Determining the Influence of Color on a Product Cognitive Structure: a Multidimensional Scaling Application

Larry Percy, Ketchum, MacLeod & Grove, Incorporated
[ to cite ]:
Larry Percy (1974) ,"Determining the Influence of Color on a Product Cognitive Structure: a Multidimensional Scaling Application", in NA - Advances in Consumer Research Volume 01, eds. Scott Ward and Peter Wright, Ann Abor, MI : Association for Consumer Research, Pages: 218-227.

Advances in Consumer Research Volume 1, 1974    Pages 218-227

DETERMINING THE INFLUENCE OF COLOR ON A PRODUCT COGNITIVE STRUCTURE: A MULTIDIMENSIONAL SCALING APPLICATION

Larry Percy, Ketchum, MacLeod & Grove, Incorporated

This paper is concerned with uncovering the influence of a stimulus' color on the perceptual space of the individual. Extensive qualitative investigation of a sauce product category indicated that color was an often articulated attribute with an expressed high value to the subjects. Within this category, as defined by the subjects, all brands but one were similar in overall visual impact: their packaging was basically the same, and their color, as seen in the bottle, was dark. The single exception, while labeled and marketed as a member of this category, was unique in its packaging, and was orange-red in color rather than dark brown or black. This exception provides the focus for this study.

If indeed, color is important to the perceptual structure of this category, some latent or real synesthetic effect may inhibit trial or usage of this brand. In this case, one might assume that the definition of a synesthetic effect would imply the tendency for a stimulus in one mode (i.e. color of the product) to elicit subjective "images" in another (e.g. expected taste or usage).

Two questions are thus raised: (1) is color in fact a dimension of the product cognitive structure; and if so, (2) is there present a synesthetic effect that could influence product selection. To answer these questions, two groups of stimuli were gathered. The first group provided brands subjects considered alike and central to defining the category in which the brand was marketed (referred to as the primary group). The second group included brands considered by subjects somewhat outside the category definition, but more similar in color to the test brand (referred to as the secondary group).

Non-metric multidimensional scaling was selected as an appropriate technique for determining how subjects perceive this overall category, and for making a determination of any possible synesthetic effect. This technique has been utilized by Gregson (1966, 1968) in analyzing a series of taste mixtures matched to histogram pictures of taste mixtures, and also in a study of taste mixture perception. Helm and Tucker (1962) have studied the structure of color perception with multidimensional scaling procedures. Also, the effects of color in a synesthetic effect has been studied with multidimensional scaling techniques. Lehmann (1972) reports on a multi-variate model of auditory-visual synesthesia where subjects related colors (and adjectives) as to their appropriateness in describing musical selections.

In this study, one first determined the cognitive space of individuals with respect to fully labeled products spanning the entire category. Next, the labels were removed and a second set of subjects studied to determine the cognitive space without the aid of a product label; in a sense, only bottle recognition and visual stimuli were operant. Finally, a third set of subjects was studied in order to determine the perceptual configuration of the products in terms of their perceived taste and usage when the stimuli were presented without label or packaging. This final experiment sought to determine the extent to which any synesthetic effect may operate, relating as it does a stimulus color to perceive taste and use.

One may hypothesize that if a synesthetic effect obtains, one would expect the final configurations to relate strongly to a color continuum along their dimensions. The darker stimuli should exhibit similar taste or usage characteristics, and these characteristics should be different from the more red or orange stimuli. One would also expect to find these stimuli arrayed more or less in accordance with the general psychological color space or natural color spectrum.

In terms of the test brand, if color is an important consideration, one would expect to find it in a perceptual proximity to the primary group of stimuli only when tested fully labeled and identified as such. Once the label identifying the test brand as a member of the primary group of stimuli is removed, it should begin to move into perceptual proximity with stimuli more visually similar (e.g. products whose color are more red or orange rather than dark brown or black).

METHOD

Sample

Female heads of household from five geographic areas representing a variety of demographics and life-styles participated in this study. Personal, in home interviews were conducted with subjects selected on an area-wide cluster basis within each city, each cluster representing a different income/life-style area. Additionally, the sample was stratified to include broad demographic representation (e.g. family size, age of respondent and outside-the-home employment) and also various category usage behavior. The interviewing criteria required each interviewer to work in at least two clusters, and to complete interviews among subjects qualified by the above stratification. This insured equal representation of behavior and demographics.

Data Collection Procedure

The study itself consisted of three experiments: in the first, 25 subjects were confronted with proximities tasks associated with the relationships among branded stimuli; in the second, 50 subjects dealt with the relationships among unbranded (i.e_ unlabeled) stimuli; in the third, 50 subjects reacted to the relationships among unidentified (i.e. unbottled) stimuli.

For the first experiment seven items representative of the general product category, plus the test brand, were chosen as stimuli. These stimuli represented the primary group of brands central to the category plus the secondary group considered peripheral. The leading brand(s) from each group were included. While the stimuli themselves are identified in the following analysis according to group and general color, the respondents were given only photographs of the stimuli with no suggestion as to an a priori categorization.

In their first task, the 25 subjects were presented with a randomly shuffled pack of twenty-eight, 3" x 5" full color photographs. Each photograph showed one of the twenty-eight possible pairs of the eight stimuli. The subjects were then asked to place the photographs into two piles; all those photographs they felt showed two similar items were to go into one pile, all those showing dissimilar items into a second. From the first pile, the photograph showing the two most similar items was selected, followed by the next most similar two items, et cetera, until all of the photographs containing similar items were ranked from most through least similar. The same procedure was then followed for the second pile until all the photographs containing dissimilar items were ranked from most through least dissimilar. These results provided a proximity measure of similarities judgements among the eight branded stimuli.

These same subjects were then given a randomly shuffled group of eight, 3" x 5" full color photographs. This time each photograph showed one of the eight stimuli by itself. The subjects were asked to group the eight photographs in any way which was sensible to them. It was emphasized that there were no right or wrong ways to group the photographs, and that they could make as many or as few groups as they liked, with as many or as few photographs in each as they felt appropriate. For each grouping of stimuli, the subjects were asked for their reasons in grouping those particular items together. This added information aided in interpretation. The results of these groupings provided a proximity measure based on an unspecified subject sorting of the eight branded stimuli.

The second experiment utilized six of the original eight stimuli, including the test brand. The two stimuli from the original eight dropped from this experiment were each one of two different brands of the same type, which additionally were of identical color. Because the thrust of this second exercise was to study the perceived similarities among unlabeled items, it was felt that identical products would be redundant, and obviously alike.

A new stimulus was added in order to extend the color spectrum represented by the original stimuli. This required going somewhat outside the general product category, but since the items were unlabeled and the added stimulus visually similar, the addition was considered reasonable. The original spectrum which ranged from black through red-orange was now continuous from black through orange.

Fifty new subjects were presented with a randomly shuffled pack of twenty-one 3" x 5" full color photographs. Each of the photographs showed one of the twenty-one possible pairs of the seven unlabeled stimuli. It should be noted that although these items were unlabeled, their color, relative consistency and bottle shape were clear. As with the earlier group of respondents, these subjects were asked to place the photographs into two piles, those showing pairs of items perceived similar and those showing pairs of items perceived dissimilar. Within each pile the photographs were ranked, providing a proximities measure of similarities judgements among unlabeled stimuli.

The final experiment involved the same seven stimuli utilized in the second experiment. This time, however, they were presented as seven 3" x 5" full color photographs, where each photograph showed a single item poured out unidentified on a plain white dish.

For each stimulus, another 50 subjects were asked to check which of a list of sixteen words they felt described how that particular item would taste. Each subject was encouraged to check as many or as few words as she felt applied. These words resulted from an extensive qualitative investigation.

Once this task was completed, the subjects were asked to check which of eleven possible usage situations they thought appropriate for each particular item. Again, they were encouraged to check as many or as few as they felt applied.

Correlations among the profiles for each stimuli provided a proximities measure of perceived taste and of perceived usage among the seven unidentified stimuli.

Analytic Plan

The first step of analysis was to determine the perceptual configuration of the full category, including the test brand, primary, and secondary groups of products. Utilizing Kruskal's M-D-SCAL program for non-metric multidimensional scaling (Kruskal, 1969), the proximity comparisons gathered in the first experiment were examined. Additional proximities measures based on stimuli sorting were also considered as a point of comparison to the more traditional and stronger proximity comparisons measures. First, the number of times each pair of stimuli were placed in the same category was used as the measure of proximity. Second, after specifying the single most important criteria along which the stimuli differed and sorting accordingly, proximities measures were examined for each of the subject-specified criteria. One was seeking at this stage confirmation that the same basic dimension(s) that establish the overall parameter(s) of the product cognitive structure also underlie the product cognitive structure related-to verbalized product differences (i.e. the subject-specified criteria).

The second step was to establish what the perceptual configuration of the stimuli would be if the items were unlabeled, and specifically what cognitive differences were evident between users and non-users of products in the category. Non-metric multidimensional scaling solutions of the data gathered in the second experiment were sought for both groups of subjects utilizing proximity comparison measures. One was at this stage particularly interested in any movement of the test brand in comparing user vs. non-user configurations, and in comparing the unlabeled solution spaces with the original perceptual configuration.

The final step in the analysis was to examine proximities measures gathered in the third experiment by correlating expected taste attributes and expected product usage among unidentified stimuli. The resulting configurations were studied for indications of synesthetic effects associated with the products, with special emphasis on the test brand's compatibility with the primary group of stimuli

RESULTS

Relationships Among Branded Items

Proximities data were collected among eight branded stimuli representative of the product category. These data were measured as similarities judgements of all n(n-l)/2 pairs. The resulting configurations yielded comprehensible and readily interpretable solutions with a good fit in as low as a single dimension (with a STRESS value of .156). As the solution vector shown in Figure 1 indicates, this single dimension is highly related to a dichotomy of stimuli groupings (remembering that our test brand is labeled as a primary stimulus). Excluding the test brand, this same dichotomy could easily be interpreted as color, suggesting that an important (if not the most important) characteristic of the perceptual reality of consumers when comparing these brands is their perceived image of the brand's color

FIGURE 1

NON-METRIC MULTIDIMENSIONAL SCALING SOLUTION VECTOR FOR EIGHT LABELED STIMULI: PAIRWISE SIMILARITY.

The two dimensional representation offered little improvement, reducing STRESS only to a level of .135, compared with the STRESS of .156 noted above for the single vector solution. A substantial improvement, however, was obtained by moving into three dimensions; here STRESS becomes a very low .0197. As the configuration shown in Figure 2 indicates, the color dimension suggested in the vector solution emerges, highly correlated with a dimension reflecting consistency: the darker the stimulus, the less thick it is apparently perceived. This perhaps explains the small reduction in STRESS moving from a single to two dimensions: there is very little to distinguish the color dimension from the consistency dimension. If this is in fact true, one may assume that the perceived color of a brand in this category is highly suggestive, in the consumer's mind, of it's perceived consistency. The third dimension, and the one providing the key to complete clarity in interpreting the consumer's understanding of the similarities among these eight branded items, is the general "use" pattern for the items.

FIGURE 2

NON-METRIC MULTIDIMENSIONAL SCALING CONFIGURATION FOR EIGHT LABELED STIMULI: PAIRWISE SIMILARITY.

It is perhaps of interest that specificity of taste does not appear among the principle criteria of judgement when comparing these branded items. Rather, it seems to be subsumed in the perceptual domains of perceived color and consistency. Later analysis examines the synesthetic effect of perceived taste and use, and color.

Analysis of the stimulus proximities collected by subject sorting yielded a configuration less precise in terms of STRESS than the above solutions resulting from the proximity comparisons (.123) in three dimensions vs. .0197). Perhaps an explanation for this lies in the fact that a single sorting for proximities measures provides no information about the differences in similarity between stimuli in the same category or differences in similarity between categories (Wish, 1972a). In terms of interpretability, there appears to be considerable agreement. As Figure 3 shows, the dimensions of color and consistency remain, while the earlier dimension of "use" has been replaced by a usage dimension that seems to be related to the "all-purposeness" of the items.

Two additional configurations were developed from stimulus proximities resulting from subject-specified sorting; one representing a composite of general taste and flavor criteria, the second a specific taste attribute. The configuration shown in Figure 4, representing the scaling solution of the subject-specified criterion proximities data on general taste/flavor, show a remarkable degree of congruence with Figure 3. It would appear that when subjects are asked to freely sort the stimuli, the criteria they use are the same as those they use when sorting the stimuli on an imagined general taste/flavor criteria. Again, as with the pairwise proximities data, it appears that a subject's perceptual considerations of color and consistency play an important role in her understanding of this category. The solution for the subject-specified specific taste criteria was generally unsatisfactory in three dimensions (a STRESS value of .205), perhaps reflecting a greater degree of cognitive variance associated with the percePtions of that specific taste attribute.

FIGURE 3

NON-METRIC MULTIDIMENSIONAL SCALING CONFIGURATION FOR EIGHT LABELED STIMULI: OVERALL SIMILARITY.

FIGURE 4

NON-METRIC MULTIDIMENSIONAL SCALING CONFIGURATION FOR EIGHT LABELED STIMULI: TASTE/FLAVOR SIMILARITY

Relationships Among Unlabeled Items

Pairwise proximities data were collected among seven unlabeled stimuli where the product itself could be clearly perceived in the bottle. These data were again measured as pairwise similarities judgements of all n(n-l)/2 pairs.

It was felt that experience with the category might enlarge the perceptual domain because of possible latent brand associations. Therefore, users and non-users of products in this category were studied separately. The resulting configuration for non-users achieved a remarkably good fit in two dimensions with a STRESS of .0066. Looking at this configuration in Figure 5, one once again finds the dimensions of color and consistency; there is no doubting that color is the principal perceptual dimension dichotomizing these products; nor that the test brand, without benefit of it's category-defining label, now clearly clusters among items more similar in color.

Among users in this category, a two dimensional configuration does not seem to be sufficient. Moving from a two dimensional to a three dimensional solution space reduces STRESS from a moderately high .221 to a very low .009. Once again we note in Figure 6 the dimensions of color and consistency, and in this case, the addition of "type". It would seem that indeed, experience with the products in this category adds to one's perception. Users appear able to imply brand (and by extension the category group) to unlabeled products, and to use this added dimension to differentiate among the unlabeled stimuli. Among these users, the test brand seems to be recognized as a primary group stimulus, though not in color or consistency. This configuration retains the basic dimensions found in the analysis of the pairwise proximities data among branded stimuli (see Figure 2), with the "type" dimension replacing the more specific "use" dimension. The added seventh stimulus assumes a logical position reflecting it's color and consistency, while filling in an empty or "new" position descriptive of it's "type".

FIGURE 5

NON-METRIC MULTIDIMENSIONAL SCALING CONFIGURATION FOR SEVEN UNLABELED STIMULI AMONG NON-USERS: PAIRWISE SIMILARITY.

FIGURE 6

NON-METRIC MULTIDIMENSIONAL CONFIGURATION FOR SEVEN UNLABELED STIMULI AMONG USERS: PAIRWISE SIMILARITY.

Relationships Among Unidentified Items

Each of the seven stimuli were presented to another group of subjects, this time poured unidentified on a plain white dish. Profile comparisons were made over a number of taste and usage variables: the perceived taste of the seven items was measured over sixteen attributes, the usage of each over eleven situations. Correlations between the various perceptual measures of the items on first the taste attributes, and then the usage attributes, were utilized as proximities data for multidimensional scaling.

Computing the configuration for perceived taste, we found a readily interpretable solution with a good fit (a STRESS value of .102) in as low as a single dimension. The solution vector shown in Figure 7 clearly suggests a dichotomy of color, indicating that there is a certain synesthetic relationship between perceived taste and color. The darker items were perceived as being more "spicy", and as one proceeds up the vector, the items become increasingly "mild". Moving into two dimensions reduces the STRESS to .0092, but offers little improvement in the interpretability of the space. The strong clustering of the primary group stimuli continues in evidence, along with the reluctance of consumers to include the test brand in that category. Perceptually, the test brand appears to be more similar to the secondary group.

FIGURE 7

NON-METRIC MULTIDIMENSIONAL SCALING SOLUTION VECTOR FOR SEVEN UNIDENTIFIED STIMULI: TASTE PERCEPTION COMPARISON.

Comparing the one dimensional vector solution among pairwise similarities of branded stimuli (see Figure 1) with the one dimensional vector solution obtained here, one sees that the basic dichotomy of stimuli groupings corresponds strongly with the taste dimension, each of which seems directly related to color. Only the test brand, when judged with its category-defining label, breaks this dichotomy of color. It should be pointed out, in fact, that in the stimuli's unembellished presentation here, the solution vector reflects a striking continuity with the natural and psychological color and hue spectrums.

The configuration for proximities data measuring similarities of perceived use among the stimuli once more yields a remarkably good fit in a single dimension; the STRESS value was a low .009. Interpreting the solution vector shown in Figure 8 reveals a clustering of items by color where the darker (and hence perceived "spicy" tasting) primary group stimuli anchor one end of the solution vector, reflected by the "redder" (and hence perceived "milder" tasting) secondary group stimuli, along with the test brand. The added orange stimulus, which was not an actual member of the product category, finds itself far out along the solution vector in the direction of a non-dark, milder product (which indeed it was).

FIGURE 8

NON-METRIC MULTIDIMENSIONAL SCALING SOLUTION VECTOR FOR SEVEN UNIDENTIFIED STIMULI: USAGE PERCEPTION COMPARISON.

DISCUSSION AND CONCLUSIONS

In this study, attention has been drawn to the dimensional representation of a specific food product category cognitive structure, and the use of multidimensional-scaling procedures of various proximity measures to recover such structure. Of particular interest was the judgement of a specific test brand within this structure, and the relationship of color to the domain. One was seeking an implicit theory about the product world, a theory which presumably underlies purchase decision-making.

Evidence from the present analysis leads one to the conclusion that indeed, color is an important component of product cognition in this category, and that a strong synesthetic effect appears evident. Changes in cognitive structures were followed by observing the way in which the product cognitions change as less information for choice decision was made available to the subjects. The test brand clearly fell outside the perceptual domain of the primary group stimuli when unaided by a category-defining label.

When the stimuli were studied as fully labeled and branded items, the test brand tended to map in the general proximity of the brands in the primary group along only one dimension; the obvious dimension of "use". In spite of the label identifying the test brand as a primary stimulus, subjects did not perceptually acknowledge similarity in other dimensions reflecting perceived attributes of color and consistency. One may intuit that were the test brand's label to identify it as a secondary stimulus, subjects would exhibit less dissonance in their expressed perceptual judgements: the only link joining the test brand to the primary stimuli is its label identification.

Once this identification is removed, the test brand convincingly maps with the secondary stimuli among non-users of products in this category. They perceive the test brand as similar to the secondary group of stimuli along the two principal dimensions of their perceptual space, consistency and color. While category users continue to recognize "type", even without labels, and tend to utilize it in their perceptual judgement of the stimuli (paralleling the labeled configuration), the important finding is that those subjects who are potential users do not. For these subjects, the unlabeled test brand is considered similar to unlabeled stimuli perceptually opposite o. the primary group. To label the test brand a primary group product would appear to be inconsistent with natural cognition.

This point is strongly reinforced in the analysis of taste and usage expectation, where a strong synesthetic effect is noted. When subjects have only the raw stimuli for judgement, they appear to translate the stimulus color to a perceived notion of taste, ascribing various taste attributes in a monotonic relation to the position on the color spectrum (i.e. if an orange stimuli were considered mild, a red stimuli would be considered less mild, and so on). The taste perception maps quite nicely in a single dimension, reflecting the color continuum running from orange through black.

The perceived usage of the stimuli, when subjects have only the raw stimuli for judgement, again follows this same orange through black color continuum. It would seem that the visual color mode reflected by the stimuli is interpreted in terms appropriate for specific usage, and that the test brand is perceived as similar to the secondary group of stimuli rather than the primary group.

The synesthetic effects noted here as associated with these stimuli offer important insight into the latent effect of a product's color in influencing consumer perception. It is quite reasonable to assume that for this category ( and by implication other food product categories) color is a principal component of product cognition, strongly influencing the determination of other more commonly considered gustatory and end-use perceptual dimensions. The strong monotonic relationship between color and basic taste and usage expectation may greatly influence possible trial; in a sense, a product is ascribed attributes on the basis of its color.

The implicit theory or color's influence on product cognition and ultimate purchase decision-making would appear to have merit. The utilization of multidimensional scaling procedures to illustrate the perceptual judgements of consumers proved useful, clearly positioning the test brand under study as more like a secondary stimulus than a primary stimulus. The dissonance created in the consumer's mind when confronting a product at the point of purchase which is perceived to be one thing while labeled another (specifically in this case a different tasting product and one seemingly more appropriate for other uses) must certainly hurt trial. Next to the test brand on the shelf would be other items in harmony with the consumer's general cognition. It would be far easier for a consumer to choose a product which "looks" like it should taste and be used as the label indicates. For the test brand to effectively compete with these brands, it would seem the color perception should be more compatible with the category.

The application of multidimensional scaling procedures over a number of different proximity measures is a particularly useful and sensitive approach to representing certain aspects of cognition. It seems clear that subjects do use a relatively small number of organizing constructs for structuring product information in this category, and that the relationship between a general color dimension and these constructs is apparent.

It remains for future research to determine the overall effects of color in determining the perceptual judgements and evaluation of consumers with regard to food products. This study has indicated that for one category of food products, sauces, color is indeed critical, and that for a particular test brand in the category, its atypical color more closely identifies it in the consumer's mind with products outside the category.

REFERENCES

Carroll, J.D. Algorithms for rotation and interpretation of dimensions and for configuration matching. Paper presented at Workshop of Multidimensional Scaling, University of Pennsylvania, June, 1972.

Coombs, C.H. A theory of data. New York: John Wiley & Sons, Inc., 1964.

Gregson, R.A.M. Theoretical and empirical multidimensional scalings of taste mixture matchings. British Journal of Mathematical and Statistical Psychology, 1966, 19, 59-76.

Gregson, R.A.M. Stimulating perceived similarities between taste mixtures having mutually interacting components. British Journal of Mathematical and Statistical Psychology, 1968, 21, 117-130.

Helm, C.E. & Tucker, L.R. Individual differences in the structure of color perception. American Journal of Psychology, 1962, 75, 437-44.

Kruskal, J.B. Multidimensional scaling by optimizing goodness of fit to a nonmetric hypothesis. Psychometrika, 1964, 29, 1-27.

Kruskal, J.B. How to use M-D-SCAL, a program to do multidimensional scaling and multidimensional unfolding. Version 5M of M-D-SCAL, October, 1969. All in Fortran IV, Bell Telephone Laboratories, Murry Hill, New Jersey.

Lehmann, R.S. A multivariate model of synesthesia. Multivariate Behavioral Research, October, 1972, 403-39.

Rapoport, A. & Fillenbaum, S. An experimental study of semantic structures. In R. N. Shepard, et al. (Eds.), Multidimensional Scaling: Theory and Applications in the Behavioral Sciences. Vol. II; Applications. New York: Seminar Press, 1972, 93-131.

Shepard, P.N., Ramney, A.K. & Nerlove, s. (Eds.) Multidimensional Scaling: Theory and Applications in the Behavior Sciences. Vol. I; Theory. New York: Seminar Press, 1972.

Wish, M. Notes on the variety, appropriateness, and choice or proximity measures. Paper presented at Workshop on Multidimensional Scaling, University of Pennsylvania, June, 1972. (a)

Wish, M. Brief Comments about dimensionality. Paper presented at Workshop on Multidimensional Scaling university of PennsYlvania. June 1972 (b)

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