Comments on &Quot;Intentions and Behaviors&Quot;

John G. Lynch, Jr., University of Florida
[ to cite ]:
John G. Lynch, Jr. (1984) ,"Comments on &Quot;Intentions and Behaviors&Quot;", in NA - Advances in Consumer Research Volume 11, eds. Thomas C. Kinnear, Provo, UT : Association for Consumer Research, Pages: 156-158.

Advances in Consumer Research Volume 11, 1984      Pages 156-158

COMMENTS ON "INTENTIONS AND BEHAVIORS"

John G. Lynch, Jr., University of Florida

If the three papers here share a common theme, it pertains to choice from among multiple alternatives. Louviere's paper comes out of a research tradition in conjoint analysis in which behavior exhibited with respect to a multiattribute alternative (i.e , choice) has long been construed as a reflection of the alternative's attractiveness relative to other options in the choice set. Barbeau and Qualls introduce the idea into the consumer satisfaction/dissatisfaction area that reactions to a consumption experience can only be fully understood by considering perceptions of the relative attractiveness of competing products that might be consumed. Miniard and Page's paper considers the implications of alternative causal configurations of components within Fishbein and Ajzen's (1975) model of behavioral intentions in the context of multi-alternative choice. I will argue that the structural results uncovered may depend upon the degree to which the analysis recognizes the interrelationships among intentions to choose one alternative and the attractiveness of competing alternatives (Jaccard 1981)

Miniard and Page

All of ny comments on Miniard and Page's paper pertain to alternative data analyses they might consider that could prove enlightening. First, in the Fishbein model rejected by Miniard and Page (see their Figure 1), Ebe is asserted to cause A_, and ,NB MC was asserted to cause SN. However, in some of Fishbein's writings (e.g., Fishbein and Ajzen 1975), he states that Eb e and AB (measured by Miniard and Page as the sum of scores on four evaluative semantic differential scales) are alternative measures of the same construct. One would assume that ENB MCj and the single item measure of SN used by Miniard and Page might also be construed as alternative measures of a common construct. If so, it would be interesting to evaluate the "alternative Fishbein model" shown below.

FIGURE 1

ALTERNATIVE CONFIGURATION OF FISHBEIN MODEL

A second analysis that might be revealing accepts the basic causal flows implied by the representation of the Fishbein model shown in Miniard and Page's Figure 1. As I understand the experiment that was performed, subjects were given a choice of recommending one of seven brands. In practice, all subjects chose either brand A or brand F. In comparing the fit of the causal configuration of model components proposed by Fishbein to that of their alternative configuration, Miniard and Page performed separate analyses for brand A and for brand F. That is, intentions to choose brand A and choice of brand A were related to beliefs, attitudes, subjective norms, etc. pertaining to brand A only, while intentions to choose brand F and choice of F were related to measures of model components for brand F alone. However, it is clear that the dependent variable of choice of brand A is simply the mirror image of the measure of choice of brand F. If attitudes, subjective norms, etc. pertaining to brand A predict choice of brand A, they are equally predictive of choice of brand F. Thus, when the data for choice A and F are analyzed separately, we have the potential for model misspecification due to omitted variables. An alternative specification is shown below.

FIGURE 2

CHOICE BETWEEN ALTERNATIVES A AND F IN A MODIFIED FISHBEIN MODEL

If any of the measure of antecedent constructs pertaining to brand A (i.e., sum biaaia, ABa, SNBia, and SNa ) covary with any of the corresponding terms pertaining to brand F, analyses that only consider measures of model components for one brand may produce misleading results.

Finally, while I am in sympathy with Miniard and Page's claim that the effects of Normative Beliefs on Subjective Norm may not be moderated by Motivation to Comply, the finding that SNBj predicts SN better than SNBjMCj is not necessarily evidence for that claim. In any c A relational test of a multiplying model, the correlation between a product term (NBjMCj) and a criterion (SN) will change with linear rescaling of the predictors (Cohen 1978). Thus, weighting NB by MC may decrease prediction, not because the multiplicative relationship is inappropriate, but because the two components are not measured on ratio scales. Holbrook's (1977) optimal scaling procedure would be an appropriate solution to the problem in the present case.

None of this is meant to imply a belief that the substantive conclusions of Miniard and Page are off target. Rather, my comments are made in the spirit of Ray and Heeler's (1975) call for comparison of conclusions suggested by multiple analysis methods applied to a given data set.

Barbeau and Qualls

I applaud Barbeau and Qualls effort to incorporate Thibaut and Kelley's (1959) idea of Comparison Level for Alternatives (CLA T) into the consumer satisfaction/ dissatisfaction area, and to relate satisfaction to the consumer's overt behavior -- in particular, subsequent choice and search behavior. While Comparison Level (CL) concepts may suffice to explain satisfaction with a chosen brand, it does not necessarily follow that one who is dissatisfied will be likely to switch brands. This, of course depends upon whether the consumer is aware of more attractive alternatives. Thus, setting aside the attribute specific aspects of Barbeau and Qualls model:

Satisfaction with

Present Brand = f (Present Outcomes - CL)

Probability of

Retaining Present Brand = g (Present Outcomes - CLALT);

Where f and g are monotonically increasing functions. One might postulate that in a case in which Present Outcomes exceed ClA I but fall below CL, the dissatisfied consumer would tend to avoid switching, but might engage in low-level search that increases his or her receptivity to marketer and nonmarketer communications about new alternatives. It is difficult to say what might happen in a case in which the consumer's Present Outcomes exceed the CL, but fall below those available by choosing some other option (CLALT1. On one hand, one might expect this to engender some dissatisfaction as the consumer is aware that his or her choice is somehow suboptimal. On the other hand, one might speculate that the consumer might be satisfied and might not be very motivated to seek out the superior alternative, but willing to try it if it is convenient to do so.

While, in general, I agree with many aspects of the Barbeau and Qualls paper, I do have two criticisms: one of their paper in particular, and one of Adaptation Level Comparison Level approaches to consumer satisfaction in general. First, Barbeau and Qualls' ideas are sometimes obscured by a proliferation of concepts in their model. For example, while the origins of Adaptation Level and Comparison Level concepts differ, its not clear that AL and CT. are fundamentally different concepts as applied to this domain. Also, it's unclear how the overall concepts of idiosyncratic factors, C. and contextual factors, S relate to their attribute specific counterparts that, in turn. determine overall Discrepancy, D.

My most fundamental criticism, though, is of the use of Adaptation Level/Comparison Level accounts of contextual effects on perception. First, evidence from perception (See Birnbaum 1974 for a review) allows one to reject conclusively Adaptation Level theory in the form proposed by Helson (1964). Moreover, the evidence seems to suggest that contrary to CL and AL sorts of ideas and to intuition, contextual factors rarely change one's fundamental perception of an object. Rather, they change how subjects use researcher-provided rating scales to reflect thoughts that are largely context-invariant. Chakravarti and Lynch (1983) reviewed this literature and concluded that context effects on perceptions are very rare, and occur primarily when subjects have very little prior experience with the content domain being rated. The practical implication of this is that while ratings of satisfaction and/or perceived attribute levels may be influenced by context manipulations, individual differences, and other factors thought to affect CL, AL, etc., parallel effects on overt behavior may be quite weak. T would urge future researchers in this area to attempt to separate effects of their theoretically critical manipulations on psychological perceptions of satisfaction from effects of those same manipulations on what Wyer (1974) has called "response language".

Louviere

The hierarchical task structure Louviere proposes is a promising and clever approach to simplifying highly complex conJoint problems. For problems with a very large number of attributes -- larger than the simple illustration Louviere presented -- respondents may have difficulty forming overall impressions of each alternative. In such cases, the researcher could ease the momentary information load on the respondent by using the two at a time tradeoff approach rather than the full profile approach, but only at a cost of asking the consumer to respond to a large number of two-way matrices. Thus, Louviere's method has two advantages: it is easier for the respondent than full profile methods, and it is inexpensive from a data-collection standpoint.

A third potential advantage is that the hierarchical method may increase the precision of parameter estimates compared with those that would be obtained by full profile conjoint analysis, so that the utility estimates might be more reliable in a test retest sense. Evidence from a number of studies (e.g., Einhorn 1971; Ogilve and Schmitt 1979; Scott and Wright 1976) shows that all models fit worse as the number of attributes to be integrated simultaneously increases. This result could be due to interprofile shifts in strategy under conditions in which descriptions of the alternatives are complex, but it could be due simply to increases in true error variance as information load increases. If so, Louviere's method may provide more precise utility estimates than full profile conjoint analysis.

However, I see that future work with the hierarchical approach must address the following issues. First, while very large full profile tasks may create a difficult information load for the respondent, he or she must deal with the product or service in all of its messy complexity when making choices out in the real world. Thus, one might argue that the measurement method should mirror the actual choice situation, as Wright and Kriewall (1980) have suggested, and that the simplified hierarchical approach may be at 2 disadvantage on this dimension.

In this regard, I might suggest that the acid test of the utility estimates derived from the hierarchical method may not be whether they recover estimates derived from a full profile ranking or rating task, but whether they recover those derived from Louviere's (1983) ingenious experimental design approach to constructing choice sets for discrete choice modeling. This might be a natural first step to take in the external validity testing of the hierarchical method that Louviere mentioned.

A related issue is whether the implied judgment model suggested by the hierarchical method corresponds to that which is used when consumers are choosing from among products or services with a very large number of attributes. First, some authors have suggested that consumers give more weight to negative information under conditions of high information load. If so, full profile results could display a number of diverging interactions that might not appear in the simpler hierarchical tasks.

Second, Louviere mentioned that it remains to be seen whether the grouping scheme used in the hierarchical task affects its ability to recover utilities generated by the full profile approach. Phelps and Shanteau (1979), studying the judgment processes of expert swine judges, found evidence that they took a large number of dimensions into account, apparently by a process similar to that Louviere requires of his subjects. First, they grouped attributes into logically related subsets, and judged the swine on each subset. Then, they combined these subgroup judgments into an overall judgment. If -consumers in the real world autonomously group attributes into subsets and follow a similar process, and if the attributes within each subset are processed configurally, it is possible that Louviere's method might group attributes "improperly" -- in the sense that the researcher's groupings do not correspond to those of the respondent - and thus, the implied judgment model produced by the hierarchical method might fail to capture certain real configural effects.

Finally, if respondents do not autonomously chunk attributes in a large full profile task, but rather, simply attend to a reduced number of attributes under conditions of high information load, the hierarchical results cause one to overestimate the number of attributes that contribute significantly to real world choice. Thus, there are three ways that the hierarchical method could go wrong: (a) by creating conditions unfavorable to the detection of (ordinal) attribute interactions caused by increased weighting of negative information under high information load; (b) by failing to capture configural chunking strategies consumers may adopt autonomously to simplify very complex decision tasks in the real world; and (c) by failing to capture the intra-subject variance in attribute weights that emerges when consumers ignore many attributes under conditions of high information load.

On the other hand, potentially relevant evidence reported to date suggests that (b) and (c) may not prove problematic. Malhotra (1982) has reviewed several studies that found that the derived importance weights for a set of attributes were unaffected by the inclusion of other attributes in the (full profile) stimulus descriptions. In these studies, information about" attributes was "embedded" in stimulus descriptions of" + k attributes, with k systematically manipulated. As far as I can tell, the" "embedded" attributes in these studies were not chosen on the basis of any logical grouping (as in Louviere's technique). Thus, the fact that the estimated part-worths of these (n) attributes were virtually the same when presented alone as when accompanied by k other attributes suggests that the success of Louviere's hierarchical technique may be fairly insensitive to the choice of the grouping scheme used to partition attributes into subsets. However, it should be noted that these results are not germane to the issue of the validity of Louviere's "overall" design that links subgroup ratings to predict overall responses to fully described options.

In sum, Louviere's hierarchical information integration technique appears highly promising. Future research is needed to test the issue of whether the overall model implied by responses to his hierarchical tasks mirrors that underlying real world choices in which consumers must deal with highly complex products and services. I have attempted to lay out some plausible reasons why this might not be the case, in hopes that future research can address the issue of when the hierarchical technique is likely to be most and least appropriate.

REFERENCES

Birnbaum, Michael H. (1974), "Using Contextual Effects to Derive Psychophysical Scales," Perception and Psychophysics, 15 (1), 89-96.

Chakravarti, Dipankar and Lynch, Jr., John G. (1983), "A Framework for Exploring Context Effects on Consumer Judgment and Choice," in R. Bagozzi and A. Tybout (eds.) Advances in Consumer Research, Vol. 10, Ann Arbor, MI, Association for Consumer Research, 289-297.

Cohen, Jacob (1978), "Partialled Products Are Interactions; Partialled Powers Are Curve Components," Psychological Bulletin, 85, 858-866.

Einhorn, Hillel J. (1971), "Use of Nonlinear and Noncompensator, Models as a Function of Task and Amount of Information," Organizational Behavior and Human Performance, 6 (1), 1-27.

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

Helson, Harry (1964), Adaptation Level Theory, New York, Harper and Row.

Holbrook, Morris B. (1977), "Comparing Multiattribute Attitude Models by Optimal Scaling," Journal of Consumer Research, 4 (3), 165-171.

Jaccard, James (1981), "Attitudes and Behavior: Implications of Attitudes toward Behavioral Alternatives," Journal of Experimental Social Psychology, 17 (3), 286-307.

Louviere, Jordan L. (1983), "Integrating Conjoint and Functional Measurement with Discrete Choice Theory: An Experimental Design Approach," in R. Bagozzi and A. Tybout (eds.) Advances in Consumer Research, Vol. 10, Ann Arbor, MI, Association for Consumer Research, 151-156.

Malhotra, Naresh (1982) "Structural Reliability and Stability of Nonmetric Conjoint Analysis," Journal of Marketing Research, 19 (May), 199-207.

Ogilve, John R. and Neal Schmitt (1979), "Situational Influences on Linear and Nonlinear Use of Information," Organizational Behavior and Human Performance, 23 (April), 292-306.

Ray, Michael L. and Roger M. Heeler (1975), "Analysis Techniques for Exploratory Use of the Multitrait-Multimethod Matrix," Educational and Psychological Measurement, 35 (Summer), 255-265.

Scott, Jerome E. and Peter Wright (1976), "Modeling an Organizational Buyer's Product Evaluation Strategy: Validity and Procedural Considerations," Journal of Marketing Research, 13, 211-224.

Thibaut, John W. and Harold H. Kelley (1959), The Social Psychology of Groups, New York: John Wiley and Sons.

Wright, Peter and Mary Ann Kriewall (1980), "State-of-Mind Effects on the Accuracy with Which Utility Functions Predict Marketplace Choice," Journal of Marketing Research, 17 (August), 277-293.

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