The Moderating Effects of Cognitive Complexity and Prior Product Familiarity on the Predictive Ability of Selected Multi-Attribute Choice Models For Three Consumer Products

Chin Tiong Tan, University of Singapore
Ira J. Dolich, University of Nebraska-Lincoln
ABSTRACT - Respondents were classified according to their levels of cognitive complexity and product familiarity with three consumer products: automobiles, apartments, and toilet soaps. Using repeated measures and the reperatory grid technique with ten brands and ten dimensions for each product class, five multi-attribute models ere tested for predictive accuracy of rank order brand preferences. Three sets of 2 x 2 x 5 ANOVA studies showed significant systematic relationships between type of model and level of cognitive complexity and significant differences attributed to models across all products and to cognitive complexity for two of the three products.
[ to cite ]:
Chin Tiong Tan and Ira J. Dolich (1981) ,"The Moderating Effects of Cognitive Complexity and Prior Product Familiarity on the Predictive Ability of Selected Multi-Attribute Choice Models For Three Consumer Products", in NA - Advances in Consumer Research Volume 08, eds. Kent B. Monroe, Ann Abor, MI : Association for Consumer Research, Pages: 140-144.

Advances in Consumer Research Volume 8, 1981      Pages 140-144


Chin Tiong Tan, University of Singapore

Ira J. Dolich, University of Nebraska-Lincoln


Respondents were classified according to their levels of cognitive complexity and product familiarity with three consumer products: automobiles, apartments, and toilet soaps. Using repeated measures and the reperatory grid technique with ten brands and ten dimensions for each product class, five multi-attribute models ere tested for predictive accuracy of rank order brand preferences. Three sets of 2 x 2 x 5 ANOVA studies showed significant systematic relationships between type of model and level of cognitive complexity and significant differences attributed to models across all products and to cognitive complexity for two of the three products.


Multi-attribute choice models have been used in marketing to understand a host of buying related behavior such as attitude structure and judgment. Considerable effort has been devoted to using these models to predict brand preferences or choices.

Several exploratory studies (Park and Sheth 1975, Park 1976, and Menasco 1976) have attempted to identify selected individual characteristics or situational variables that may influence model usage and to investigate how these characteristics and variables affect judgment. Such a research direction should be encouraged, as it enhances an in-depth understanding of the interaction between brand choice, individual characteristics and situational differences.

The notion that a person acquires product familiarity while progressing through different stages of the learning curve is well documented in the consumer behavior literature (Howard and Sheth 1969). Sheth and Raju (1973) further proposed that different choice models, some more complicated than others, may be used at these different learning stages. In other words, consumers with different levels of familiarity with a product may rely on different types of brand choice models. This issue of product familiarity and use of choice model is a major consideration of the present study.

Marketers have made numerous attempts to link individual characteristics such as personality variables to buying behavior. In this study, the personality variable of interest is the complexity-simplicity of one's cognitive structure. Treating one's cognitive style as a personality structure is common in psychology (Schroder 1971), and its relationship with processing behavior is well documented in the works of Schroder, Driver and Streufert (1967). In that tradition, this personality characteristic is used as a moderating variable to examine its effect on the prediction of choice models.

This study centers on three specific consumer behavior issues: personality traits, choice models and familiarity with products. The major research objectives are to test propositions about the moderating effects of cognitive complexity and product familiarity on brand preferences derived from multi-attribute choice models.

Cognitive Complexity

Cognitive structure is viewed as a system of attributes or dimensions which form the basis of one's perceptual and judgmental processes, Kelly (1955), Bieri et al, (1966), and Schroder et al. (1967) showed that persons differ in cognitive structure and that such differences result in response variations. The relationships and interdependencies of dimensions within these systems bring about certain properties or characteristics that are considered unique to a person. Cognitive complexity is one such characteristic.

Cognitive complexity has been defined differently by psychologists (refer to Bieri (1971) for a review). The present study adopts the definition of Bieri (1955)-namely, the extent to which a person uses dimensions in a differentiated manner to construe objects. A complex person evaluates objects in a more multi-dimensional way and has a more differentiated sat of dimensions available than does a simple person.

Psychologists have found cognitive complexity to mediate a vide range of behaviors such as attitude change (Lundy and Berkowitz 1957), information processing (Petronkon and Perlin 1970), salience of dimensions (Mueller 1974), and stereotyping (Keening and King 1962). Marketers have recently investigated the construct. Among them, Park and Sheth (1975) found mixed results with the effect of cognitive complexity on choice behavior. The present study re-examines how cognitive complexity affects accuracy of prediction for several judgment models.

Product Familiarity

The extent to which a person is familiar with a product class depends largely on factors such as prior knowledge, usage, and purchase of brands in the product class. Marketers conceptualize one's familiarity to be a multidimensional concept comprised of elements such as awareness, knowledge, and actual usage of brands from the product class. Thus far, empirical studies on prior familiarity with products have been limited. Park (1976) found it to influence types of judgmental rules. However, another study showed familiarity to have little effect on brand preferences (Monroe 1976). The present study investigates how one's prior product familiarity may affect multi-attribute choice model efficiencies in predicting brand preferences.

Choice Models

Most of the marketing studies utilizing multi-attribute choice models disregarded individual differences and situational variations which might affect chair usage. It is only recently that marketing researchers have begun to look to individual and situational differences as causes of the poor results found in some studies. Sheth and Raju (1973) have postulated that different processing rules for brand choices are used in different contexts. If such is the case, multi-attribute models, such as Fishbein's (1967) and Rosenberg's (1956), when used across different product classes for all types of consumers, would inevitably lead to weak results in some instances. The weighted linear-compensatory rule of their models cannot be assumed to hold for all buyers for all purchases.

The generalized model assumes that evaluations of a brand are based on multiple attributes that vary in importance. Thus, one's brand choice is related to his overall evaluation which is determined by aggregating individual evaluations adjusted for differences in importance.

This research effort utilizes four versions of the sum-mated model and the lexicographic model as described below:

1. Unweighted summative model I (full version) --subject's ratings on ten attributes are added.

2. Unweighted summative model II (partial version) --subject's ratings on the five most important attributes are added.

3. Weighted summative model I (full version) -- subject's ratings on ten attributes are weighted by importance values and summed.

4. Weighted summative model II (partial version) --subject's ratings on the five most important attributes are weighted again by importance ratings and summed.

5. Lexicographic model -- a sorting procedure in which the attributes are ranked according to importance, and alternative brands evaluated sequentially starting with the single most important attribute downward. If a tie arises, successive dimensions are used. The sorting process continues until a rank ordering of alternatives is completed.

This study, then, investigates hey these various choice models may provide different predictive accuracy for persons with different characteristics and familiarity with products. Specifically, the major research objectives are to test the following hypotheses:

H1: The predictive accuracy of selected choice models differs between cognitively complex and cognitively simple persons.

H2: The predictive accuracy of selected choice models differs between persons with different levels of product familiarity.


Product Selection

Three products were required to differ in importance and to have multiple brands or options that were generally known to the respondents. Automobiles, rental apartments and toilet soaps were shown to meet the above criteria using a group of respondents similar to the final sample.

Research Instruments

Respondents indicated brand knowledge by the percentage of brands checked on a list of brands provided for the purpose. A person's familiarity with a product class was measured as the proportion of brands in the product class that one knew something about.

A new research approach was adopted to operationalize the choice models. Instead of imposing a standard set of attributes (dimensions) for evaluation of brands, each respondent was asked to provide his/her own attributes. Such a procedure should partially eliminate the problem of non-salient attributes. The brands (stimuli) to be evaluated were also self-selected from the larger list of brands. Arbitrary responses to unknown brands were thereby avoided.

Kelly's Repertory grid (Kelly 1955) was used to organize attributes and then to rate brands. In essence, it is a matrix which captures a person's repertoire of self-generated dimensions used in the evaluation of objects. Table 1 shows such a grid.

Cognitive complexity is computed using the matching of dimensions procedure as proposed by Bieri and colleagues (1966). In brief, a cognitively simple person, when using the grid, would not be able to use many of the dimensions in a differentiated manner, and hence would evaluate the objects identically. The ratings in the grid are compared with one another for agreement on objects to determine the number of identical ratings. Every dimension is compared with all others. In such a manner, the entire repertoire of dimensions is examined to determine one's relative complexity. A simple person will have a high number of matches (similar to identical semantic differential profiles). A low score therefore denotes a more differentiated structure and cognitively mere complex. Table 2 summarizes the results of this procedure using the data from Table 1.

Data Collection

Respondents were upperclass undergraduate students and each was paid two dollars for participation. Ninety-six persons formed the final sample. They were arranged in small group settings and detailed instructions were presented verbally and in writing with special care taken to explain the Repertory Grid routines. A repeated measure design was used in that each subject provided evaluations for all three product classes. Product presentation in the questionnaire booklets was randomly distributed to prevent order effects. Three Reperatory Grids were included in each booklet.

The grid was a 10 X 10 matrix. Each respondent was instructed to pick the ten most familiar brands from a list of twenty-two and to write the brand names at the column headings. Starting from row one to row ten, the respondent was to list ten different attributes/dimensions he or she commonly used to evaluate brands of that product. Each dimension was then to be used with a six-point Likert type bipolar scale having values of +3, +2, +1, -1, -2, -3 to rate all ten brands on the grid. To assist in the generation of dimensions, three randomly pre-selected cells were circled on each row of the grid. The subject was told to look at the three brands and to consider a way, dimension, or characteristic for which two of them were similar and yet different from the third. Each dimension generated was written down on the side of the row and the procedure continued until ten bipolar dimensions were elicited. (See Kelly 1955, and Bieri et al. 1966 for details on the procedures).

In addition, preference rankings for the ten brands and importance rankings for the ten self-generated dimensions were collected. The sequence of tasks was identical for each product class.


Each subject's rank ordered brand choice prediction was derived for each model and correlated with the independently obtained preference ranking. The Spearman Rank Order Correlation was used as a measure of the accuracy of prediction of the models. In other words, the "goodness of fit" criterion was used to compare predictive ability of competing models. These correlation coefficients were then used as the dependent variables in the various analysis of variance studies.

For each product class, frequency distributions of the subjects' cognitive complexity scores and product familiarity indices were determined. The distributions were then dichotomized into the high (upper) and low (lower) halves for analyses. The final ANOVA model for each product class is a 2 X 2 X 5 repeated measure design with two fixed factors (2 levels of cognitive complexity and 2 levels of product familiarity) and one repeated factor (5 levels, one for each choice model).









Table 3 summarizes the main and interaction effect significance tests for each product class. Mean scores are shown in Table 4.

Product Familiarity

Product familiarity as defined herein was not a statistically significant main effect. The overall group means for the high and low familiarity subjects are rather close. However, the high familiarity group consistently shows slightly higher mean scores than the low familiarity group across all three product classes. No further individual comparison is performed.

Cognitive Complexity

It was previously proposed that the predictive accuracy of selected choice models differs between cognitively complex and cognitively simple persons. And, the ANOVA results shown in Table 3 indicate significant complexity main effects for two of the three product classes. Since familiarity with product class was nonsignificant for all three products, this main effect was eliminated for presentation in Table 4.

The significance tests on cognitive complexity can be seem more clearly from Table 4 where overall mean differences for automobiles and apartments are shown to be statistically significant at 0.01 and 0.05 levels, respectively. These individual comparisons highlight differences among the five models and explain the significant main effect shown in Table 3. In the two product classes where the complexity factor is significant, four out of five individual comparisons are significant as well. All of the comparisons for toilet soaps are insignificant.

Choice Models

All three product classes show significant main effects on choice models (Table 3). This issue has been considered by many researchers and evidence of variability in predictive accuracy of different models has been well documented (Wilkie and Pessemier 1973). In general (Table 4), the weighted summative models (3 and 4) provide higher correlations than the unweighted models, (1 and 2) the full version models (1 and 3) show higher correlations than the partial version models (2 and 4), and the summa-rive models (1 through 4) give more accurate results than the lexicographic model.


Of the various interactions, the one that shows up well is between type of model and level of complexity. Both automobiles and apartments show significant interactions. In addition, the three-way interaction is significant for the automobiles. These model-complexity interactions are presented graphically in Figure 1. As can be seen, the high and low complexity subjects do not provide similar results across the five models.


The present study found respondents differing in cognitive complexity to provide differences in accuracy of prediction for five choice models. The cognitively complex subjects appear to have consistently better predictions than the simple subjects, irrespective of the types of choice model considered. This may be due to the complex persons' more differentiated cognitions which enable then to see things in a more multi-dimensional fashion. For them, the more refined and differentiated evaluations of brands (full models) resulted in choice rankings which corresponded closely to actual preferences. Conversely, cognitively simple persons may lack the ability to see differences in brands. Brand choice predictions based on the evaluation of such dimensions may not realistically reflect actual preferences for this latter group.

For the cognitively simple subjects, the weighted summative models consistently out-perform the others. In the case of the cognitively complex subjects, the unweighted model appears to be the best predictor. In general, the differences between the full and the partial models appear to be more substantial for the cognitively complex subjects. One explanation may be that the additional attributes ere meaningless to the cognitively simple subjects due to their simplistic cognitive processes, whereas, more attributes enable the cognitively complex subjects to further differentiate and thus, predict more accurately.



Cognitive complexity as a moderating variable is found here to be significant in the two product classes of highest importance. When the product is less important and rather simple such as toilet soap, no difference in predictive accuracy is found. In other words, the cognitively simple subjects seemed to be as capable as the cognitively complex subjects in evaluating stimulus objects that are inherently simple. Of the three products, the former subjects have the highest predictive accuracy in toilet soap (rs = 0.60) which is significantly higher than the other two product classes (rs = 0.46 and 0.43 for automobiles and apartments respectively). For the latter subjects, differences among the three products are not statistically significant (rs = 0.61, 0.55 and 0.63 for automobiles, apartments and toilet soaps, respectively).

Prior familiarity with the product class was not shown to be a moderating variable in choice model predictability in this study. Persons with high and low product familiarity did not show any significant differences in the predictive accuracy of the models studied. An alternative explanation is that evaluation of brands is related more to cognitive structure than to product familiarity. Hence, a person, regardless of his/her level of brand familiarity, can still generate meaningful brand evaluations through unique cognitive capacities. Such a theoretical rationale has support. Bieri (1971) and Hall (1966) have claimed one's cognitive style to be a consistent trait manifested in different situations rather than domain specific (experience related) characteristic.


This research, although exploratory in nature, does provide a new perspective to a high interest area in marketing. Introducing moderating variables to the study of multi-attribute models and brand choice unveils significant details which are otherwise unnoticed.

This research is an attempt to refine our use of multi-attribute models. Of two variables with high potential as moderators, one -- cognitive complexity -- was found to be a significant factor. Persons of different levels of cognitive functioning were found to provide differences in predictive accuracy for five multi-attribute choice models.


Belk, R. (1975), "Situational Variables and Consumer Behavior," Journal of Consumer Research, December.

Bieri, J. (1955), "Cognitive Complexity-Simplicity and Predictive Behavior," Journal of Abnormal and Social Psychology, 51.

Bieri, J., (1971), "Cognitive Structures in Personality,'' in H. M. Schroder and P. Suedfeld, (ed.) Personality Theory and Information Processing, New York: Ronald Press.

Bieri, J., Attins, A. L., Briar, S., Leaman, B. L, Miller, H., and Tripodi, T. (1966), Clinical and Social Judgement: The Discrimination of Behavioral Information, New York: Wiley and Sons.

Fishbein, M. (1967), "Attitudes and Prediction of Behavior," in M. Fishbein (ed.), Readings in Attitude Theory and Measurement, New York: John Wiley and Sons, Inc.

Hall, M. F. (1966), The Generality of Cohesive Complexity-Simplicity, an unpublished Ph.D. Dissertation, Department of Psychology, The Vanderbilt University.

Higgins, J. C. (1959), "Cognitive Complexity and Probability Preferences," unpublished manuscript, Department of Psychology, University of Chicago.

Howard, J. and Sheth, J. (1969), The Theory of Buying Behavior, New York: Wiley and Sons, Inc.

Kasulis, J. and Zaltman, G. (1976), "Message Reception and Cognitive Complexity," in Perreault, W. (ed.), Advances in Consumer Research, Vol. IV.

Kelly, G. A. (1955), Psychology of Personal Constructs, New York: Norton Press.

Koening, F. W. and King, M. (1962), "Cognitive Simplicity and Prejudice," Social Forces, 40, March.

Koening, F. W. and King, M. (1964), "Cognitive Simplicity and Out-Group Stereotyping," Social Forces.

Lundy, R. and Berkowitz, L. (1957), "Cognitive Complexity and Assimilative Projection in Attitude Change," Journal of Abnormal and Social Psychology, 55.

Meansco, M. (1976), "A Further Exploration of the Modern Effects of Cognitive Complexity Upon Consumer Choice Behavior,'' Working Paper on the Bureau of Business and Economic Research, University of Iowa, July.

Monroe, K. (1976), "The Influence of Price Differences and Brand Familiarity on Brand Preferences," Journal of Consumer Research, June.

Mueller, W. W. (1974), "Cognitive Complexity and Salience of Dimension in Person Perception," Australian Journal of Psychology.

Nakanishi, M, and Bettman, J. (197&), "Attitude Models Revisited: An Individual Level Analysis," Journal of Consumer Research, December.

Park, W. and Sheth, J. (1975), "Impact of Prior Familiarity and Cognitive Complexity on Information Processing Rules," Communication Research, Vol. 2, No. 3, July.

Petronko, M. and Perin, T. (1970), "A Consideration of Cognitive Complexity and Primacy-Recency Effects in Impression Formation," Journal of Personality and Social Psychology, Vol. 15.

Rosenberg, M. J. (1956), "Cognitive Structure and Attitudinal Effect," Journal of Abnormal and Social Psychology, Vol. 53.

Schroder, H. M. (1971), "Conceptual Complexity and Personality Organization," in Schroder and Suedfeld, (eds.), Personality Theory and Information Processing, New York: Ronald Press.

Schroder, J., Driver, M., and Streufert, S. (1967), Human Information Processing, New York: Holt, Rinehart and Winston.

Sheth, J. and Raju, P.S. (1973), "Sequential and Cyclical Nature of Information Processing Models in Repetitive Choice Behavior," in Wards, S. and Wright, P. (eds.), Advances in Consumer Research, Vol. 1.

Stiles, G. (1974), "Determinants of Industrial Buyer's Level of Information Processing: Organizations, Situations and Individual Differences," in D. Hughes and M. Ray (eds.) Buyer/Consumer Information Processing, Chapel Hill, North Carolina: University of North Carolina Press.

Tripodi, T. and Bieri, J. (1966), "Cognitive Complexity, Perceived Conflict, and Certainty," Journal of Personality, 34.

Wilkie, W., and Pessemier, E. (1973), "Issues in Marketing's Use of Multi-Attribute Attitude Models," Journal of Marketing Research, November.

Wilson, D. T. and Tan, C. T. (1977), "Dimensional Complexity of Cognitive Structure: A Personality Trait in Decision Making," in Stolen, J. D. and Convey, J. J. (eds.), Proceedings of the American Institute of Decision Sciences.

Wright, P. (1975), "Consumer Choice Strategies: Simplifying Vs. Optimizing," Journal of Marketing Research, February.

Zajonc, R. B. (1968), "Cognitive Theories in Social Psychology," in The Handbook of Social Psychology, Vol. 1, Reading, Mass.: Addison-Wesley.