The Explanatory Power of Values in Preference Judgements: Validation of the Means-End Perspective

ABSTRACT - Means-end theory proposes that product preferences are significantly influenced by personal values which serve to give relative salience to the consequences derived from consuming the product and to the product's attributes. In this study, the chain from product attributes (A), to consequences (C), to values (V) is uncovered through an in depth interview technique, laddering. An individual's chains of A/C/V's are then related to their pairwise judgements of product preferences and perceptions through ordinal regression. This paper will statistically assess the hypothesis that values contribute significant explanatory power for product preferences, over and above the explanatory power of consequences or attributes. In addition, the paper explores how this explanatory power differs by level of product usage.

Citation:

W. Steven Perkins and Thomas J. Reynolds (1988) ,"The Explanatory Power of Values in Preference Judgements: Validation of the Means-End Perspective", in NA - Advances in Consumer Research Volume 15, eds. Micheal J. Houston, Provo, UT : Association for Consumer Research, Pages: 122-126.

Advances in Consumer Research Volume 15, 1988      Pages 122-126

THE EXPLANATORY POWER OF VALUES IN PREFERENCE JUDGEMENTS: VALIDATION OF THE MEANS-END PERSPECTIVE

W. Steven Perkins, The Pennsylvania State University

Thomas J. Reynolds, The University of Texas at Dallas

ABSTRACT -

Means-end theory proposes that product preferences are significantly influenced by personal values which serve to give relative salience to the consequences derived from consuming the product and to the product's attributes. In this study, the chain from product attributes (A), to consequences (C), to values (V) is uncovered through an in depth interview technique, laddering. An individual's chains of A/C/V's are then related to their pairwise judgements of product preferences and perceptions through ordinal regression. This paper will statistically assess the hypothesis that values contribute significant explanatory power for product preferences, over and above the explanatory power of consequences or attributes. In addition, the paper explores how this explanatory power differs by level of product usage.

INTRODUCTION

The means-end model (Gutman, 1982) assumes that a consumer's personal values influence product choice by giving salience to the concrete attributes of the product and the benefits it provides. Understanding the link between product choice and consumers' value systems has been successfully applied to product positioning and assessing product strategy (Reynolds and Gutman, 1984). This paper will test the hypothesis that values add statistically significant explanatory power to preferences over and above product benefits and attributes.

Products go beyond functional properties to have meaning in the consumer's life, according to means-end theory. Consumers move from the concrete level of product attributes to the positive benefits provided by consuming a product on to the highest level of abstraction, personal values (Reynolds, 1985). Consumer preferences may be explained in terms of this chain from attributes to consequences to values. While a great deal of research has been based on attributes and even benefits, surprisingly little has attempted to make the step to values.

One effort to make this final link between personal values and preferences has been through in-depth, individual interviews with consumers, termed laddering. Starting at the lowest level in the chain, consumers discuss what attributes they use to discriminate between products and why they are important. Next they are probed to think about what consequences they derive from the product and finally why this consequence is important to them. At this highest level, values define what consequences a person sees as desirable and the consequences in turn define what attributes appear important. Therefore, in means-end theory the product is translated from concrete attributes (A) to an abstract consequence (C) and then linked directly to self through values (V). These A/C/V chains can then be used to explain differences in product preferences (Reynolds and Perkins, 1986).

This paper focuses on the contribution of the values level in explaining preference. The major concern here is whether including values improves our predictive ability. Assuming that including the values level does prove to be worthwhile, a second research question relates to how the A/C/V chain differs by intensity of product usage. As a consumer uses the product, the cognitive links in the chain change, thus the ability to predict preference changes also. For instance, the attribute level may not mean as much to a heavy user who does not choose based on distinctions between products' physical features. At the other end of the chain, values may add even more explanatory power to heavy users' preferences than to light users, because the product is more closely linked to the self.

BACKGROUND

Two small-scale consumer studies (Reynolds, Gutman, and Fiedler, 1984; Reynolds and Jamieson, 1984) have reached the same tentative conclusion--the value level of the means-end chain produces the greatest fit with preference judgements, followed by the consequence level second and the attribute level third. On the other hand, if perceptual rather than preference judgements are considered, the fit does not increase from attributes to consequences to values. Means-end theory argues that preference and perception stem from different processes. Physical attributes serve as the basis for perceptual distinctions between products, but preferential differences develop from within the consumer. Preference is based on how the product is personally meaningful to the consumer.

In a study related to job performance appraisal, Jolly, Reynolds, and Slocum (in press) found that values did add significant explanatory power beyond attributes and consequences. Of course, job performance is a very values laden topic. This proposition has not been statistically tested for relatively low involvement topics, such as consumer products.

The second research question concerns how the explanatory power of values changes with the level of product usage. A great deal of recent research has examined the effects of product experience on a consumer's cognitive structure (see Alba and Hutchinson, 1987, for a review). Cognitive structure refers to the knowledge a consumer has about products and how that knowledge is organized. As product experience increases, consumers accumulate more knowledge about more products, learning how to make finer distinctions between them. Their understanding of the product category will be more complex and abstract compared to a "novice" consumer. Also with experience knowledge becomes organized by the function or meaning of products rather than their surface features.

As an exploratory extension of the basic research question addressed in this paper, the A/C/V structure of "expert" consumers will be compared with that of "novice" consumers. Framed in means-end terminology, the change in cognitive structure due to increased product experience would predict that "novices" attend to the concrete attributes of a product as compared to "expert" consumers who are more concerned with the benefits that can be derived from the product and the meaning of the product to them. Though experienced users know more about product attributes than others, they have moved beyond that level of analysis to a deeper level of meaning. For the purposes of this study, "expert" and "novice" consumers were identified by their product usage rate. The explanatory power of the values level of the means-end chain should be greater for the preference judgements of experts than for novices.

METHODOLOGY

Sixty consumers who were familiar with salty snack foods participated in an extensive study of product perceptions and preferences and their relation to product attributes, consequences, and values. The subjects, two-thirds of whom were women, lived in a large metropolitan area. Each consumer worked individually with a marketing researcher through a two hour interview that consisted of several ratings, rankings, and open-ended questions about salty snack foods, including assessment of competitive advertising. A key element of the interview was the laddering procedure which essentially determines the higher level meanings of product attributes (Gutman and Reynolds, 1979). The main focus of the study was on nine competing brands in the salty snack category.

To obtain pairwise perception ratings for the nine brands, consumers received a set of thirty-six cards each of which named a pair of brands. The researcher instructed the subject to sort each card into one category along a scale from "very similar" (1) to "very dissimilar" (9). In effect the consumer rates every pair of brands on a nine point scale. Following a procedure suggested by Cooper (1973), the preference rating task was repeated in a nearly identical fashion. First the consumer was asked which member of the pair was preferred and then placed the card in the appropriate pile corresponding to a nine point scale ranging from "about equally preferred" (1) through to "totally prefer one over the other" (9). The resulting ratings were arranged into two proximity matrices per consumer. Thus both psychological distances and intensity of preferences between brands are measured as multidimensional constructs, using the same basic nine point scale. Both judgements represent distances between pairs of stimuli, a similarity and a preference distance. The similarity task was performed first, a different rating scale task was then performed, then the preference judgements.

The researcher then "laddered" each consumer starting with concrete bipolar attribute constructs elicited from the subject in a triadic sort task (Kelly 1955). The "reason for importance" was obtained after asking the respondent which pole of the bipolar construct was preferred and why. This method of probing is followed for each respondent by "why is that important to you" questions, until the consumer can no longer continue "up the ladder." For example, common attributes identified for salty snacks were "crunchy," "flavor," and "taste." The consumer may state that they like more "crunchy" snacks because they are "good with dips," "filling" and "relaxing." These consequences of eating crunchy snacks could possibly lead to the personal values of "self-enjoyment" and "belonging." Following the laddering procedure, consumers rated all nine products on the degree to which they possess or facilitate the attributes, consequences and values uncovered. Each consumer elicited at least two complete ladders, determined by having one element at each A/C/V level. The specific A/C/V chains differed by subject.

Consumers also rated their frequency of use in each of seven occasions, such as lunch, snacks, or entertaining. The five point rating scale ranged from "five or more times a week" (1) to "less than once a month" (5). An overall frequency of use index was obtained by summing over all occasions for each consumer (Table 1). The fifteen heaviest users averaged almost twice the level of consumption of the lightest users.

TABLE 1

FREQUENCY OF PRODUCT USE

At the end of the interview subjects had (1) rated every pair of brands for perceptual similarity and (2) preference intensity as well as (3) attribute, consequence, and value ratings specific to each brand for the elements of two A/C/V ladders and (4) the frequency of use by occasion.

COGNITIVE DIFFERENTIATION ANALYSIS

The relationship of the pairwise preference or perception judgements to the attributes, consequences and values is estimated by Cognitive Differentiation Analysis (CDA), which is described fully in Reynolds and Sutrick (1986) and extended in Reynolds, Weeks, and Perkins (in press). Briefly, CDA is an individual level analysis which uses ordinal regression to relate each vector of brand ratings to the proximity matrices. The procedure resembles property fitting where a matrix of distances is submitted to multidimensional scaling then the vectors of ratings regressed on the derived scaling stimulus coordinates. CDA has several advantages-over property fitting including direct ordinal regression of the matrix onto one or more vectors of product ratings. In addition, standard regression measures such as R-squared, estimated Betas, F and t statistics are calculated as well as ordinal measures of association specific to this technique.

Because the ordinal regression requires decomposing the data into pairs of pairs, the predicted dependent variable must be "refolded" back into pairs corresponding to the original pairwise judgements. The squared Pearson correlation between the predicted and actual judgements, termed "R-squared refolded" or R2r, assesses the fit between product ratings and pairwise product judgments. With nine salty snack products in this study, subjects made 36 paired comparisons. Because of the pairs of pairs decomposition, there are 630 unfolded predicted values but only 36 "refolded" predicted values. Thus a correlation of .27 (or about .07 if squared) is significant at a .05 level. The analyses in this paper use the R2r's as the dependent variable.

DESIGN OF THE ANALYSES

This paper concerns the amount of variance explained, R2r, by value level ratings compared to consequences and attributes. If values significantly improve the ability to predict preferences, this would argue for going beyond product attributes and even beyond the consequence level to include personal values. One method of statistically testing the explanatory contribution of values is by nesting CDA regression models. Using a forced order of entry, attributes are entered first, consequences second, and values third, with the incremental change in R2r indicating the increased explanatory power of the set of variables. In effect, the analysis answers how much the R2r increases when more independent variables are included in the CDA regression equation. Six CDA equations were estimated per consumer--attributes only, then attributes and consequences, and finally all three, for perception and preference judgements. Previous means-end research using CDA has used this methodology to study the incremental gain in R2r from attributes to consequences to values for job performance appraisal (Jolly, Reynolds, and Slocum, in press).

Results

Mean CDA R2r values for the six equations are presented in Table 2. The mean over all perception measures equals .19 compared to .32 for preference. Also the R2r increases from .15 for attributes to .28 for consequences and .35 for values. CDA R2r increases from attributes to consequences to values for both preference and perception, but increases much more for preference than for perception.

TABLE 2

MEANS FOR PREFERENCE AND PERCEPTION TASKS BY ATTRIBUTE, CONSEQUENCE AND VALUE MEASURES

An analysis of variance on the CDA R2r data was performed across all sixty consumers and six equations per subjects with task (preference and perception) and measure (attributes, consequences, and values) as the main effects. The equation shows significant differences (Table 3) for both main effects and their interaction as expected.

TABLE 3

ANALYSIS OF VARIANCE RESULTS

Figure 1 plots the six task by measure means and shows the relevant pairwise -test values, all of which are significant (p<.01). It is obvious that values do significantly increase the predictive power for both preference and perception judgements over attributes and consequences.

FIGURE 1

PLOT OF ATTRIBUTES, CONSEQUENCES AND VALUE R2r FOR PERCEPTION AND PREFERENCE T-TEST RESULTS BETWEEN PAIRS

To measure the incremental gain in R2r, a regression equation was estimated which allowed for different coefficients by task and measure. This dummy-variable regression (not a CDA regression) provides the same result as the previous ANOVA but specifically tests whether the mean R2r's differ by test and measure. The regression coefficients indicate the increase in R2r due to adding consequences to attributes or adding values and consequences to attributes. It also measures the differences between perception and preference R2r's. The resulting equation (Table 4) finds that means significantly differ as expected. All coefficients are significant at a .05 level. Using attribute measures as the basis, both perception and preference R2r increase by including consequences and increase again by including values, but the rate of increase is greater for preference.

TABLE 4

REGRESSION EQUATION FOR TASK BY MEASURE

Interpreting the regression equation simply requires adding up the appropriate coefficients. When only attributes were used to explain perception or preference matrices, the R2r averages the intercept alone, .15. For perception judgements, the incremental change in CDA R2r due to adding consequences to attributes was .05 while including both values and consequences added .10 to the intercept. On the other hand, for preference judgements two coefficients must be added to the intercept. Including consequences increases the amount of variance explained for preferences by .22 (.05 for consequences and .17 more for preferences). Finally, when both values and consequences were added to attributes to explain preferences, the average R2r increased by .30 (.10 plus .20). In general, including the value level ratings adds .05 to the average CDA R2r in explaining perception over and above including consequences and attributes. However, when the CDA regression was performed on the preference judgements, the average increase in R2r due to including the value level ratings was .08.

An additional set of analyses involved the consumers' frequency of product use. An analysis of variance with task, measure, and frequency of use results in all three main effects being significant (Table 5). As will be detailed later, R2r decreases with greater product usage. In other words, across all three measures and both tasks the average amount of variance explained declines with heavier use.

TABLE 5

ANALYSIS OF VARIANCE RESULTS

The task by frequency interaction is significant. In order to illustrate the importance of this interaction, the usage index will be categorized into three levels; the 25% of consumers who use the product the most frequently were designated the "heavy users," the 25% who use it the least, the "light users," and the 50% in the middle as "medium users." Means for these three groups by preference and perception are displayed in Table 6.

TABLE 6

MEANS FOR PREFERENCE AND PERCEPTION TASKS BY FREQUENCY OF USE

Note that across users the mean perception value does not change much from .16 to .20, but for preferences the heavy users have a mean of .27 which increases to .42 for light users. Preference means are higher than perception means, but light users have a much larger difference between the two.

Though the three way interaction of task by measure by frequency of use was not significant, inspecting only the preference task finds a very interesting relationship. For all consumers, the highest R2r was achieved when values were included in the CDA regression equation; however, the magnitude of the R2r for attributes and consequences decreases dramatically with frequency of use. Table 7 shows that heavy users have an attribute R2r of .09 compared to .26 for lighter consumers. Also consequences increase from .29 for heavy users to .47 for light users.

TABLE 7

PREFERENCE MEANS BY FREQUENCY OF USE AND ATTRIBUTES, CONSEQUENCES, AND VALUES MEASURES

While all consumers increase about 20 points when consequences are added to attributes, the gain from adding values to consequences differs by usage level. Heavy users increase by 14 points compared to only 6 for.light users. In fact, for heavy users the mean CDA R2r increases 48% by including values, but only 13% for light users. This difference is portrayed in Figure 2 which shows a small change from consequences to values for light users but a larger change for heavy users. Also the pairwise t-tests are presented, all of which are significant (p<.05).

FIGURE 2

PLOT OF ATTRIBUTES, CONSEQUENCES AND VALUES FOR HEAVY AND LLGHT USERS WITH T-TEST RESULTS BETWEEN PAIRS

In summary, there are two major results--first, values add significant explanatory power over consequences and attributes, especially for preference. Secondly, values contribute more to the explanation of preference for heavy users than for light users.

CONCLUSIONS

This research has confirmed the hypothesis that personal values add significantly to explaining product preferences. As predicted by means-end theory, products have meaning to a consumer beyond the attribute or even benefit level. In contrast, explaining the perceptual differences between products is not improved as much by including values.

The more exploratory aspect of this research concerned the effect of product usage rate on the means-end chain. Values appear to contribute more to explaining heavy users preferences than light users. There may be stronger ties between the self and the product for heavy users. On the other hand, attributes capture almost half of the total variance explained for light users indicating that they may be making their choices primarily on concrete product features.

These results have important implications for product positioning and strategy development. First, all three levels in the A/C/V chain should be considered in order to develop a complete picture of consumer preference. Failure to include values will reduce the predictive ability of the model, but more importantly it will miss important information that ties closely to product choice.

Secondly, depending on the target market, different levels of the chain should be stressed in advertising strategy. Values are especially influential for heavy users, but light users or trial users are more attuned to physical attributes. Evidently as usage increases consumers' A/C/V chains evolve from being primarily attribute driven to value driven.

REFERENCES

Alba, Joseph and J. Wesley Hutchinson (1987), "Dimensions of Consumer Expertise," Journal of Consumer Research, 13 (March), 411-454.

Cooper, Lee G. (1973), "A Multivariate Investigation of Preferences," Multivariate Behavioral Research. 8, 253-272.

Gutman, J. (1982), "A Means-End Chain Model Based on Consumer Categorization Processes," Journal of Marketing, 46 (Spring), 65-72.

Gutman, J. and T. J. Reynolds (1979), "An Investigation of the Levels of Cognitive Abstraction Utilized by Consumers in Product Differentiation," in Attitude Research Under the Sun. ed. J. Eighmey, Chicago: American Marketing Association.

Jolly, J., T. J. Reynolds, and J. W. Slocum, Jr. (in press) "Application of the Means-End Theoretic for Understanding the Cognitive Bases of Performance Appraisal," Organizational Behavior and Human Performance

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

Reynolds, T. J. (1985), "Implications for Value Research: A Macro Vs. Micro Perspective," Psychology and Marketing, 2 (4),297-305.

Reynolds, T. J., and J. Gutman (1984), "Advertising is Image Management," Journal of Advertising Research. 24 (1), 27-36.

Reynolds, T. J., J. Gutman and J. Fiedler (1984), "Understanding Consumer's Cognitive Structures: The Relationship of Level of Abstraction to Judgements of Psychological Distance and Preference," in Psychological Processes of Advertising Effects: Theory, Research, and Application. eds. A Mitchell and L. Alwitt, Erlbaum.

Reynolds, T. J. and L. Jamieson (1984), "Image Representations: An Analytical Framework," in Perceived Qualities of Products, Services, and Stores. eds. J. Jacoby and J. Olson, Lexington, Ma: Lexington Books. 115-138.

Reynolds, T. J. and W. S. Perkins (1986), "Cognitive Differentiation Analysis: A New Methodology for Assessing the Validity of Means- End Hierarchies" in Advances in Consumer Research Volume 14, eds. Paul Anderson and Melanie Wallendorf, Provo, UT: Association for Consumer Research.

Reynolds, T. J. and K. Sutrick (1986), "Assessing the Correspondence of One or More Vectors to a Symmetric Matrix Using Ordinal Regression," Psychometrika, 51 (1), 101-112.

Reynolds, T. J., D. Weeks and W. S. Perkins (in press), "CDASCAL: An Algorithm for Assessing the Correspondence of One or More Vectors to a Symmetric Matrix Using Ordinal Regression," Psychometrika.

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