Low Involvement Versus High Involvement Cognitive Structures

John L. Lastovicka, Temple University
David M. Gardner, University of Illinois
ABSTRACT - Multidimensional scaling of compact car perceptions is used to examine differences in cognitive structure between those highly involved with compact cars and those who are not. The empirical results, in line with prior theory, suggest a less differentiated and integrative structure for the low involved.
[ to cite ]:
John L. Lastovicka and David M. Gardner (1978) ,"Low Involvement Versus High Involvement Cognitive Structures", in NA - Advances in Consumer Research Volume 05, eds. Kent Hunt, Ann Abor, MI : Association for Consumer Research, Pages: 87-92.

Advances in Consumer Research Volume 5, 1978      Pages 87-92


John L. Lastovicka, Temple University

David M. Gardner, University of Illinois

[Preparation of this paper was aided by University of Illinois participation in the General Motors Intercollegiate Marketing Competition. The authors acknowledge with thanks the assistance of graduate students Patrick Merrill, Bruce Newman, Gary Pope, and Kwok-Cheung Wong for use of their data.]


Multidimensional scaling of compact car perceptions is used to examine differences in cognitive structure between those highly involved with compact cars and those who are not. The empirical results, in line with prior theory, suggest a less differentiated and integrative structure for the low involved.


Krugman's (1965) low involvement learning model has recently received renewed attention by consumer behavior researchers (Maloney, 1977; Banks and Hart, 1977). Krugman contended that television advertising is a special low involvement communication situation in which receiver responses are akin to the passive learning of nonsense syllables. He further suggested that the repetition of advertising resulted in a replacement of old brand perceptions with a new set of beliefs. This new cognitive structure was said to guide brand choice behavior without changing attitude (affect) first.

Krugman's original research stimulated consumer behavior researchers to conceptualize the advertising process in a situation-specific "micro theoretical" manner. For example, Ray et al. (1973) present alternative hierarchies of effect for different levels of involvement. One hierarchy, the standard COGNITIVE + AFFECTIVE + CONATIVE order, is labeled the learning hierarchy, and is seen as most appropriate for high involvement decisions which typically deal with high priced, high risk products. Another hierarchy, a COGNITIVE + CONATIVE + AFFECTIVE order, a low involvement hierarchy, is seen as being appropriate for much repetitive brand choice behavior of inexpensive, low risk products. Figure 1 presents the two hierarchies.



Despite the extensive research on low involvement consumer behavior (Maloney, 1977; Banks and Hart, 1977; Bowen and Chafee, 1974; Chafee and McLeod, 1973; Hupfer and Gardner, 1971; Preston, 1970; Robertson, 1976; Rothschild and Ray, 1974; and Ward, 1975), very little is known about the details of the differences between the cognitive structures in the two alternative hierarchies of effects.

One general assumption in this literature is that a low involvement cognitive structure is much less complex than a high involvement cognitive structure. Based on this assumption, it has been suggested that those advertising low involvement products discuss fewer product attributes in their advertising copy (Rothschild, 1977; Lastovicka and Gardner, 1977). Yet this global hypothesis, of less complex cognitive structure in low involvement, has not been specifically put to empirical test. The intent of this paper is to test this global hypothesis.


One approach of demonstrating the effects of involvement on cognitive structure is to compare and contrast groups of respondents who differ in their level of involvement with a given product class. In this study subjects highly involved with compact cars were compared to subjects who had a low level of involvement with compact cars. Although a correlational approach necessarily creates ambiguous statements, such an approach provides an opportunity to richly demonstrate the powerful effects of involvement on cognitive complexity.

Schroder, Driver and Streufert (1967) discuss several aspects of cognitive complexity in information processing. Two important aspects are "differentiation" and "integration." Broadly defined, differentiation refers to dimensionality or the number of dimensions used by an individual in processing information; and integration refers to the extent dimensions are interrelated or used simultaneously. The influence of involvement on both differentiation and integration were tested with these hypotheses:

H1: A unidimensional model of car model similarities will produce a better representation of car model perceptions for low involved consumers than for high involved consumers. More complex multi-dimensional representation of brand perceptions will be better models of the cognitive structure of the high involved, but not necessarily the low involved consumers.

H2: For a given dimensionality, highly involved consumers will tend to be more integrative and rely simultaneously on several perceptual dimensions. Low involved consumers will rely primarily on a lesser number of perceptual dimensions.


One hundred twenty-seven respondents from an introductory marketing course completed a questionnaire dealing with low price, medium price and luxury domestic compact cars. Responses were collected on:

1. Twenty-two lifestyle items. A 1-5 strongly disagree --strongly agree scale was used.

2. Forty-five paired comparison dissimilarity items for ten different compact cars. A 1-9 extremely similar --extremely dissimilar scale was used to measure perceptions of these ten compact car models: Pontiac Ventura, AMC Pacer Ford Maverick, AMC Hornet, Mercury Comet, Oldsmobile Omega, Buick Skylark, Ford Granada, Mercury Monarch, Chevrolet Nova. These ten cars include the "look-alike" parity models of Granada-Monarch, Nova-Omega-Skylark-Ventura, and Maverick-Comet.

3. In addition to the forty-five paired comparisons for the ten stimuli, a repeated set of ten paired comparison items were completed by each subject. Test- retest reliability on the paired comparison task could then be measured for each subject.

4. A battery of familiarity questions. Subjects were asked how familiar they were with each of the ten car models on a 1-7 not at all familiar--very familiar scale.

5. A battery of demographic questions including age, sex and perceived social class.

6. A battery of ten questions in which respondents were asked to sort each of the car models into different categories of acceptability. The question posed was:

"How acceptable to you are each of the following car models?"


This last battery often questions is based on the Sherif and Sherif (1967)Own Categories Procedure measurement of involvement. The Sherif method starts with a list of items relevant to an attitude. This may be objects, pictures, or verbal statements about some topic or issue. In the current research the items are ten brand names of compact cars.

In the Own Categories Procedure respondents are asked to sort items into categories of acceptability. Considerable research (Sherif and Howard, 1961; Sherif, Sherif and Nebergall, 1965) confirms that the items that an individual accepts, rejects and towards which he is neutral or noncommittal varies systematically with his personal involvement. Specifically the Sherifs found:

Proportional to his lack of involvement, the number of positions the individual accepts and rejects become approximately equal and his latitude of noncommitment increases. This means that highly involved persons have a much broader latitude of rejection than persons less concerned, and that they remain noncommittal toward fewer positions, even when not required to evaluate all of them. (Sherif and Sherif, 1967, p. 191, italics added.)

Typically, an involvement index is created in which the ratio of the number of items rejected to the number of items accepted, adjusted by the number of neutral items, is computed. With the current data such a measure was undefinable for many respondents for whom the number of items accepted and/or rejected was zero.

However, an involvement index which relied on the number of items which subjects were neutral or noncommittal about was operable with the current data. Depending on their degree of noncommitment or neutrality, respondents were rated in involvement with the following index:


where Xij is the ith individual's response to the jth of the ten acceptability questions and 4 is the neutral point on the 1-7 scale. Following Sherif, a high involvement index number computed from (1) indicates a small degree of noncommittal and a high level of involvement. A low index number represents a large latitude of non-committal and a corresponding low level of involvement.

By recoding the responses to the acceptability questions for the ten makes and models of compact cars using (1), an involvement index with a range of 0-30 was computed for each respondent.


A pair of multidimensional scaling (MDS) models, Young's TORSCA (1968) and Carroll and Chang's INDSCAL (1970), were used to uncover the underlying dimensions used by the respondents in evaluating the similarities between the ten compact cars.

Information processing researchers Schroder, Driver and Streufert (1967) cite aggregate level MDS methods as a good measure of differentiation or the number of dimensions used by individuals in processing information. Individual differences MDS models such as INDSCAL, which indicate to what degree respondents use the underlying dimensions, offers direct measurement of integration or the degree to which the dimensions are used simultaneously and are interrelated. Also, MDS researchers Shepard, Romney and Nerlove (1972) show MDS useful in recovering degree of cognitive complexity. MDS, therefore, seems well suited for tapping cognitive complexity.

Preprocessing the Data

Since each respondent provided two sets of judgments for ten of the possible forty-five paired comparisons, a reliability index was computed for each respondent with Spearman's rank order correlation. Those 79 respondents of the 127 whose reliabilities were above the critical level of .54 were retained for both the TORSCA and INDSCAL analyses. The average Spearman correlation for this retained group of 79 was .77.

Once involvement indices were computed, the retained sample of 79 was divided into three, approximately equal sized, involvement groups: low, medium, and high. The low group consisted of thirty respondents whose involvement indices were 17 or less. Twenty respondents, with indices between 18 and 21, were the medium group; and another twenty-nine respondents, with indices of 22 and over, were the high group. The mean involvement indices for the low, medium and high groups were 14.28, 19.90 and 24.70, respectively.

Hypothesis One

Average, standardized similarity matrices for the paired comparison data were computed for the high and low involvement groups. The result was two 10 x 10 average matrices of car model dissimilarities. The high and low involvement dissimilarity matrices were scaled at an aggregate level using the TORSCA algorithm. In each case, solutions were obtained in one through four dimensions. Figure 2 shows the relationship between the Kruscal's stress statistic (the goodness of fit between the original dissimilarities data and the n-dimensional MDS configuration) and the dimensionality of the solution. Examination of Figure 2 should be done in the same spirit as the "root staring" procedure in factor analysis in which eigenvalues are plotted versus their number. This comment is made because established statistical tests for comparing stress levels are not available. Just as in "root staring," then, subjective judgment must play a large role in the current analysis.

The stress measures for the high and low involvement scalings shown in Figure 2 are in support of the first hypothesis. A simple unidimensional model of dissimilarities is a much better representation of cognitive structures for the low involvement group than the high involvement group. This simplest scaling indicates that a more complex model is needed for the high involvement group. For the two and three dimensional scaling solutions, differences in stress are not great. Yet these two and three dimensional solutions offer slightly better fit for the high involvement group. Finally, for the four dimensional scalings, the most complex models built in this analysis, the results are as theory predicts. Such further complication does not provide a better representation of cognitive structure for the low involvement group, yet better fit is marginally obtained for the high involvement group.

Hypothesis Two

The INDSCAL model was used to examine cognitive complexity in terms of integration. This model assumes a common stimulus space, with differential weighting of the dimensions of this common space for each respondent. A respondent's position in the INDSCAL person space represents the salience he assigns to each of the dimensions in the common-stimulus space. Thus, the weights can be used to estimate an individual stimulus space for each person. The individual stimulus configurations are based on the common stimulus space, but are differentially "stretched" in accord with the square roots of the respondent's own weights.

The two dimensional group stimulus space in Figure 3 was interpreted with the classic LUXURY and SPORTY dimension.

The person space, showing each of the 79 respondent's saliences for the two dimensions of the common stimulus space, is shown in Figure 4. Using visual clustering the respondents were grouped into three clusters representing three integrative styles. A "left" cluster of 29 respondents who made disproportionate use of the SPORTINESS dimension was identified. A "center" cluster of integrative respondents who made roughly equal use of both of the common stimulus space dimensions has 24 respondents. The third cluster, a "right" cluster of 26 respondents, contains those who made disproportionate use of the LUXURY dimension.

The next step of the analysis was to test the influence of involvement on integrative complexity. In a two group discriminant analysis, involvement along with lifestyle, familiarity and demographic measures were used to predict membership in the integrative "center" cluster or the nonintegrative "left" and "right" cluster. Though the prime interest is in detecting the influence of involvement on integration, it was felt the predictive power of this variable should be compared with others. Examination of normalized discriminant function weights allows easy comparative assessment of the predictor variables.

Rather than use the entire battery of the 32 lifestyle and familiarity measures as predictors in the discriminant analyses, two independent principal axes factor analyses were used on these measures to create a set of six parsimonious underlying factors.

The first factor analysis conducted on the 22 lifestyle items revealed four common factors. After a varimax rotation the four factors were easily interpretable. The first factor, an automotive knowledge factor, had high positive loadings on items such as: "I know a lot about cars;" and "People often come to me for information.'' The second factor, a functional automotive preference factor, had high positive loadings on items like: "I like to drive a car that will hold up in an accident;" and high negative loadings on items such as: "You can tell a lot about a person from the model of car he drives." The third factor, an image preference factor, loaded highly on items like: "I think cars are a mark of status;" and "I like to have the best looking car on the road." The fourth lifestyle factor, an automotive thrill-seeking factor, loaded highly on items such as: "I like to drive fast;" and "I like to listen to music while driving." Factor scores were then estimated on these four lifestyle factors.

The second factor analysis was conducted on the familiarity ratings for the ten compact cars. Two factors were extracted in this analysis; the first was an overall familiarity factor and the second was familiarity with certain models of parity cars. Familiarity with the differences between the "look-alike" parity models of Omega, Nova, Skylark and Ventura, for example, was independent of overall familiarity. Factor scores were also estimated for these two familiarity factors.

Using lifestyle factor score estimates, familiarity factor score estimates, the involvement index, age, social class and a dummy variable for sex, a significant discriminant function was found to differentiate between the integrative and non-integrative groups. The obtained value of Rao's F-ratio approximation, F10,68= 1.96, is just significant at the .05 level. Using the discriminant function weights shown in Table 1, 70.8 percent of the integrative group and 72.7 percent of the non-integrative group could be correctly classified. Predictive validity was further tested using U-method pseudo jack-knife classification (Crask and Darden, 1977). With the U-method a classification function was computed for each of the 79 cases with that case omitted from the computations. Since each of these functions was used to classify the left-out case, a less biased classification occurs.

Based on the U-method jack-knifed classification functions, 50 percent of the integrative group and 67.3 percent of the non-integrative group were correctly classified.

Examination of the relative size of each standardized discriminant weight in Table 1 gives an indication of the relative importance of the ten variables in differentiating between the two groups. As hypothesized, involvement is found to be a good explanatory variable for cognitive integration. The weight for the involvement index, .806, is almost twice the size of the next largest weight. Examination of the cluster means on the discriminant function shows that the "center" cluster,









the integrative group which used both the INDSCAL stimulus space dimensions equally, has the highest score. It seems then that equal use of both dimensions occur primarily under high involvement. Low involvement cognitive structures seem less integrative and rely primarily on one dimension.


The general findings of this study are in line with prior theory. Low involvement cognitive structures do seem to be simpler than high involvement structures in at least two ways. First, low involvement structures seem less differentiated as they can be represented adequately with fewer dimensions than high involvement structures. Second, low involvement structures tend to be less integrative. In the current data, a two space map of a low involved individual's compact car perceptions is typically most reliant on one dimension. The simultaneous, integrative approach is apparently not worth the effort on the part of the low involved consumer.

Despite the support for the global hypothesis, several points must be kept in mind.

First, the differences found in cognitive structure can only be said to be potentially due to involvement. Crucial differences besides involvement may be responsible for the observed phenomena. Further research in this area should include both use of experimentation and multiple measurement approaches to involvement.

Second, cognitive structure, the dependent measure, was measured only along two dimensions: differentiation and integration. Cognitive differences in terms of discrimination, for example, were not examined.

Third, differences have been examined between individuals for a given product. Though such an approach is useful for market segmentation, the real thrust of low involvement consumer behavior concerns differences between products.

In conclusion, this study should be seen as a very basic exploration into the nature of the differences between high and low involvement cognitive structures. The paper should help to underscore that there is a difference between the low involved consumer and the high involved consumer that most researchers have implicitly assumed.


Banks, S. and E. W. Hart, "Advertising and Promotional Methods," in Robert Ferber, ed., A Synthesis of Selected Aspects of Consumer Behavior (Washington, D.C.: National Science Foundation, 1977, in press).

Bowen, L. and S. H. Chafee, "Product Involvement and Pertinent Advertising Appeals," Journalism Quarterly, 51 (Winter, 1974), 613-21.

Caroll, J. D. and J. J. Chang, "Analysis of Individual Differences in Multidimensional Scaling Via an N-Way Generalization of Eckart-Young Decomposition," Psychometrika, 35 (1970), 283-319.

Chafee, S. H. and J. M. McLeod, "Consumer Decisions and Information Use," in S. Ward and T. S. Robertson, eds., Consumer Behavior: Theoretical Sources (New York: Prentice-Hall, 1973).

Crask, M. R. and W. D. Perreault, Jr., "Validation of Discriminant Analysis in Marketing Research," Journal of Marketing, 14 (February, 1977), 60-8.

Hupfer, N. and D. Gardner, "Differential Involvement With Products and Issues: An Exploratory Study," Proceedings of the Association for Consumer Research, Second Conference, 1971.

Krugman, H. E., "The Impact of Television Advertising: Learning Without Involvement," Public Opinion Quarterly, 29 (Fall, 1965), 349-56.

Lastovicka, J. L. and D. M. Gardner, "Components of Involvement," in John C. Maloney, ed., Attitude Research Plays for High Stakes (Chicago: American Marketing Association, 1977, in press).

Maloney, John C., ed. Attitude Research Plays for High Stakes (Chicago: American Marketing Association, 1977, in press). (Proceedings of the American Marketing Association's Eighth Annual Attitude Research Conference. All papers from two sessions: "Must Consumers Think Before They Act?--A Review of Low Involvement Theory" and "Low Involvement Versus High Involvement Situations--Some Guidelines for Action," deal with the involvement issue.)

Preston, L. L., "A Reinterpretation of the Meaning of Involvement in Krugman's Models of Advertising Communication,'' Journalism Quarterly, 47 (Summer, 1970), 287-95.

Ray, M. L., A. G. Sawyer, M. L. Rothschild, R. M. Heeler, E. C. Strong, and J. B. Reed, "Marketing Communication and the Hierarchy of Effects," in P. Clarke,' ed., New Models for Mass Communication Research (Beverly Hills, Cal.: Sage, 1973).

Raymond, C. Advertising Research: The State of the Art (New York: Association of National Advertisers, 1976).

Robertson, Thomas S., "Low Commitment Consumer Behavior," Journal of Advertising Research, 16 (April, 1976), 19-27.

Rothschild, M. L., "Advertising Strategies for High Involvement and Low Involvement Situations," in John C. Maloney, ed., Attitude Research Plays for High Stakes (Chicago: American Marketing Association, 1977, in press).

Rothschild, M. L., and M. L. Ray. "Involvement and Political Advertising Effect: An Exploratory Experiment," Communication Research, 1 (July, 1974), 264-84.

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

Shepard, R. N., A. K. Romney and S. B. Nerlove, eds., Multidimensional Scaling, Volume I. (New York: Seminar Press, 1972).

Sherif, M. and C. I. Hovland. Social Judgment: Assimilation and Contrast Effects in Communication and Attitude Change (New Haven: Yale University Press, 1953).

Sherif, M. and C. W. Sherif, "The Own Categories Procedure in Attitude Research," in Martin Fishbein, ed., Readings in Attitude Theory and Measurement (New York: Wiley, 1967).

Sherif, W. C., M. Sherif and R. E. Nebergall. Attitude and Attitude Change: The Social Judgment Involvement Approach (New Haven: Yale University Press, 1965).

Ward, Scott, "Marketing Strategies for 'Low Ego-Intensity/Low Risk' Products," Intercollegiate Case Clearing House, 1975.

Young, F. W., "TORSCA, An IBM Program for Nonmetric Multidimensional Scaling," Journal of Marketing Research, 5 (August, 1968), 319-21.