An Approach to Measuring Thought Patterns and Gauging Causal Schemata

William R. Dillon, University of Massachusetts
Chris T. Allen, University of Massachusetts
Marc G. Weinberger, University of Massachusetts
Thomas J. Madden, University of Massachusetts
ABSTRACT - Introduced here is a measurement technique that may allow identification of thought patterns of the sort suggested bs attribution theory. Basically, the approach entails a two-stage procedure wherein thought verbalization data are recast into thought categories; these provide the framework for gathering similarity data that are input into a functional scaling method. A sample application illustrates the approach and furnishes data for assessing its face and criterion-related validity.
[ to cite ]:
William R. Dillon, Chris T. Allen, Marc G. Weinberger, and Thomas J. Madden (1982) ,"An Approach to Measuring Thought Patterns and Gauging Causal Schemata", in NA - Advances in Consumer Research Volume 09, eds. Andrew Mitchell, Ann Abor, MI : Association for Consumer Research, Pages: 281-286.

Advances in Consumer Research Volume 9, 1982      Pages 281-286

AN APPROACH TO MEASURING THOUGHT PATTERNS AND GAUGING CAUSAL SCHEMATA

William R. Dillon, University of Massachusetts

Chris T. Allen, University of Massachusetts

Marc G. Weinberger, University of Massachusetts

Thomas J. Madden, University of Massachusetts

ABSTRACT -

Introduced here is a measurement technique that may allow identification of thought patterns of the sort suggested bs attribution theory. Basically, the approach entails a two-stage procedure wherein thought verbalization data are recast into thought categories; these provide the framework for gathering similarity data that are input into a functional scaling method. A sample application illustrates the approach and furnishes data for assessing its face and criterion-related validity.

INTRODUCTION

The general problem of assessing communication effects has been and will continue to be of major interest to consumer researchers. Communication research has evolved from simplistic, black-box models to recent approaches which emphasize and specify cognitive mediators of communication effects. Current work in assessing and explaining the impact of information stimuli is dominated by cognitive structure (e.g. Fishbein and Ajzen, 1975) and cognitive response models (e.g. Wright, 1980), or a combination of the two (e.g. Lutz and Swasy, 1977).

A paradigm which adopts a cognitive processing perspective, but has been employed to only a limited degree in explaining communication effects (e.g. Smith and Hunt, 1978), is attribution theory. As characterized by Kelley (1972; 1973), attribution theory entails a general conception of the way people think about and analyze cause and effect data; the theory posits that informational stimuli will evoke configurations of thoughts in the attributor. A major thesis in this paper is that attribution theory offers richness for conceptualizing thought processes in communication contexts that has not been adequately explored by consumer researchers. The purpose of the paper is to introduce, and report an exploratory application of, a measurement technique that may facilitate identification of thought patterns of the sort suggested by attribution theory, and thus prove useful as a tool for assessing communication effects.

ATTRIBUTION PROCESSES AND COMMUNICATION EFFECTS

A key concept in Kelley's (1972; 1973) theory in the situation where a person is making a causal inference from a single observation of some event (say, a single exposure to some persuasive message), is the causal schema or array. [Unfortunately, attribution theory is not precise in delineating situations where single inference rules, as opposed to the covariation principle (based on multiple observations over time), are appropriate. The assumption here is that single inference rules are relevant where the attributor views a single informational stimuli. Mizerski, Golden, and Kernan (1979) provide a discussion concerning the apparent controversy inherent in such an assumption (see p. 133).] A causal schema refers to the way a person thinks about plausible causes in relation to some observed event (Kelley, 1973); it is the framework within which bits and pieces of relevant information are fit in order to draw reasonably good causal inferences (Kelley, 1973). Consumer causal schema is a largely unexplored area (Mizerski, Golden, and Kernan, 1979), but may prove particularly useful for understanding communication effects, because if one can identify the type of causal array evoked by a message, one should be able to make a statement about the inferences the attributor will draw.

Kelley seems to be suggesting that if one were to examine verbalized thoughts (Wright, 1980) elicited after exposure to information stimuli, at least in some instances, certain patterns should be identifiable that indicate the message receiver evoked a causal array. That is, specific thought statements should appear linked in logical configurations that reveal the type of causal inference derived. For instance, the following pattern or set of thoughts would indicate the causal array referred to as augmentation (Kelley, 1972; 1973) had been evoked by an automobile advertisement about handling characteristics: "that was an extremely difficult handling test"; "yet the auto performed flawlessly"; "this auto must be very maneuverable". Such a pattern, of course, indicates that the attributor was making a very strong inference about a characteristic of the object (the auto) in question.

Conceptually, Kelley's notion of causal schema is appealing because of the emphasis on identifying thought configurations to explain message impact. As noted by Wright (1980) approaches to classification of thought verbalization data used in "cognitive response theories" to date uniformly key on individual thoughts, and in so doing may be ignoring important information about the communication's impact that would be revealed in patterns. There is limited evidence derived via verbalization data that persuasive communications do evoke attributional analyses generally (Smith and Hunt, 1978), but little has been done in the way of examining causal arrays via analyses of thought configurations.

GAUGING CAUSAL SCHEMATA

Given one accepts the notion that examining causal schemata through analyses of thought patterns would be a useful exercise, the key issue becomes "how does one go about it"; indeed, limited progress in applying the attribution paradigm may be at least as much a function of measurement issues as conceptual ones. To gauge the causal arrays evoked by realistic informational stimuli, one's measurement technique must possess inherent flexibility to allow identification of a variety of thought configurations. Such flexibility is a must because different attributors can react to the same information stimuli in very divergent ways (Kelley, 1973): this can be due to differences in attributors' prior beliefs that affect their perceptions of the information, or differences in the causal schemata they have learned to use and rely on in making inferences from information of the sort they are being exposed to. Notably, although two types of schemata--augmentation and discounting [Under the discounting principle (Kelley, 1973), the role of a given cause in producing a given effect is discounted by the attributor if other plausible causes are also present. Augmentation operates in situations where nonstimulus or external causes may inhibit the observed effect, and thus serve to heighten the impression that a stimulus cause is present and is a potent force (Kelley, 1973).] -- appear to be evoked so frequently they have achieved the status of attribution principles (Mizerski, Golden, and Kernan, 1979), Kelley's (1972; 1973) work indicates there is potentially an infinite variety of possible schemata, underscoring the need for a flexible measurement device.

Free-flow thought verbalizations provide the ultimate in flexible measurement, and Wright (1980) notes judges could be instructed to code such data according to whether reasoning chains of theoretical interest are evidenced. However, verbalization data are cumbersome to analyze and difficult to quantify, and a problem that continues to plague use of these thought weighting (Wright, 1980) - would seem even more problematic when examining thought patterns. For example, should one equal weight all thoughts in a given configuration, or do some thoughts dominate the pattern and appear to be more intense than others? Free-flow methods to not deal very well with such questions (Wright, 1980). For gauging causal arrays then, it would be desirable to have a technique which generally furnishes more quantifiable output than the free-flow approach, and that specifically allows one to address the weighting issue.

Described in the next section is a measurement approach which possesses inherent flexibility, and ultimately supplies more quantifiable output, and seems to deal with the weighting issue in a more appealing way, than free-flow verbalizations. An application of the approach is discussed, and data from this application are used in demonstrating face and criterion-related validity. Like others, this is not a methodology without potential limitations: construct validity issues are raised in the concluding discussion. It is not being suggested that cognitive (attribution) processes are actually revealed via this technique; rather, similar to what Wright (1980) assumes about verbalization data, the proposed method can be portrayed as yielding an indication and description of a process that purportedly takes place when persons are exposed to informational stimuli.

A THOUGHT MATCHING APPROACH

The approach involves a two-stage procedure which integrates thought verbalization and functional scaling methodologies. In the first stage subjects' free-flow reports of message-evoked thoughts are used to define a set of plausible causes; in essence, one is determining the thought categories and descriptions of the content of the categories that persons are evoking in responding to a particular message. The second stage derives functional scale weights for the thought categories or plausible causes via a metric modeling of a second group of subjects' perceptions of the similarity between their thoughts and a variety of thought profiles. A sample application seems the most efficient way of describing the approach in detail.

Context of the Sample Application

In June of 1978 Consumers Union released via press conference and its magazine Consumer Reports highly negative information about Chrysler's Horizon and Omni automobiles. All the major network evening news programs carried the story which included a dramatic film demonstrating the purported handling deficiencies of these new autos. Knowing of the impending news releases Chrysler prepared a reply to Consumers Union's condemnation. Videotapes of the Consumers Union story, Chrysler's reply, and a fifteen minute segment of the NBC Nightly News were obtained; this material was used in developing the informational stimuli. [Five different treatment configurations were utilized; they differed in terms of how much negative or unfavorable product information they contained.] The broadcast message concerning Horizon/Omni is treated as the focal event in subject's attributional analyses.

Subjects

Two groups of students were utilized. In stage one, twenty-five graduate students were asked to report the thoughts that had occurred to them as they viewed the information about Horizon/Omni. Sixty-eight undergraduates participated in stage two. As an initial step, pre-exposure belief and intention data were collected from students in the second group. Two weeks later they viewed the videotapes, performed the thought matching task, and again provided belief and intention (post-exposure) information.

Stage One: Prescreening the Informational Stimuli

Immediately after viewing information about the Chrysler/ Consumers Union controversy, subjects completed questionnaires which in part asked for a listing of all thoughts that had passed through their minds during viewing. The request was general in nature (i.e., contained no specific priming), no severe space limits were imposed, and ten minutes were allotted for the listing. [Wright (1980) reports that a fairly short time limit may be preferred to guard against made-up thoughts. However, in the context of this use of thought verbalizations, we felt that to potentially exclude valid thoughts would run the risk of overly restricting the generality and relevance of the thought profiles.] Afterwards, a focus group session was conducted to clear up potential ambiguities regarding participants' thought verbalizations, and then they were debriefed and dismissed.

The thoughts were evaluated independently by four judges (the authors) in identifying the various categories of plausible causes for the information that subjects thought about as they viewed the stimuli. The separate content analyses proved similar and revealed that, for the most part, thoughts tended to fall into three categories related to: (1) the safety of the automobile (Horizon/Omni) itself; (2) the general credibility/fairness of Consumers Union; and (3) the validity of the testing procedures used by Consumers Union. Below are examples of thoughts listed in each category:

l. Safety of the automobile itself.

"This confirmed my opinion--the car is unsafe."

"There's something wrong with this car."

"I wondered whether this auto is any less safe than others."

"I thought that this auto is as good as any other compact."

2. Credibility/fairness of Consumers Union.

"I questioned the intent of Consumers Union."

"A low rating from such a reputable source must mean that the auto should be avoided."

"I thought about the fairness of Consumers Union."

"Consumers Union is biased against Chrysler Corporation.

3. Validity of the testing procedure.

"I questioned whether the test simulated real driving conditions."

"I wondered under what circumstances a driver would have cause to twist the wheel 90 degrees and release."

"I thought about the control characteristics of the auto and how poor they were as demonstrated by the test."

"The test appeared legitimate."

Within each thought category, a variety of sentiments were expressed. For example, the automobile was perceived to be unsafe by some, but about as safe as other subcompacts by others; to some, Consumers Union was perceived as fair, yet others called its credibility/fairness into question; the test procedure was considered legitimate by some, while others questioned its realism. In general, the evaluative nature of subjects' thoughts reflected either support, disbelief or uncertainty about the car's safety, Consumers Union's credibility/fairness, and/or the testing procedure's validity. Notably, careful examination of the thought verbalizations also showed that certain thought patterns tended to occur together, that thought linkages did exist, and that the thought statements had a definite evaluative component.

Stage Two: Thought Profiles and Thought Matching

Twenty-seven unique thought profiles containing three thought statements, one from each of three thought categories, were generated from the framework shown in Figure 1. Notice that statements within each category were framed in such a way as to reflect support, disbelief or uncertainty about the automobile's safety, the fairness of Consumers Union, and the test's validity. After the second group of subjects watched the videotaped material, they were asked to recall the thoughts that passed through their minds while viewing the information about the Chrysler/Consumers Union controversy. Subjects were given a stack of twenty seven "3x5" cards; each card held one of the twenty-seven thought profiles. (A sample profile card is displayed in Figure 2.) Subjects were first instructed to sort the carts into three piles: they placed in one pile carts containing statements very similar to their thoughts; in a second, cards containing statements somewhat similar to their thoughts; and in a third, cards with statements dissimilar to their thoughts. Each time instructions were given, strong emphasis was put on subjects recalling the thoughts about the Chrysler/Consumers Union controversy that came to their minds as they watched the videotape. After the sorting, subjects recorded a similarity judgment for each of the profiles on a nine point scale ranging from (1) "these statements completely matched my thoughts" to (9) "these statements did not match my thoughts at all" By rating the similarity of all twenty-seven profiles, each participant evaluated every possible combination of thoughts.

Functional Scaling

Utilizing these similarity judgments, a simple metric scaling of the profile data is possible through use of a dummy-variable coding procedure. Since each profile is defined by three thought statements, it can be uniquely represented by two dummy variables, where each dummy reflects the absence or presence of each thought statement.

For example, from the framework of Figure 1, the code (00,10,01) indicates the profile with thought statements:

-Horizon/Omni is less safe than other small cars.

-I have no way of knowing how fair Consumers Union is in their product ratings.

-Consumers Union's procedure seems very realistic and reflects what a person might encounter when driving.

The functional scaling is accomplished by use of standard regression analysis. With dummy-variable coding the takes the form:

Yij = b0+b1d2i(1) + b2D3i(1) + b3D2i(2) + b4D3i(2) + b5D2i(3) + b6D3i(3)    (1)

where Yij (i = 1,2,..., 27) denotes the jth person's similarity rating of the ith belief profile, = denotes least squares approximation, b6 denotes the intercept term, b1, b2,..., b6 denote the partial regression coefficients, D2i(k) and D3i(k) denote the respective second and third thought statements for the kth (k s 1,2,3) thought category.

FIGURE 1

THOUGHT CATEGORY DESCRIPTIONS

FIGURE 2

INSTRUCTIONS AND SAMPLE PROFILE CARD

The partial regression coefficients will indicate the extent to which a subject reported that a given thought statement matched his or her own thoughts. Thus, it is possible to describe the pattern of plausible causes/thought categories evoked in the person's thinking, and through simple computations assign scale values (weights) to the plausible causes. These weights furnish an indication of the relative intensity of the various individualized thoughts within a configuration. Note that unlike previous approaches using free response data, the weights are not self-explicated but are derived unobtrusively via the functional scaling procedure.

FACE VALIDITY OF THE APPROACH

Data from the sample application are used here to examine face validity, and in the next section, to evaluate criterion-related validity. Face validity is assessed in terms of (l) how well the twenty-seven profiles seemed to allow subjects to characterize their thoughts, and (2) the plausibility of the kinds of thought patterns identified.

Distribution of R2's

Functional scaling of the thought profile data entailed fitting sixty-eight regressions, one for each subject; the dependent measure in each was the reported similarity score assigned to a particular profile. The observed R2's suggest that the twenty-seven profiles did allow subjects to adequately report their own thoughts, offering support for face validity of the approach. The median R2 was 0.77; fifty-seven subjects had R2's > .6, and just seven had R2's < .5. Where poor fits did surface, the difficulty of the data collection task might have been a problem, or it may have been that for some persons the profiles did not allow them to accurately represent their thoughts; alternatively, some may have had no thoughts at all. In any case, the efficacy of a given measurement application can be assessed in part from the fitting process; this would appear to be an attractive feature of the method.

Derived Thought Configurations--Discounters and Enhancers

Table 1 presents summary output of the functional scaling method for two sample subjects (labeled Persons D and E). Low values for partial regression coefficients (p.r.c.'s) mean that the respective plausible cause apparently matched a thought that had come to the person's mind while watching the videotape. [All results are based on raw data; no standardization was necessary because means and standard deviations were similar across subjects.] (The p.r.c.'s have been scaled so zero is the smallest value.) By identifying those thoughts with p.r.c.'s of zero, a "most similar" overall thought profile can be identified.

The modal thought profile is different for the two sample subjects: Person D thought the Horizon/Omni as about as safe as other small cars, Consumers Union fair, and the test procedure unrealistic; Person E thought the Horizon/ Omni less safe, Consumers Union fair, and the test realistic. While attribution theory does not supply clear rules for selecting an attributional focus and an attendant theoretical paradigm (Mizerski, Golden, and Kernan, 1979), it seemed apparent in the present context that the object of the attribution process was the automobile. Thus, Persons D and E appear quite different in terms of their likelihood of making an internal or stimulus attribution about the auto's safety; this difference appears related to their thoughts about the test procedure external factor, but not their thoughts about Consumers Union--another external factor.

The p.r.c.'s supply additional insight about one's message-evoked thoughts: if within a thought category they are all of the same magnitude, then either all three thought statements were considered plausible causes, or, more likely, they tended not to match the person's thoughts at all. On the other hand, if the within thought category p.r.c. spread is large, then one thought was highly discriminal and reflective of "the" plausible cause. To provide a method for capturing idiosyncratic differences weights were assigned to the thought statements which reflect the size of the associated p.r.c. as well as the discriminal power within a category. The weights were calculated by taking the ratio of the exponential of each within category p.r.c.; [The transformation used to compute the weights was of the form: EQUATION a = 1,2,3; j = 1,2,3; where wj(a) denotes the estimated weight for the jth thought statement of the ath thought category, e = 2.718..., bj(a) is the p.r.c. (with algebraic sign reversed) associated with the jth thought statement of category a. This type of transformation induces weights which will be larger as the thought category's discriminality increases; in addition, the rank order of weights within a thought category is invariant over the transformation.] these weights are shown for subjects D and E in the last two columns of Table 1. This type of normalization is analogous to the conditional logit specification used in consumer choice models (McFadden. 1976).

TABLE 1

A COMPARISON OF TWO PERSONS' (D AND E) THOUGHT PATTERNS

Referring to Table 1, the within category p.r.c. ranges for the internal and external causes are similar for Persons D and E; thus, although they held different thoughts in the case of the test procedure, the weight assigned to their respective most similar thought is about the same for both persons. Note that for the internal cause category, marked differences in thoughts were exhibited. Person D's did not seem to question the safety of the Horizon/Omni, whereas Person E's did.

Given their apparent thought patterns, it is reasonable to expect that Person D would be less likely than E to revise his beliefs about the auto. Indeed, D's thoughts seem to identify a causal schema--discounting- familiar to attribution researchers; D discounted the negative information about the auto's safety because he perceived the test was unrealistic. Conversely, E perceived the test realistic, Consumers Union generally fair, and these thoughts seemed to enhance E's suspicions about the auto's safety. These "discounting" and "enhancing" configurations typified many of the participants' thought processes. Just as was done for D and E, all subjects were classified as discounters, enhancers, or neither based on observed patterns in their respective p.r.c.'s.

Unlike discounting, enhancing is not a schema examined in prior attributional research, but would appear to be simply the flip-side of discounting: just as external factors can serve to inhibit internal or stimulus attribution, they also may function to enhance it. Kelley's (1972; 1973) work seems to portray augmentation as the schema which is the natural opposite of discounting; however, thought patterns like those of subject E do not fit Kelley's rather strict definition of augmentation. The fact that theoretically-plausible thought patterns were identified establishes face validity for the approach; the identification of a plausible, but somewhat unexpected thought configuration--enhancing--seems to demonstrate the technique's desired flexibility.

CRITERION-RELATED VALIDITY

Discounting, Enhancing, and Stimulus Attribution

Although direct measures of stimulus attributions were not taken, the output of the functional scaling procedure does provide an indicator for such attributions in the derived weights. Larger weights on the though; "the auto is less safe" should reflect greater likelihood of a causal attribution about the auto/a stimulus attribution (e.g., D's weight was .03, E's was .91). If in fact the discounter and enhancer thought patterns are realistic portrayals of causal schemata actually evoked, persons so classified should differ on their stimulus attribution scores.

Each subject's thought configuration was categorized judgmentally as being suggestive of discounting, enhancing, or neither, based on the pattern of p.r.c.'s derived for their thoughts on the two external factors. The process resulted in 41 enhancers, 17 discounters, and 10 unclassifiable patterns. An analysis of variance was then run on stimulus attribution scores for enhancers versus discounters. The first row of Table 2 shows the stimulus attribution scores are indeed significantly higher for subjects whose causal arrays reflected enhancing.

Discounting, Enhancing, and Cognitive Structure

Because it is reasonable to expect that causal explanations play a role in determining a plan of action and decision making, variations in the nature and composition of causal schemata evoked should yield consequent changes (i.e., revisions) in cognitive structure (Fishbein and Ajzen, 1975; Mizerski, Golden, and Kernan, 1979). Stronger stimulus attributions, for example, should yield stronger changes in beliefs. In the following analysis, difference scores (postexposure-preexposure) are used to be consistent with the definition of an attribution as a revision of a belief in light of new information (Fishbein and Ajzen, 1975).

TABLE 2

ANALYSIS OF DEPENDENT VARIABLES -- DISCOUNTERS VS. ENHANCERS

Stimulus attribution scores were correlated with (1) (postexposure-preexposure) belief scores about the safety of Horizon/Omni, and (2) the change in the likelihood of considering Horizon/Omni if purchasing a new automobile as evinced in (postexposure-preexposure) intention scores. Results were encouraging: the correlations were r = 0.72 (p <.001) for belief revisions and r = 0.64 (P<.001) for intention revisions. Thus, a substantial amount of variation in the change in beliefs and purchase intentions was accounted for by the stimulus attribution scores. These scores, of course, are just weights derived from the functional scaling procedure.

Enhancers should have experienced greater changes in their beliefs and purchase intentions concerning Horizon/Omni than discounters. As a final step in the investigation, an analysis of variance was run on the (postexposure-preexposure) belief and intention scores, and the results are summarized in Table 2. As expected, the second and third rows of the table show that mean revision scores are significantly higher for enhancers than for discounters. Since higher mean revision scores reflect more unfavorable changes in beliefs and purchase likelihoods (remember, the treatments contained negative or unfavorable information about Horizon/Omni), the results suggest a strong measure of criterion-related validity for the measurement approach.

CONCLUDING REMARKS

Our purpose was to introduce a measurement approach for identifying theoretically-plausible thought patterns evoked by informational stimuli. In one exploratory study it is impossible to deal with all questions concerning such a new approach, but the results of this effort seem promising. Further refinement may yield a technique that will Prove useful both for theory testing, and in practical applications like evaluating the impact of advertisements.

Successful use of the method depends on accurately defining the number of thought classes and thought statements potentially evoked in response to a message. If past studies in the attribution area are any indication, the typical research design should allow the thought process to be characterized by a 1 number of thought categories and within thought statements. Where the ability to adequately characterize the thought process with a limited number of thought profiles is suspect, the researcher can adopt (orthogonal) fractional factorial designs which will allow the estimation of weights on an unconfounded basis. This approach would be analogous to that which is used in conjoint analysis (Green, 1974; Green and DeSarbo, 1978).

The major concern about the approach at this point involves construct validity. Presenting people with thoughts to rate as similar/dissimilar to their own may encourage them to report thoughts they did not have; moreover, the structure imposed on the reporting by the thought matching procedure may culminate in derivation of artificial thought patterns. Stage one of the approach provides a check against the latter concern since one can assess whether theoretically-plausible patterns emerged in the free-flow verbalization data. In this application, analysis of the free-flow data did indicate the presence of thought patterns similar to those derived via functional scaling of the thought matching data. The former concern is more problematic; similar concerns have been expressed about free-flow data (e.g. Lutz and Swasy, 1977) because asking persons to list their thoughts presumes they had thoughts. Construct validity of both the proposed approach and thought verbalization methods generally might be heightened simply by giving subjects a legitimate option to report they had no thoughts.

Empirical research can help lessen concerns about construct validity. Experiments could be designed using treatment messages which theoretically would yield different attributions: the technique would be used to examine whether the expected differences in attributions were evidenced in subjects' thought patterns. Further, research might be designed to allow one to make causal inferences about whether identified thought patterns actually mediated the effect of divergent persuasive messages on beliefs. Such research seems to be the logical "next step" in developing the method introduced here.

REFERENCES

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

Green, Paul E. (1974), "On the Design of Choice Experiments Involving Multi-Factor Alternatives," Journal of Consumer Research, 1, 61-68.

Green, Paul E. and DeSarbo, Wayne S. (1978), "Additive Decomposition of Perceptions Data Via Conjoint Analysis," Journal of Consumer Research, 5, 58-65.

Kelley, Harold (1972), Causal Schemata and the Attribution Process (New York: General Learning Press).

Kelly, Harold (1973), "The Process of Causal Attribution," American Psychologist, 28, 107-128.

Lutz, Richard J., and Swasy, John L. (1977), "Integrating Cognitive Structure and Cognitive Response Approaches to Monitoring Communications Effects," Advances in Consumer Research, Vol. IVS 363-371 (ed.) W. Perreault, Jr., Atlanta: Association for Consumer Research.

McFadden, Daniel (1976), "Quantal Choice Analysis," Annals of Economic and Social Measurement, 363-390.

Mizerski, Richard, Golden, Linda L., and Kernan, Jerome B. (1979), "The Attribution Process in Consumer Decision Making," Journal of Consumer Research, 6, 123-140.

Smith, Robert E., and Hunt, Shelby (1978), "Attribution Processes and Effects in Promotion Situations," Journal of Consumer Research, 5, 149-158.

Wright, Peter (1980), "Message-Evoked Thoughts: Persuasion Research Using Thought Verbalizations," Journal of Consumer Research, 7, 151-175.

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