Some Conceptual and Empirical Issues Regarding Cognitive Structure

ABSTRACT - This paper discusses conceptual, methodological, and application issues raised by the Laroche et. al., Richins and Bloch, and Durand and Lambert papers. It is concluded that while the notions of brand categorization and cognitive complexity are potentially important cognitive variables in the study of buyer behavior, much must be done to better understand the theoretical premises of each concept. Lack of clear theoretical definition has been a major factor in the employment of measurement techniques that fail to effectively "tap into" the construct of interest. Recommendations are made for-improving methodologies and suggestions offered for future research.

Citation:

Philip A. Dover (1983) ,"Some Conceptual and Empirical Issues Regarding Cognitive Structure", in NA - Advances in Consumer Research Volume 10, eds. Richard P. Bagozzi and Alice M. Tybout, Ann Abor, MI : Association for Consumer Research, Pages: 705-708.

Advances in Consumer Research Volume 10, 1983      Pages 705-708

SOME CONCEPTUAL AND EMPIRICAL ISSUES REGARDING COGNITIVE STRUCTURE

Philip A. Dover, Boston University

ABSTRACT -

This paper discusses conceptual, methodological, and application issues raised by the Laroche et. al., Richins and Bloch, and Durand and Lambert papers. It is concluded that while the notions of brand categorization and cognitive complexity are potentially important cognitive variables in the study of buyer behavior, much must be done to better understand the theoretical premises of each concept. Lack of clear theoretical definition has been a major factor in the employment of measurement techniques that fail to effectively "tap into" the construct of interest. Recommendations are made for-improving methodologies and suggestions offered for future research.

Laroche et. al. Paper

The Brisoux-Laroche paradigm for consumers' brand categorization strategies extends earlier work on the categorization processes developed by Howard and Narayana-Markin. Their conceptualization suggests that brands consumers are aware of can be classified into either a processed or foggy set. The latter includes brands the consumer knows exist but for which he has no specific knowledge. The Laroche et. al. paper presents survey data that appears to generally support the Brisoux-Laroche categories of evoked, hold, reject and foggy. While research that adds to our understanding of how brand information is stored in memory and subsequently retrieved and used in brand choice is much needed the contribution of the present-research must first be critically examined from both conceptual and methodological perspectives.

Conceptual Issues: The essential components of the Brisoux-Laroche model are shown below.

FIGURE 1

PART OF THE BRISOUX-LAROCHE MODEL

The authors emphasize that the model is applicable in a routinized response behavior situation. That is, consumers have become familiar with the brands within the processed set either through actual experience or information assimilation (e.g., advertising, publicity) and brand choice is an automatic and cognitively stress-free process. As such, some information processing has taken place and a consequent cognitive structure formed toward each brand. It is to be expected that consumers can easily and reliably articulate their beliefs, attitudes and intentions-to-purchase these brands.

The situation is rather different, however, for brands within the foggy set. It is postulated that "such brands do not have significant meaning as they cannot really be distinguished in terms of the evaluative criteria of that product class." In hierarchy of effect terms only brand name awareness has been attained, and then only possibly at a minimal level (a non-semantic memory trace, ;f. Krugman, 1977?). It thus becomes meaningless to measure product attribute knowledge, attitude and intention -to- purchase toward foggy set brands as such cognitive structure has not yet been developed. These measures in the Laroche et. al. study are likely to be artifactual and therefore should be treated with extreme caution.

Although stated intention -to- purchase may be non-existent (a 0.25% chance that a foggy set brand would appear -in the next 10 toothpaste purchases, according to the Laroche study), this is not to say that foggy set brands will not appear in future purchases. Let us assume that the product of interest is perceived as a low involvement category. In this situation passively learned information can precipitate a purchase decision prior to brand evaluation. For example, passive exposure to a TV commercial may lead to the storage of a few bits and pieces of information in the consumer's mind without associated cognitive processing. A visit to a supermarket, and observation of point of purchase displays of the advertised brand, may stimulate recall of the commercial and encourage trial. Brand evaluation will only follow actual experience with the brand. For foggy set brands contained within a low involvement hierarchy, viz.

Awareness ----> Action ----> Attitude

a communication campaign involving simple, mainly visual, repetitive, TV dominated messages should help move brands from the foggy to the processed set. The movement of brands from the foggy to the processed set becomes more difficult when dealing with high involvement situations. Consumers must be encouraged to fully process brand information before trial can occur. It should be noted, however, that once this processing starts the brand moves from the foggy to the processed set. I rather wonder whether the foggy set is a relevant concept with a high involvement product category. It seems likely that any brand one is aware of in a personally important category requires some classificatory activity. Information processing becomes an ongoing dynamic, active pursuit.

The significance of the foggy set concept may well hinge on whether the product category is perceived as high or low involvement by the consumer. How can toothpaste, the test product in the Laroche study be classified? Use can be wade of Sherif's theory of Social Judgment (Sherif et. al. 1965). Sherif describes an individual's position on an issue based on the individual's involvement with the issue. Sherif operationalizes this concept of involvement by identifying a latitude of acceptance (the positions the individual accepts), a latitude of rejection (positions the individual rejects), and 8 latitude o, noncommitment (positions toward which the individual is neutral). A highly involved individual having a definite opinion about the issue would accept very few positions and reject a wide number of position (narrow latitude of acceptance and wide latitude of rejection). An uninvolved individual would find more positions acceptable (wide latitude of acceptance) or would have no opinion about the issue (wide latitude of noncommitment). For our purposes, brands can be substituted for issues while the analogy between latitudes and processed sets should be clear. Set size data can be taken directly from the Laroche study to indicate the respective sizes of the latitudes of acceptance, noncommitment, and rejection. Rounding off the numbers we have:

FIGURE 2

CLASSIFICATION OF TOOTHPASTE BRANDS

It is not clear whether foggy set brands should be included in the non-commitment or rejection latitudes. That decision should not, however, affect the determination of toothpaste as low or high involvement. Although there is no exact procedure for classifying individuals as high or low involvement, Toy (1982) has specified that subjects who place less than 40% of brands in the acceptance and neutral regions should be classified as highly-ego-involved. Although the selection of 40% is arbitrary, Toy has shown that results are consistent with other operationalizations of the involvement concept.

It can be seen that substantially more than 40% of the toothpaste brands are placed within the latitudes of acceptance or non-commitment (6 out of 8.5 or 70%). This suggest that toothpaste is a low involvement product category. -Further testing should be undertaken to establish the range of product categories, on a low to high involvement continuum, to which the foggy set notion is applicable. It may prove that the Narayana-Markin model is appropriate for high involvement situations while the Brisoux-Laroche categorization relates to low involvement products. If this is so, the measurement of cognitive structure variables for foggy set brands has little value and it may prove beneficial to rename the foggy set as the unprocessed set.

Methodological Issues: Three suggestions can be briefly made for methodological improvement in data collection.

Only brands in the evoked, rejected and foggy set were directly measured. Identification of the hold set was indirect, being inferred as brands not mentioned in the other three sets. Future research should directly measure the hold set, using the same card sort techniques as in the Laroche study. The question should read something like:

"Of those brands which you've heard of, and have some knowledge about, are there any which you don't know whether you would buy or not."

The foggy set would then become the "left-over" brands. If similar results were attained from the two studies then this would greatly improve confidence that people actually do think in the four postulated categories. Finally, as a further validation the conceptual statements for all four sets should be shown to consumers as the basis for the card sort technique.

It is recommended that more sensitive measures be taken to determine the cognitive structure towards brands in each set. For example, the use of free elicitation techniques (e.g., Olson & Muderrisoglu, 1979) overcomes many of the demand characteristics endemic in pen and paper questionnaires, especially for brands towards which Little knowledge is held. Particularly sensitive data can be recovered on informational and attitudinal components of brand structure. Although free elicitations are substantially more time consuming to collect than structured questionnaires, this appears an acceptable trade-off at this embryonic stage of model development.

Moreover, the resulting protocol data allows observation of a number of cognitive traits (e.g., cognitive complexity, depth of processing) that may help understand the rules employed in allocating brands to various sets.

Greater attention should be given to individual difference analysis. It is hardly surprising that trial proved a significant predictor variable in the multiple discriminant analysis. Foggy set brands will not have experienced trial nor possibly will have many hold set brands. Moreover, trial will have a marked impact on such criterion variables as knowledge, attitude and confidence.

The importance of considering product specific ego-involvement has already been mentioned. Cognitive complexity, the subject of the remaining papers in this session, is also worth exploring. There is some evidence to suggest that greater degrees of product familiarity are associated with more dimensional knowledge structure (Conover, 1981). Further, product familiarity is thought to lead to a more systematic organization of cognitive structure (Hayes-Roth's notion of "unitization"5. Could these postulates be employed in better understanding brand allocation to the various sets?

Application Issues: What are the implications for marketing strategy of categorizing brands into sets? Despite the title of this session referring to "empirical issues" none of the authors have explored the potential application of their findings. Given that the set position-of a brand can be identified it becomes important to devise marketing programs that will best meet brand objectives. Examples may be:

TABLE

This should allow for some interesting experimentation that measures set size and cognitive structure both before and after exposure to relevant marketing stimuli.

Richins and Bloch Paper

As with the preceding paper, Richins and Bloch are concerned with identifying and measuring differences in cognitive structure towards brands in a product class. In particular, they are interested in the concept of cognitive complexity and find support for their hypothesis that degree of complexity is directly related to strength of brand preference. Unfortunately, once again a number of conceptual and methodological issues clout interpretation and require examination.

Conceptual Issues: Richins and Bloch conceptualize cognitive structure and cognitive complexity as synonymous terms. This is an erroneous assumption. Cognitive structure can be thought of as the content of cognitive space and the interrelationship between these cognitive components. Cognitive complexity is one important measure of cognitive structure characteristics. As Scott (1969) suggests it embraces two types of cognitive activity. The first is dimensionality which is measured by the number of salient concepts stored in memory and used to differentiate among concepts. The second is articulation, the number of category representations or levels of each salient dimension in memory. Thus, people with both high dimensionality and articulation evidence "cognitive complexity" in certain tasks. It should be noted that there are other properties of structure such as levels of abstraction and degree of structure organization or maturity.

However, a major problem with cognitive complexity work to date is lack of consensus on both conceptual and operational definitions of the term. There 45 general agreement that it is a multidimensional concept but problems arise in identifying different aspects of it. Scott, as mentioned above, equates complexity with the extent (number of attributes or dimensions-in a cognitive domain) and richness (degree of articulations of cognitive structure formed toward the topic of interest. Bieri (1966), on the other hand, is concerned with the perceived evaluative similarity of items (often referred to as cognitive differentiation), regardless of structural dimensionality and articulation. This has led Metcalfe (1974) to observe that...cognitive differentiation is a measure only of how much subject's constructs distinguish between elements, whereas cognitive complexity additionally reflects the hierarchical arrangement of the constructs" (in Fransella and Bannister, 1977, p. 62).

Conceptual confusion has spilled over into measures of cognitive complexity. Little convergent validity has been found for the various complexity measures (Kuusinen and Nystedt, 1975), due mainly to lack of a clear theoretical definition of the nature of complexity. The most widely used measurement approaches are based on grid procedures developed from the Role Construct Repertory Test. It is an adaptation of the Reptest that Richins and Block use with consequent serious interpretational difficulties.

Methodological Issues:

Richins and Bloch's first hypothesis tests for differences in dimensions used by shoppers with weak and strong brand preferences. They use Bieri's approach of summing the number of identical construct ratings assigned to each brand across all brands. As a result they claim that "shoppers with stronger brand preferences use more dimensions in evaluating products than consumers whose preferences are less emphatic." But is this the case? They have, in fact, created an index of differentiation. That is, certain subjects are better able to affectively differentiate between a given number of dimensions. This is because we have a situation in which the cognitive structure is already created for the subject. It comprises three brands, and six attributes, each assessed on a seven point semantic differential scale. We can only therefore search for differences in pre-determined structure.

There also appear difficulties with Richins and Bloch's measures of articulation or discrimination. For each attribute, a respondent's rating of each brand was compared with his ratings of the two other brands. The absolute values of the difference scores resulting from these comparisons were summed and considered a measure of discrimination. But again is this so? Recall that discrimination relates to the number of distinctions made along dimensions in use. The provision of seven-point semantic differential scales certainly does not allow subjects to articulate their own dimensional levels Even if we could assume that points on the semantic-differential scale represent subject articulation then measuring the difference between scores on the scale does not measure discrimination. For example, a subject may score brand A as 1 and brand B as 5 on an attribute scale. Another subject may score the same brands as 2 and 3 on the same scale. Difference scores are 4 and 1 for each subject but the level of articulation is the same. That is, both subjects have used two levels of the specific attribute. Richins and Bloch's differences measure has some similarities with the index of differentiation; it should not be confused with measures of articulation or discrimination.

Durand and Lambert suggest in their excellent paper (comment follows) that "conceptual and methodological questions inherent in buyer behavior studies of cognitive complexity" be resolved in order to "lessen the likelihood that a number of studies reporting inconsistent and contradictory findings will be conducted before potential conceptual and methodological artifacts become evident." In addition to the need for conceptual clarity noted above, I'd like to recommend that Kelly's repertory grid (or other elicitation techniques) be used to identify naturally occurring cognitive structure toward relevant topics. This will avoid studying complexity under artificially created structural premises.

Application Issues: It seems implicit in the calculation of dimensionality,-differentiation, and articulation scores that consumers process information in an aggregate, compensatory form. That is, all information contained in topic-specific cognitive structure is used. This need not be so, especially with low involvement products. Let's take a simplified example. Scenario 1 shows reptest scores for 3 brands prior to brand A advertising. Scenario 2 shows the same reptest scores following brand A advertising that emphasizes the benefits associated with attribute 1.

TABLE

The post-communication purchase of A suggests that attribute 1, because it now provides a determinant attribute, has become the catalyst for action in a noncompensatory processing system. Note that the overall cognitive complexity score has not changed and therefore, in this situation, provides little diagnostic value. It would be interesting to try and understand why and when cognitive complexity is an important concept. In part this requires more individual difference analysis to provide explanatory variables (e.g., ego-involvement; risk perception; processed set categorization).

Durand and Lambert Paper

Durand and Lambert fulfill a much needed function in identifying deficiencies in current methods of interpreting grid and free elicitation procedures for cognitive complexity measures. Although their focus is on complexity, many of their points are relevant to a broad range of memory structure assessment issues. Certainly the authors of the preceding papers would have benefitted substantially in their methodology development from a prior reading of the Durand and Lambert paper. Main points include:

Salience and number of attributes may not be invariant across subjects

The salience and number of brands considered may not be the same across subjects

The number of salient attributes considered per brand may not be invariant for each subject

It is argued that as humans are so idiosyncratic measures of cognitive structure should allow variance across subjects. This requires unstructured techniques in which individual differences in subjects can be observed.

There are, of course, problems in employing such open ended methods as repertory grids and elicitation techniques. As Durand and Lambert indicate it may prove difficult to select the correct probe in order to activate the topic relevant memory schema. This may be overcome by using multiple probes (e.g., using various product use/choice situations; employing different levels of abstraction) although the authors speculate that this may result in elicitation from multiple cognitive domains. This may prove less troublesome than first thought as closely linked probes may reveal similar structures that reinforce the belief that topic relevant schema has been identified. For example, some unpublished exploratory work of my own with probes of different levels of abstraction showed remarkable consistency in eliciting certain attributes. Such attributes may be superordinate in a cognitive core that is applicable to a range of product related domains. Clearly, such a notion requires considerable further testing but may help in answering a further question posed by Durand and Lambert..; how to identify salient attributes from single or multiple probes. Attributes that appear repeatedly in multiple, linked probes can be considered salient. In addition, attributes that emerge only in a single domain may be identified by their order of elicitation, with early elicitations being more salient than later ones (Fishbein, 1967).

The use of grid and protocol procedures is very time consuming and constrains research to small sample sizes. This is defensible with such exploratory conceptual work as the study of brand categorization and cognitive complexity but becomes of concern in more general research (e.g., attitude formation and change). It may prove possible to classify respondents into groups that exhibit considerable homogeneity of cognitive structure. Examples may be level of ego-involvement, processed set membership, and purchasing pattern profiles. Durand and Lambert offer empirical evidence that constructs vary across objects from the same domain. Identification of the above classificatory groups may provide control of such variance.

Durand and Lambert raise the point that the number of objects (brands) as well as the number of constructs must be considered in measuring cognitive complexity. The problem here is that complexity is judged entirely on the basis of dimensionality. Limited numbers of objects and/or constructs may mask substantial levels of differentiation and articulation. Moreover, I am not sure that object and construct similarity should be treated similarly. Maturing of a cognitive structure (especially for a high involvement product) is likely to result in substantial numbers of elicited constructs. Here dimensionality may be a good surrogate for cognitive complexity. This may not be the case with object elicitation. High involvement situations may lead to fewer (a smaller latitude of acceptance) rather than more brands being considered in a choice situation. This suggests an inverse relationship between object and construct complexity. Durand and Lambert show evidence of little association between object and construct dimensions in the aggregate but controlling the analysis for level of involvement may support the "inverse" Postulate.

Although the Durand and Lambert paper does not offer any clear recommendations for clarifying the complexity measurement problem, it should nevertheless be compulsory reading for researchers considering work in the cognitive complexity and other cognitive structure areas. It strongly underlines the need for more careful consideration of conceptual and assessment issues than has been evidenced to date

REFERENCES

Bieri, James (1966), "Cognitive Complexity and Personality Development" in Experience, Structure and Adaptability (ed.) O. J. Harvey, NY: Springler.

Conover, Jerry N. (1982), "Familiarity and the Structure of Product Knowledge," in Advances in Consumer Research: Vol. 9 (ed.) Andrew Mitchell, St. Louis: Association for Consumer Research. 494-498.

Fishbein, Martin (1967), "A Consideration of Beliefs, and their Role in Attitude Measurement," in Readings in Attitude Theory and Measurement (ed.) M. Fishbein, NY: John Wiley.

Fransella, Fay and Bannister,Don (1977), A Manual for Repertory Grid Technique, London: Academic Press.

Hayes-Roth, Barbara (1977), "Evolution of Cognitive Structures and Processes," Psychological Review, 84, 260-78.

Krugman, Herbert E. (1977), "Memory Without Recall, Exposure Without Perception;" Journal of Advertising Re- search, 17 (August), 7-12.

Kuusinen, J., and Nystedt, L. (1975), "The Convergent Validity of Four Indices of Cognitive Complexity in Person Perception," Scandinavian Journal of Psychology, 16, 131-136.

Metcalfe, R. J. A. (1974), "Own vs. Provided Constructs in a Reptest Measure of Cognitive Complexity," Psychological Reports, 35, 1305-1306.

Olson, Jerry C., and Muderrisoglu, Aydin (1979)," The Stability of Responses Obtained by Free Elicitation: Implications for Measuring Attribute Salience and Memory Structure," in Advances in Consumer Research: Vol. 6 (ed.) William L. Wilkie, Ann Arbor, MI: Association for Consumer Research, 269-275.

Scott, William A. (1969), "Structure of Natural Cognitions," Journal of Personality and Social Psychology, 12, 261-278.

Sherif, Caroline W., Sherif, Muzafer and Nebergall, Robert E. (1965), Attitude and Attitude Change: The Social Judgment Involvement Approach, Philadelphia: Saunders.

Toy, Daniel R. (1982), "Monitoring Communication Effects: A Cognitive Structure/Cognitive Response Approach," Journal of Consumer Research, 9 (1), 66-76.

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