Context Effects in Product Perception

Michael E. Johnson, The University of Chicago
ABSTRACT - An experiment demonstrates the existence of context effects in consumer perceptions of product similarity. Such effects violate the measurement assumptions of familiar models of product perception such as multidimensional scaling. The observed violations are explained by a theory that represents products as sets of qualitative features.
[ to cite ]:
Michael E. Johnson (1981) ,"Context Effects in Product Perception", in NA - Advances in Consumer Research Volume 08, eds. Kent B. Monroe, Ann Abor, MI : Association for Consumer Research, Pages: 112-115.

Advances in Consumer Research Volume 8, 1981      Pages 112-115

CONTEXT EFFECTS IN PRODUCT PERCEPTION

Michael E. Johnson, The University of Chicago

[The author thanks J. Edward Russo and the Consumer Behavior Laboratory of the Graduate School of Business of the University of Chicago for their assistance in conducting the experiments and preparing the paper. Financial assistance was provided by the Red and White Foundation and the Center for Research in Marketing at the University of Chicago.]

ABSTRACT -

An experiment demonstrates the existence of context effects in consumer perceptions of product similarity. Such effects violate the measurement assumptions of familiar models of product perception such as multidimensional scaling. The observed violations are explained by a theory that represents products as sets of qualitative features.

INTRODUCTION

Through years of advertising and use occasions, Budweiser beer has come to be associated with such distinctive features as "Clydesdales", "beechwood aged", and so on. When marketers ask consumers to rate the similarities among beers, their dimensional models of similarity assume that such nondimensional features have no influence. The resulting dimensional representation or "perceptual space" of the brands is assumed to reflect only the differences along uniformly relevant dimensions, such as a beer's "lightness".

The goal of this study is to show that product perceptions are affected by unique features as well as by shared dimensions. Such a demonstration contradicts the measurement assumptions of the familiar models of product perception (multidimensional scaling, multidiscriminant analysis, and conjoint measurement). More specifically, my aim is to show that the salience of different features (and dimensions) will vary with context. Where context can be controlled, such features should influence product comparisons in predictable directions.

After outlining a psychological theory that describes similarity judgments as the result of a feature matching process, the generality of the theory is tested in a consumer products context. The results support such a feature based approach to product perception and attest to the sensitivity of perception to changes in context.

SIMILARITY JUDGMENT AS A FEATURE MATCHING PROCESS

A dimensionally based technique such as multidimensional scaling (MDS) presumes a full analogy between the cognitive concepts of similarity and dissimilarity on the one hand, and the (Euclidean) geometry of spatial proximity and distance on the other (Cunningham and Shepard 1974). Specifically, MDS postulates that perceived similarity among objects is a monotonically decreasing function of the distances between those objects represented as points in n-dimensional (metric) space. Therefore, such a method is based on both dimensional and metric assumptions.

However, Tversky (1977) has demonstrated that these assumptions may be inappropriate in describing people's perception of similarity. The metric assumptions require that similarity be symmetric. That is, the similarity of a, the subject, to b, the referent, is necessarily identical to the similarity of b to a. The metric models also assume that perceived similarities among objects should be perfectly negatively correlated with the corresponding perceived dissimilarities. In many cases, however, neither of these assumptions holds. For example, Tversky (1977) found subjects' ratings of the similarity of North Korea to Red China to be greater than the similarity of Red China to North Korea. He also showed that when two stimuli in a group appeared to have both more common and more distinctive features than any other pair of stimuli in the group, these two were often judged to be both the most similar and the most dissimilar pair in the group. For instance, the USSR and the United States received both the highest similarity and the highest dissimilarity ratings among the countries judged by Tversky's subjects. Such findings lead one to question the general applicability of both the dimensional and metric assumptions of MDS. Indeed, dimensional interpretations may be inappropriate for a large class of product categories and, if applied, will produce misleading product spaces.

Tversky offers an alternative description of how people make similarity judgments. Judging similarity is a feature matching process.

It seems more appropriate to represent faces, countries, or personalities in terms of many qualitative features than in terms of a few quantitative dimensions. The assessment of similarity between such stimuli, therefore, may be better described as a comparison of features rather than as the computation of metric distance between points (Tversky 1977, p.328.)

When faced with a similarity task, people extract and compile from remembered information a limited list of relevant features on the basis of which they perform the required task. As formally stated by Tversky, the similarity between two objects, s(a,b), where a and b are associated with feature sets A and B respectively, is:

s(a,b) = qf(A B) - af(A-B) - bf(B-A)

The similarity between two objects is a function of their common features (A B), features of a but not of b (A-B), and features of b but not of a (B-A). Similarity can be a function of just common or just distinctive features or both depending on the values of the parameters q, a, and b.

Because format can influence the value of these parameters, Tversky's theory has the power to account for diverse empirical observations. For instance, the North Korea/Red China asymmetry can be explained by a small value of B (in conjunction with Red China's having more distinctive features than North Korea). In other words, when one stimulus is the subject and the other the referent, the distinctive features of the referent do not receive as much weight in the overall similarity judgment as the distinctive features of the subject. When, in addition, one stimulus has more distinctive features than the other, an asymmetry occurs. Thus, when North Korea is the subject its features map quite well into those of Red China. But when Red China is the subject its features will not map as well into those of North Korea because of the greater number of distinctive features associated with Red China. Similarly, the difference between similarity and dissimilarity judgments can be accounted for by a greater emphasis on common features in the former case (let q" SIZE="2> >> a = b in Equation 1) and a greater emphasis on distinctive features in the latter case (let a" SIZE="2 = b >> q).

The power of the feature model to explain context effects lies in the weights given to the different sets of features. By controlling the context or format within which judgments are elicited, one should be able to strategically manipulate the relative size of these parameters and, as a result, influence product perception.

It should be noted that the empirical violations of MDS that Tversky reported may be explained by other than his feature based set theoretic model. Attempts have been made to reconcile these violations with both metric (Cooper 1979, 1980) and antimetric (Krumhansl 1978) distance functions. The important issues here, however, are the generality of Tversky's findings in a consumer products context and the ability to predict such contextual effects. As long as Tversky's model is a good predictor of contextual effects it will be of obvious usefulness to marketing researchers.

TESTING THE THEORY USING CONSUMER PRODUCTS

Method

Experiment 1 tested the asymmetry hypothesis for consumer products. Five product categories were used. Some of these categories were selected because they contain several homogeneous brands with widely varying market shares, such as soft drinks (colas and non-colas) and beers. The others represent product names within a larger, heterogeneous product category where imitation or substitute products must compete with established or prototype products. These products include frozen desserts {frozen yogurt versus ice cream), appliances {toaster versus waffle iron), and fruits (orange versus tangerine).

Assuming that high share or established products are associated with more distinctive features than their counterparts, consumer ratings of similarity in a subject/referent format should be asymmetric. For example, the similarity of Shasta Cola to Coke should be larger than the similarity of Coke to Shasta Cola. Given the nature of these product categories, the predictions that follow are made relative to the specific stimulus pair involved. For example, Coke is the market leader for colas followed by Pepsi and R.C. with brands such as Shasta and Canfield (a brand local to the Chicago area) bringing up the rear. If the features sets associated with the various products roughly correspond to their relative market shares, Tversky's theory predicts that asymmetry should hold, for example, in such cases as R.C. versus Coke and Shasta versus R.C..

Experiment 2 tested the relationship between similarity and dissimilarity judgments among soft drinks (colas and non-colas), beers, and fruits. These categories were chosen because they represent cases where there exist more than one low share or lesser known product and more than one relatively high share or established product. Assuming that high share products are associated with both more distinctive and more common features, then both the similarity and the dissimilarity ratings among them should be enhanced. This, in turn, should reduce the negative correlation between similarity and dissimilarity. For example, Coke and Pepsi should be both relatively more similar and more dissimilar than other pairs of colas.

Subjects

Participants included students from the University of Wisconsin and residents of the Madison area recruited from ads in the city newspaper and posters on campus. All of the assumptions concerning market shares and associated features were relative to this target market. For example, because Old Style beer is a market leader in this area, it is assumed to be associated with relatively more distinctive features.

Procedure

Stimulus pairs within each product category were randomized and divided into two stimulus sets, (A,B) and (C,D). Subjects were divided into four treatment groups. Treatment group I received subject/referent questions of the form "How similar is A to B?", Group II received questions of the form "How similar is C to D?". Group I was also asked to judge the similarity of pairs of products from stimulus set (C,D) with no subject/referent distinction while group II did likewise for stimulus set (A,B). Group III received questions of the form "How similar is B to A?" and then was asked to rate dissimilarities among stimulus set (C,D). Finally, group IV vas given questions of the form "How similar is D to C?" and then was asked to rate dissimilarities among stimulus set (A,B). These four treatment groups were used to avoid having subjects rate subject/referent similarities on the same stimulus pairs they faced in the similarity versus dissimilarity condition.

All of the judgments were elicited in a questionnaire format. Responses were made by placing an X on 125mm rating scales below each question. The scale ranged from "Not At All Similar" to "Very Similar" or from "Not At All Dissimilar" to "Very Dissimilar". After obtaining the judgments, memory probes were obtained on all of the products used in the study. The purpose of the probes was to check on the validity of the assumptions made concerning associated features.

RESULTS

Subject-Referent Asymmetry

The results support the generality of Tversky's criticisms in a subset of the product categories. Significant asymmetry was found to hold for both types of soft drinks and for beers. For example, the similarity of Shasta Lemon Lime Soda to Seven Up was significantly greater than the similarity of Seven Up to Shasta (t=2.15, P< .05). Both Pepsi and R.C. were found to be more similar to Coke than vice versa (t=1.42 and t=1.40 respectively, P< .10). Meanwhile, Shasta Cola was more similar to both R.C. and Pepsi than R.C. and Pepsi were to Shasta (t=2.22 and t=1.72 respectively, P< .05). A sample result for beers finds both Huber and Leinenkugel (two local Wisconsin beers) being more similar to Old Style than vice versa (t=1.95, P< .05, and t=1.55, P< .10, respectively).

A summary of the results for experiment 1 is presented in Table 1, Column 2 presents the average t value of stimulus pairs for which asymmetry was predicted in each product category. Columns 3, 4, 5, and 6 present the number of stimulus pairs involved, the standard deviation on the t's, a summary t value for each product category, and the significance of each summary value respectively.

TABLE 1

AVERAGE ASYMMETRY WITHIN PRODUCT CATEGORIES

Similarity versus Dissimilarity

The results of experiment 2 are remarkably consistent with the results of experiment 1. In those cases where asymmetry was significant, a relatively poor relationship was shown to hold between judgments of similarity and judgments of dissimilarity. Likewise, where subject/referent judgments were symmetric, strong negative relationships existed between similarities and dissimilarities. Table 2 presents the least squared regression results which compare the similarity judgments to the dissimilarity judgments for each of the product categories. Columns 2, 3, 4, and 5 present the regression constants, beta coefficients, r2 statistic, and r2 adjusted for the number of independent variables, respectively.

TABLE 2

SIMILARITY VS DISSIMILARITY REGRESSION RESULTS

The cola and beer categories provide the most interesting results. The adjusted r2 for colas is a dismal .076. This small value is due for the most part to the fact that, as predicted, Pepsi and Coke were relatively more similar and more dissimilar than any other pairs of colas (see Figure 1). Non-colas do moderately well, obtaining a fit of .585 (see Figure 2). The market leaders in beer, Old Style and Budweiser, were also relatively more similar and more dissimilar than other pairs of beers (see Figure 3). The best fit of all occurs for fruits (see Figure 4). Such results are consistent with those in experiment 1, where fruits failed to demonstrate asymmetry.

A preliminary analysis of the thought listings supports the assumptions made concerning associated features. A detailed analysis of the nature of the associations is, however, pending.

FIGURE 1

COLAS

FIGURE 2

NON-COLAS

FIGURE 3

BEERS

FIGURE 4

FRUIT

SUMMARY AND CONCLUSIONS

The results of this study support the generality of Tversky's theory in a consular products context. Contrary to the dimensional assumptions of existing market segmentation techniques, context has a significant and predictable influence on judgments of similarity among products within certain product categories.

Judgments of similarity were found to be poorly related to judgments of dissimilarity for certain products. For example, Coke and Pepsi were found to be both more similar and more dissimilar than other pairs of colas. Therefore, if one builds a product space for colas using similarity judgments, Coke and Pepsi will wind up being very close in proximity. But if one uses dissimilarity judgments as inputs to the construction of such a space, Coke and Pepsi will be relatively distant. Obviously this will have important implications with respect to the interpretation of the market and its use in the development of marketing strategies.

Whether a product is used as a subject or referent in a comparison may also influence the degree of similarity among certain products. If a product is associated with more distinctive features than other products in the same market, such features will detract more from judgments of similarity when that product is the subject rather than the referent in the comparisons. For example, because Coca-Cola is associated with an many distinctive features, a low share cola attempting to position itself close to Coke must be careful not to use Coke as a subject in any direct comparisons of the two products.

The consistency of the results in experiment 1 with those in experiment 2 provide convergent validity for the results as a whole and for Tversky's theory. In many cases it is simply more appropriate to view a product as being associated with sets of features than as varying on a small number of shared dimensions. As to why the homogeneous product categories were the only ones prone to such effects is the subject of the author's current research.

REFERENCES

Cooper, Lee G. (1979), The Reference Space Underlying Judgments of Similarity and Preference and Multiattribute Measures on Alternatives, working paper, Graduate School of Management, University of California, Los Angeles.

Cooper, Lee C. (1980), Discovering >Homogeneous Groups from Comparative Judgments, working paper, Graduate School of Management, University of California, Los Angeles.

Cunningham, James P., and Shepard, Rodger N. (1974), "Monotone Mapping of Similarities into a General Metric Space," Journal of Mathematical Psychology, 11, 335-363.

Krumhansl, C. L. (1978), "Concerning the applicability of geometric models to similarity and spatial density," Psychological Review, 85, 445-463.

Tversky, A. (1977), "Features of Similarity," Psychological Review, 84, 227-352.

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