly to BI. Hence, there is no compelling justification for Miniard's hypothesis that SN should receive a significant beta weight. If SN only concerned the neighbor, then this postulate might be warranted. But, SN includes attributions to many others besides the neighbor, and in a projection task, these other attributions are unlikely to remain constant across treatments. Miniard, unjustifiably, implicitly assumes this constancy condition.

Hence, Miniard's claim that versions 1 and 2 of the FBI model fail to correctly identify the importance of normative influences is highly suspect. Since SN includes important others, the neighbor my or may not be included in this set. Further, since no information concerning husband, parents and other friends was given, SS is probably largely a projection of one's own AB, which would imply high cor(AB, SN), where AB dominates and thus receives the significant beta weight. The nonsignificance of model 2 simply shows that general MC is unrelated to influence in this particular situation (which is not surprising given lack of general information about any member of the offered referent set). The significant results in models 3 and 4 support this analysis (in each case there was significant cor(SN,Bl) and SN had a significant beta). That is, treatment information was situation-specific, and when MC was similarly specific, having only data on the neighbor gave this referent a dominant influence on SN, which then behaved in line with expectations. The 57 subject influence check implied that this would occur, but did not imply that when general SN or MC were employed, significance would result. Only when measures were specific enough to capture the neighbor's stated influence in these specific treatment situations was SN likely to be significant - and it was.

The only point Miniard really makes is that SN and AB are typically correlated, which questions the assumption that AB and SN are independent components of intent. However, this issue bas been covered extensively in Warshaw (1980) and elsewhere. The real issues, such as separating out identification, internalization and compliance aspects of normative influence were not even touched upon by Miniard.

REFERENCES

Ajzen, I. and Fishbein, M. (1977), "Attitude-Behavior Relations: A Theoretical Analysis and Review of Empirical Research", Psychological Bulletin, 84, 888-919,

Holbrook, M. and Maier, K. (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.

Kassarjian, H. (1971), "Personality and Consumer Behavior: A Review", Journal of Marketing Research, 8, 409-14.

Raju, P., Bhagat, R. and Sheth, J. (1975), "Predictive Validation and Cross-Validation of the Fishbein, Rosenberg and Sheth Models of Attitudes", in M. Schlinger (ed.), Advances in Consumer Research, Vol. II, Chicago. Association for Consumer Research.

Warshaw, P. (1980), "A New Model for Predicting Behavioral Intentions: An Alternative to Fishbein", Journal of Marketing Research, 17, 153-72.

Wicker, A. (1971), "An Examination of the 'Other Variables' Explanation of Attitude-Behavior Inconsistency", Journal of Personality and Social Psychology, 19, 18-30.

Zuckerman, M. and Reis, H. (1978), "Comparison of Three Models for Predicting Altruistic Behavior", Journal of Personality and Social Psychology, 36, 498-510.

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