A Framework For Exploring Context Effects on Consumer Judgment and Choice

ABSTRACT - Judgments of a stimulus may be affected not only by the characteristics of the stimulus itself, but also by the other stimuli that accompany it in the judgment context. This paper provides a framework for exploring context effects on consumer judgment and choice processes. First, an outline of the judgment process is used to identify the stages where context effects may occur. Next, the literatures in experimental and social psychology, behavioral decision theory and consumer research are selectively reviewed for evidence regarding context -effects on judgments of both uni-dimensional and multidimensional stimuli. Finally, the findings are integrated to identify the situations and processes contingent upon which context may influence consumer judgments and choice.

Citation:

Dipankar Chakravarti and John G. Lynch, Jr. (1983) ,"A Framework For Exploring Context Effects on Consumer Judgment and Choice", in NA - Advances in Consumer Research Volume 10, eds. Richard P. Bagozzi and Alice M. Tybout, Ann Abor, MI : Association for Consumer Research, Pages: 289-297.

Advances in Consumer Research Volume 10, 1983      Pages 289-297

A FRAMEWORK FOR EXPLORING CONTEXT EFFECTS ON CONSUMER JUDGMENT AND CHOICE

Dipankar Chakravarti, University of Florida

John G. Lynch, Jr., University of Florida

[Order of authorship is alphabetical and both authors contributed equally to writing this paper.]

ABSTRACT -

Judgments of a stimulus may be affected not only by the characteristics of the stimulus itself, but also by the other stimuli that accompany it in the judgment context. This paper provides a framework for exploring context effects on consumer judgment and choice processes. First, an outline of the judgment process is used to identify the stages where context effects may occur. Next, the literatures in experimental and social psychology, behavioral decision theory and consumer research are selectively reviewed for evidence regarding context -effects on judgments of both uni-dimensional and multidimensional stimuli. Finally, the findings are integrated to identify the situations and processes contingent upon which context may influence consumer judgments and choice.

INTRODUCTION

Consumer decision making involves many different types of judgments. For example, a consumer examining automobiles in a dealer's showroom may be choosing among the cars in the showroom, ranking a particular car in a preference rank order of a set of cars, judging how similar a particular car is to another, assessing how "economical" a particular car is, evaluating the overall "desirability" of the car and so on. In many situations, these judgments may be held private whereas in others the consumer may be required to express them overtly (e.g., on an 110 scale of "desirability" provided by a researcher).

Like many other judgments, consumers' judgments of a marketing stimulus are usually made in the presence of other stimuli. These other stimuli that are present or judged simultaneously define the "context" of the judgment. Often, the same stimulus may be judged in varying contexts. For example, a particular car in the dealer's showroom may be judged in the context of other cars in the showroom and/or cars that the consumer may have previously owned or examined. One may hypothesize that the context in which the car is judged may influence the judgment in some way.

FIGURE 1

A GENERAL OUTLINE OF THE JUDGMENT PROCESS

This paper deals with "context effects" in judgment and choice. The term refers to changes in the judgment process or its outcome as a function of the other stimuli to be judged. First, the paper presents an outline of the judgment process and identifies the stages where context effects may occur. Next, the paper selectively reviews the literatures in experimental and social psychology, behavioral decision theory and consumer research to examine the available evidence regarding context effects in judgment and choice. Judgments of uni-dimensional and multi-dimensional stimuli are discussed separately because of differences in the research methodologies used t-o identify the nature and locus of context effects in the two instances. Finally, the paper integrates the findings to identify the situations and processes contingent upon which context may influence consumer judgements and choice. Future research ties are also discussed briefly.

A TYPOLOGY OF CONTEXT EFFECTS IN JUDGMENT

The Judgment Process

Figure 1 shows an outline of the judgment process similar to those given by Birnbaum (1974, 1982). The outline is fairly general and can accommodate many different types of judgment. For this discussion, the scenario could be a consumer evaluating the "desirability" of a car in a dealer's showroom and required to overtly express this judgment on an 1-10 rating scale in a questionnaire study.

A large number of cues (f_i, f_j, f_k, f_l, etc), could potentially enter the consumer's Judgment. These cues may be retrieved from long term memory or externally available (e.g., from visual inspection of the car, information on the window-sticker, etc.). Only a subset of these cues, say {f_i, f_j}, may be attended to and mentally encoded or represented as {f_i, f_j}. Thus, fi might be information on a car's window sticker saying that it delivers 31 mpg, and fi might be an encoded rePresentation of this, e.g., "the car is fairly fuel-efficient". From the semantic representation, f, of each cue, the consumer abstracts its implication or scale value, s, for the judgment to be made. Thus, s might be the part-utility of this level of fuel-efficiency. Other cues would be similarly transformed (e.g., f_j -> f_j -> s_j). If the original input cues about the stimulus were objectively quantifiable, each translation process could be viewed as a separate "psychophysical" function H. i.e., s_i = H_i (f_i), s_j = H_j (f_j), etc.

The scale values associated with the various encoded stimulus inputs are integrated using some combination function, C, to form a private integrated impression, Y_ij, of the stimulus, i.e., Y_ij = C(s_i, s_j). In the example, Yij would be a private evaluation of the car's desirability. The overt rating of the car, R_ij, would be related to Y_ij by some monotonic "judgment function", J, i.e., R_ij = EQUATION. The judgment function may be viewed as merely translating the unobservable integrated impression into an overt response, e.g., a rating of the car-on a 1-10 "desirability" scale.

A Typology of Context Effects

Consumers' overt judgments (e.g., the "desirability" rating of the car), could be influenced by context in at least three potential ways. First, context may influence which aspects of a stimulus are selected for processing. Second, context may influence the scale values, i.e., the subjective, encoded representations of the stimulus cues. Finally, context may influence the judgment function that translates private evaluations to overt ratings.

Selection of Aspects for Processing: Context could influence which aspects of the stimulus are processed. In terms of Figure 1, such a selection could potentially occur (a) when aspects are selected for "encoded representation" and (b) in the combination process, Y_ij = C(s_i, s_j) that generates the integrated impression.

First, during the selection of aspects for encoded representation, context may selectively direct attention to some aspects of the car and not to others or it may selectively facilitate recall of some information from long term memory and inhibit recall of other information. For instance, context may activate a particular schema. This could cause the consumer to direct attention to schema-relevant features or to ignore and edit out schema-irrelevant features (Beattie -1982, cf. Alba and Hasher 1989). Alternatively, the activated schema could Cause the consumer to retrieve certain information from long term memory. This information would then be included in the encoded representation whereas it may not have been otherwise (Gilovich 1981).

Second, context could influence the combination process in a variety of ways. For instance, the processing agenda may change as a function of context. Contexts differing in the number of stimuli, number of features and the distribution of stimulus values on these features could differentially encourage the use of object-based versus attribute-based (e.g., lexicographic) decision rules. Also, context may influence the order in which features are examined in processing rules such as "elimination by aspects" (Tversky 1972). Alternatively, the context may influence the combination process if it caused a consumer to ignore some represented attributes of the stimulus and include others while making the over all judgment. If a compensatory combination process were involved, this could be viewed as affecting the weights" of various "salient" attributes.

Effects on Scale Values: A second type of context effect involves changes in the stimulus representation that occur because the location of a represented aspect along the dimension of judgment (e.g., s_i = H_i (f_i) has actually changed. Thus, context could change one's perception of a particular feature of a car (e.g., of how fast it is). For instance, if the input information described how many seconds it took the car to accelerate from O to 60 mph, context could potentially change the relationship between the numerical information and subjective "fastness". Alternatively, context could change the relationship between perceptions (of "fastness") and scale value (e.g., the part utility indicating how good or bat it is for a car to be that fast). As shown in Figure 1, both these types of effects are located in the realm denoted by the "psychophysical" function, H.

Effects on the Judgment Function: A third way in which context might alter one's overt rating of a stimulus is by changing the judgment function, R_ij = J(Y_ij) that translates the private overall evaluation into a rating on a (1 to 10) scale. Even if context does not affect the overall subjective representation of the car, w , in any of the ways described previously, it could influence how cars of specific levels of desirability are assigned specific ratings (e.g., 3,4,5 etc.). Thus, in terms of Torgerson's (1958) "law of categorical judgment", based upon Thurstone's work, context could alter the "category boundaries" that define how a rating scale is to be used without changing the subjective values of items to be rated. A familiar example of such an effect of context occurs in situations where an instructor alters the grade he or she assigns a student's essay, because of the relative quality of the essays written by other students in the class, without altering his or her private opinion of how good or bad the essay is.

This last type of context effect is confined to the judgment function and is termed a change in "response language" (Wyer 1974). It should be contrasted with the two preceding types of context effects which are "representational" in their nature and locus. Thus, a change in response language does not reflect a change in the subjective representation of the stimulus as a whole or of its represented parts,whereas context effects on selection of aspects and upon scale values influence the stimulus as subjectively experienced.

The above typology of context effects and the framework shown in Figure 1 is used in the rest of this paper to organize and interpret the empirical evidence on context effects in judgment and choice. In doing this, it is important to distinguish between judgments of uni-dimensional versus multi-dimensional stimuli, since the judgment processes and the potential loci of context effects in these two stimulus domains are somewhat different.

CONTEXT EFFECTS ON UNIDIMENSIONAL JUDGMENTS

Overview

Certain stimuli may vary or be described to judges only in terms of a single feature. For instance, subjects may be asked to judge the heaviness of lifted objects of varying weight but equal volume. Because the stimuli are uni-dimensional, subjects presumably would attend to the one dimension on which the stimuli vary. Hence, selection of aspects for "encoded representation" is not a relevant locus of context effects. Also, integration of information on diverse dimensions is not necessary. Hence, context effects on the "weights" of various stimulus dimensions in the combination function are also not relevant. Thus, for such uni-dimensional judgments, "representational" context effects could occur only as a function of changes in scale value.

Other "unidimensional" judgments involve complex stimuli that vary in terms of several features retrieved from long term memory. However, the stimuli are described to subjects only in terms of overall labels and the researcher cannot determine exactly what subjective features are being processed. For instance, a subject may be asked to judge how liberal or conservative various opinion statements are, e.g., "Government workers should be forced to retire by the age of 65." While the subject's response may be based on an integration of features retrieved from long term memory, the researcher has no means of distinguishing between the various types of "representational" context effects, i.e., between context effects on selection of aspects for "encoded representation", on the "scale value" of one or more aspects, or on the "weights" assigned to different aspects in some integration process.

For these reasons, context effects in judgments of unidimensional stimuli may be classified only broadly as to whether the effects are "representational" in nature or are confined to effects on the "response language" (i.e., the judgment function). Whereas a more fine-grained classification of the various types of "representational" context effects is possible for multi-dimensional stimuli, uni-dimensional judgments may be studied in the less elaborate framework shown in Figure 2.

FIGURE 2

JUDGMENTS OF UNI-DIMENSIONAL STIMULI

Consider a subject asked to overtly rate the desirability of a set of cars described only in terms of gas mileage (in mpg). Numerical gas mileage information, fi, is encoded and its implication, si, for the judgment of desirability is abstracted in accordance with some "psychophysical" function, s_i = H(f_i). Since there are no other stimulus dimensions, no combination process is necessary. The overt rating of desirability, R_i, is related to the underlying scale value, by some monotonic judgment function, R_i = J(s_i).

Most research on context effects in judgments of unidimensional stimuli comes from the social judgment literature. Across a variety of settings and stimulus domains, it has been shown that judgments of a stimulus (e.g., a person or object) are affected not only by the characteristics of the stimulus itself but also by other stimuli present in the judgment context. "Contrast effects" are probably the most well-documented type of contextual effect in this literature. Thus, when judges rate a series of stimuli along a given dimension, ratings of a target stimulus are related inversely to the objective values of the context stimuli that are judged concurrently (Manis and Armstrong 1971, Ostrom and Upshaw 1968, Sherif and Hovland 1961, Sherif, Sherif and Nebergall 1965).

While context effects, and in particular, contrast effects, are well-documented, there has been some controversy about their interpretation. The debate concerns whether context truly changes what one thinks about an object, or only how one overtly reports one's context invariant thoughts on some externally provided scale.

In other words, the issue is whether context changes the representation (i.e., the scale value, s_i, in terms of Figure 2) or whether the effect is confined to the "response language" used (i.e., the judgment function, J).

"Representational" and "Response Language" Accounts

The difference between "representational" and "response language" accounts is illustrated in Figure 3 in terms of a contextual phenomenon called "range effects". Range effects refer to variations in judgments of a set of core stimuli as a function of the range of physical values or other stimuli present in the judgment context.

FIGURE 3

REPRESENTATIONAL AND RESPONSE LANGUAGE INTERPRETATIONS

Consider two core stimuli, A and B, to be judged along a given dimension in two contexts. In a baseline situation, A and B are judged together with two context stimuli, X and Y, which have physical values slightly lower and slightly higher respectively than A and B. In a second context, the stimuli to be judged are the same, except that Y is replaced by Z, which has a markedly higher physical value. Thus, the second context differs from the first in that the physical values of the stimuli cover a wider range.

The baseline condition is shown in Figure 3 (Panel A). The physical values, fi, of the stimuli are shown on an arbitrarily numbered scale of physical values. For example, if the stimuli were cars, the f_i could be numerical gas mileage information (in mpg). The scale values, s_i, are also shown on an arbitrarily numbered scale and could represent the part-utilities of the various levels of gas mileage. The overt ratings, R_i, are shown on a scale going from -3 to +3. These ratings are akin to judgments of the cars' fuel-efficiency on experimenter-provided scales in a questionnaire study.

In such a study, the overt ratings, R_i, of the stimuli may be affected by the judgment context. Consistent with previous contrast effect findings, A and B should be rated lower, on average, in the wide range condition relative to the baseline condition. Also, the differences between the ratings of A and B should be less in the wide range condition relative to the baseline condition. Thus, in terms of the example, cars that deliver, say 25 and 30 mpg, should be rated as (a) less fuel-efficient and (b) more similar on fuel-efficiency in the presence of a car delivering 50 mpg versus in the presence of a car delivering a less extreme gas mileage, san 35 mpg.

Panel B (Figure 3) gives the "representational" account of these observed context effects on the ratings, R_i. The result is attributed to context-induced changes in the "psychophysical" function, s_i = H(f_i). According to this interpretation, the locations of A and B on the psychological continuum, (s_" and s_B), actually change when the range of physical values of the context stimuli is extended. In the wide range condition, these psychological scale values are both lower and closer together than their counterparts in the baseline condition. The mapping, R_i = J(s_i), between the psychological scale values and overt numerical ratings remains unaffected by the range of the context stimuli.

Panel C (Figure 3) shows the "response language" interpretation. According to this view, context does not affect the subjective representations, (s_" and s_B), of the stimuli. However, when asked to report his or her perceptions of the stimuli, the subject assigns the highest possible rating (+3 to the highest stimulus (Y or Z), and assigns the lowest rating (-3) to the lowest stimulus (X). The remaining stimuli are rated in a manner reflecting their positions on the subjective continuum relative to s_X and s_Y or s_Z.

Empirical Evidence

Empirical evidence to date tends to provide more support for response language explanations of context effects. Researchers have reasoned that if this interpretation is correct, context effects on ratings should be relatively isolated phenomena, i.e., they should not be manifested in behavior, nor should they impact on perceptions of the stimulus along conceptually related dimensions. For example, Upshaw (1962, 1965, 1969) found that subjects' prior attitudes affected their ratings of attitude-related statements in a Thurstone-type procedure for scaling items to be included in an attitude questionnaire. However, Upshaw argued that subjects with differing initial attitudes rated a given opinion statement differently not because they perceived the position of the statement along the attitude continuum to be different. Rather, the difference was due to subjects using their own attitudes to define one endpoint of the response scale when their attitudes were more extreme than the attitudes implied by any of the opinion statements being rated. In a particularly elegant study, Upshaw (1978) generated context effects by manipulating the way in which subjects anchored one of a possible set of congeneric attitude scales. The observed contextual effect remained confined only to the measure that was explicitly manipulated and did not appear in any other attitude measure. Since a true representational change in attitude should have manifested itself in the responses to the broader set of congeneric attitude scales, Upshaw concluded that the observed effects were due to changes in response language.

Parducci's range-frequency theory of category judgment (Parducci 1965, Parducci and Perret 1971) suggests processes similar to those hypothesized by Upshaw. Parducci argued that category judgments represent a compromise between two principles. First, the range principle asserts that numerical ratings, R_i, of a set of stimuli should be linearly related to underlying subjective perceptions, s_i. The nature of this linear relationship is determined by a process whereby the rater assigns the lowest possible category rating to the "lowest" stimulus and assigns the highest possible rating to the "highest' stimulus. Differences among the ratings of the remaining stimuli are proportional to the differences between their scale values, s.. Second, the frequency principle holds that the rating of a stimulus is directly proportional to its ordinal position in the set. It is as though the subject wishes to use each rating scale category with equal frequency. (See Birnbaum 1974 for a more general interpretation of the frequency principle). According to range-frequency theory, the final category judgment is a weighted average of the judgments implied by these two principles.

While in his less technical work Parducci (1969) has implied that context effects on ratings may be representational in nature, his position is not explicit. Other authors (e.g., Kanouse and Hanson 1979) have speculated that range and frequency effects might be given a representational interpretation. However, Birnbaum (1974) has shown that frequency effects, at least, are mediated solely by changes in response language, i.e., by changes in the judgment function, J. Using a functional measurement analysis, Birnbaum demonstrated the context-invariance of the implied psychophysical function. His data also systematically violated the predictions of Helson's adaptation level theory (1964), perhaps one of the best-known representational accounts of context effects on ratings. As previously noted, Upshaw interpreted effects closely related to range effects in terms of response language.

Despite the evidence supporting response language explanations, the belief persists that representational context effects occur, but have not yet been experimentally detected. Intuitively, one may recall experiences in which context affected not just what one said but also how one thought about and behaved toward a person or object. Sherman, Ahlms, Berman and Lynn (1978) have shown that under specific circumstances contrast effects on ratings may be accomPanied by corresponding effects on behavior. Subjects judged the importance of a social issue in the context of either a set of important or unimportant issues. As was expected, the target issue was rated as more important in the latter context. The researchers found that contrast effects on ratings were explainable in terms of response language in that they did not carry over to issue-related behavior if the rating was not salient at the time of the decision to perform the behavior. However, if the rating itself was made salient, it served as an informational input that induced corresponding changes in overt action.

CONTEXT EFFECTS ON MULTI-DIMENSIONAL JUDGMENTS

As discussed, the weight of the evidence regarding context induced variations in the ratings of uni-dimensional stimuli seems to support a response language interpretation. In a sense though, uni-dimensional judgment tasKs in which both the psychophysical function, H, and the judgment function, J, are monotonic make it difficult to conclusively demonstrate that context effects on ratings are representational in nature. As long as the rank order implicit in the ratings of a set of core stimuli is preserved, it is always possible to interpret context effects as due to a change in the monotonic judgment function. J. However. if changes in the rank order do occur, interpretation of the results as due to changes in response language would require the implausible assumption of a non-monotonic judgment function. Since the monotonicity of J is a very general and non-restrictive assumption, it would seem far more reasonable to explain such a result in terms of a change in the subjective representations of the core stimuli.

Experiments that ask subjects to make integrated judgments based upon two or more informational inputs may provide a better opportunity to conclusively demonstrate representational context effects. First, it is more plausible to expect contextually induced changes in the rank order of such stimuli. (See Chakravarti and Lynch 1982, for an illustration.) Second, in judgments of multi-dimensional stimuli there are three ways in which a change in representation may occur: (a) by changing which aspects are selected for representation; (b) by changing the scale values of the represented inputs and (c) by changing the way in which the scale values are combined to make integrated judgments. Only the second possibility exists in judgments of uni-dimensional stimuli.

Evidence From Experimental and Social Psychology

Empirical research on context effects in judgments of multi-dimensional stimuli may be discussed in terms of the previously presented framework in Figure 1. Apart from response language effects, context effects of a representational nature could occur for at least three reasons: (a) a change in the aspects selected for encoded representation; (b) a change in the scale values of the encoded aspects; and (c) a change in how the scale values of diverse aspects are weighted in some combination process.

Three studies by Mellers and Birnbaum (1982a, 1982b, Mellers 1982) may be interpreted in terms of Figure 1. Mellers and Birnbaum (1982a) presented subjects with the task of assigning grades to students on the basis of two numerical test scores. A common (core) set of test score pairs was judged in several different contexts. Results showed that contextual effects due to the distribution of scores on the separate tests had minimal influence on subjective representations (scale values) of the scores (e. g., of what the scale value of a score of 70% would be). Since the distribution of context scores on each test did not cause any change in the rank order of grades assigned to test score pairs in the core set, Mellers and Birnbaum reasoned that the most parsimonious interpretation was that context effects influenced only the monotonic judgment function, R_ij = J(Y_ij), that translates private integrated impressions to overt ratings.

It should be noted that in the above study, information about the two test scores to be integrated was in naturally equivalent units. Mellers and Birnbaum (1982b) found that in a psychophysical task requiring cross-modality comparisons or combinations of stimuli (i.e., in judgments or the combined magnitudes of pairs comprised of circles that varied in size and dot patterns that varied in darkness), the distributions of stimulus values along two physical input dimensions did affect the subjective scale values derived for stimuli common to each context. Since the rank order of ratings of (circle & dot pattern) pairs in a core set changed systematically as a function of context, the results cannot be interpreted in terms of context-induced changes in response language.

Mellers (1982) reported a conceptually similar result in her research on judgments of salary equity. Subjects were given scatter diagrams displaying numerical merit ratings of faculty members in a given department along with their salaries, and were asked to judge how under-benefitted or over-benefitted each faculty member was relative to the rest of the department. A core set of descriptions of certain faculty members was judged in one of several contexts that varied the distributions of merit ratings and salaries within the department. The rank order of judged inequity in the core set changed systematically as a function of context. Mellers proposed that subjects judged a situation as equitable when the salary of each faculty member relative to the distribution of faculty salaries matched the merit rating of the faculty member relative to the distribution of faculty merits. Her research suggested that these relativistic within-mode judgments were well-described by range-frequency theory.

In summary, these studies show that in judgments of multi-dimensional stimuli, variations in context do not produce systematic rank order reversals when the dimensions are in equivalent units (Mellers and Birnbaum 1982a). However, when such judgments involve cross-modality comparisons or combinations, systematic rank order reversals are produced (Mellers and Birnbaum 1982b, Mellers 1982). For reasons stated earlier, these latter findings cannot be interpreted as response language changes, and in terms of Figure 1, would lie in the realm of representational context effects. However, the rather simple stimuli used in these studies minimized the likelihood that context effects on the selection of aspects for "encoded representation" played a significant role.

Evidence From Consumer Research

Chakravarti and Lynch (1982) investigated whether context effects influence the trade-offs implied by conjoint analysis. Two separate experiments used the same basic research strategy. A "core" set of descriptions of hypothetical subcompact cars was constructed by choosing a fractional replicate of a 2n design in which several attributes (e.g., gas mileage, price, warranty, etc.) were each varied over two levels. Several sets of context profiles were constructed by letting these same attributes take on different levels. All subjects rated the overall desirability of cars in the core set together with cars from one of the context sets. The context sets extended the range of a particular attribute. For instance, the two levels of gas mileage in the core set were 97 and 30 mpg, whereas the context set levels would be 32 and 35 mpg, 43 and 46 mpg or 51 and 54 mpg. The manipulated attributes were those said to be most important by Consumer Reports and by subjects in pilot studies. All subjects were pre-qualified on the basis of car ownership.

Chakravarti and Lynch were interested in how judgments of the cars in the core set were affected by the particular context set with which they were judged and especially in whether extending the range of an attribute (e.g., gas mileage) might reduce the subjective differences in utility between the two levels of gas mileage in the core set, while leaving the utility differences between the two levels of other attributes unchanged. In these circumstances, context could change the implied rank order of preferences among the core set options.

Results showed that while the raw ratings of the core stimuli were influenced by context, there were no significant changes in the rank order of overall evaluations of the core set options. In terms of Figure 1, the results were interpreted as showing that in a typical conjoint analysis task, context may change the judgment function, J. However, there was no evidence of any change in psychological representations (either in terms of overall evaluations or in terms of representations of separate attributes) of the cars described as a function of context.

Hutchinson (1983, this volume) presented subjects with the names of real brands of cars and asked them to judge the prices of those cars. A core set of relatively inexpensive cars was judged together with either a "narrow range ' context set comprised of mid-priced cars or a "wide range" context set of expensive cars. Half of the subjects in each group made absolute judgments of prices (in dollars) while the other half rated how expensive the cars were on a 1-9 scale. In each of the four range x judgment type conditions, Hutchinson classified the subjects as being high or low in experience with purchase of cars. After making the above respective judgments, all subjects estimated the likelihood that they would buy each car previously judged.

Without recounting the findings and the interpretive rationale in complex detail (and acknowledging that Hutchinson presents a somewhat different view), the results were basically as follows. When subjects were estimating prices of the cars, there were no apparent context effects on underlying representations of price. However, context effects on overt price estimates, interpretable as response language changes, did occur. These were most pronounced for subjects who had made rating scale (rather than absolute dollar) price estimates. In particular, consistent with prior contrast effect findings, cars in the core set were rated lower, as well as more similar, in price in the wide range versus the narrow range condition. (See Figure 3). These context effects did not carry over to subsequent purchase likelihood judgments with one exception. This exception occurred for the low purchase experience subjects who had previously made rating scale price estimates. Their purchase likelihood judgments differentiated less among the cars as a function of estimated price in the wide range condition relative to the narrow range condition. In contrast, for experienced subjects, the correlation between the purchase likelihood judgments and the estimated prices was not affected by context since context had not influenced the subjective representations or scale values of price. In summary, Hutchinson's results show that context effects of a representational nature affected the purchase likelihood judgments only when subjects were low in experience and had made a context-dependent rating scale judgment immediately prior to making the purchase likelihood judgment.

Huber, Payne and Puto (1982) report a fascinating effect of context on choice probabilities, rather than rating scale judgments. They found that the assumption of "regularity" of choice probabilities was violated by adding an "asymmetrically dominated" alternative to the choice set. "Regularity", a property assumed by virtually all models of probabilistic choice (Luce 1959, Tversky 1972), asserts that the probability of choosing a from a choice set {a, b} must be greater than or equal to the probability of choosing a from a choice set {a,b,c}. Huber et al found that for two dimensional options, if c was "asymmetrically dominated", i.e., dominated by a but not by b, the addition of c to the choice set actually increased the probability that a would be chosen. Since the element c, in such cases, is more similar to a than to b, and since previous theorizing (McFadden 1980, Tversky 19,9) postulated that new options should attract choices at the expense of more similar alternatives, the result is particularly surprising. For the present review, it should be noted that since context affected overt behavior, the effects cannot be interpreted in terms of response language.

AN INTERPRETIVE FRAMEWORK FOR CONTEXT EFFECTS

As previously noted, context effects in judgments of unidimensional stimuli have generally been viewed as due to response language changes. In such judgments, context effects on ratings were found to be a fairly isolated phenomenon and did not carry over to other judgments or to behavior. However, the inherently simple nature of the stimuli and the underlying judgment may have made it difficult to demonstrate contextual effects on the scale values or subjective representations of the stimuli.

FIGURE 4

LOCUS OF CONTEXT EFFECTS IN CROSS-MODALITY STIMULUS JUDGMENTS

The evidence regarding context effects in judgments of multi-dimensional stimuli is considerably more complex. Whereas context effects on the judgment function (i.e., response language changes) have been commonly noted, the evidence suggests that effects of a representational nature may or may not occur contingent upon the judgment situation. This section provides a framework in which the evidence may be interpreted and the phenomenon explored further.

Mellers and Birnbaum (1989a) found that in judgments of multi-dimensional stimuli,where the stimuli were in the same modality or in equivalent units, context produced no systematic reversals in the rank orders of the stimuli. However, reversals did occur in judgments that involved cross-modality comparisons or combinations (Mellers and Birnbaum 1982b, Mellers 1982). These findings suggest that in cross-modality comparisons or combinations, the stimuli must be judged within the context of stimuli in the same modality prior to comparison across modalities (Birnbaum 1982). This is as if two context invariant psychological values, s_i and s_j, that are in incommensurable units, are transformed into context dependent judgments, R_i and R_j. These implicit, intermediate numerical responses, rather than the scale values, s_i and s_j, are then combined to determine the private/integrated impression, Y_ij. This notion reconciles the findings of context invariant subjective scale values in uni-dimensional judgments and in combinatorial judgments of stimuli in the same modality with findings of context dependent derived scale values in cross-modality stimulus comparison or combination tasks.

By this explanation, context normally has no impact on the psychological representation (i.e., the scale value) of stimulus information along a given dimension. However, certain novel tasks may require the subject to integrate dimensions that cannot be readily compared, because (a) one or more of the informational dimensions is described in unfamiliar units or (b) the integration task requires the subject to make trade-offs that are unfamiliar. In such cases, it is as though an additional processing stage is interposed between evaluation of the scale values of informational inputs and integration of the information. (Compare Figures 1 and 4).

The hypothesis of such an additional processing stage may seem ad hoc, especially since the evidence reviewed makes it impossible to distinguish this explanation from the hypothesis that context changes the scale values, s_i and s_j. However, there are strong parallels between this explanation and the previously cited Sherman et al (1978) findings pertaining to contrast effects on ratings of the importance of a target issue (recycling). Recycling was rated as more important when judged with other trivial issues relative to when judged with issues of greater import. However, the same context manipulation did not have parallel effects on behavior, i.e., a decision to participate in a local recycling project. Therefore, Sherman et al concluded that the contrast effects on ratings of the importance of recycling were attributable to changes in response language rather than to changes in the underlying subjective representation of the importance of recycling. This is analogous to the Mellers and Birnbaum (1982a) findings that when multidimensional stimulus inputs (test scores) which did not need to be rated separately for combination (because they were in equivalent units) were judged, context did not change the underlying representation (scale values) of the test scores and thus did not affect the rank order of the ratings of the test score pairs.

However, Sherman et al (1978) also found that when subjects had rated the importance of recycling and the ratings were made salient at the time recycling behavior was measured, contrast effects on ratings, though based only on response language changes, carried over to issue-related behavior. It was as if once ratings were made, -they replaced the original scale values. The argument presented by the present authors is essentially similar, i.e., that implicit ratings of informational stimulus inputs replace scale values in cross-modality comparison and combination tasks that require unfamiliar trade-offs to be made

One other published study in the psychology literature seems to show context induced changes in the psychological representation of stimuli that cannot be attributed to changes in response language. Torgerson (1965) had subjects judge the similarity of kite shapes that varied in length and "bottom-heaviness". His results showed that the impact of an unit change in physical length (of the kites) in MDS solutions grew smaller as the range of lengths of kites to be compared was increased. Here too, the task was one in which the trade-offs required to make judgments of overall similarity were probably unfamiliar to most subjects. Thus, the psychology literature that was reviewed is consistent with the tentative generalizations offered in this paper.

The consumer research literature also provides a fairly consistent picture. The Chakravarti and Lynch (1982) experiments found no systematic rank order reversals as a function or manipulations of the range of attribute values in a conjoint analysis task using subcompact cars described on several attributes as the stimuli. However, context effects, consistent with the response language interpretation were in fact noted

It is important to note that the subjects in these studies were ore-qualified on the basis of car-ownership. Also, the attributes manipulated were chosen from those said to be most important by Consumer Reports and by subjects in pilot studies. Thus, even though the car attributes were in different objective units, subjects were probably quite familiar wit" how they traded off against each other to affect the subjective desirability of the cars. In other words, with familiar stimuli, the joint distribution of the stimuli in the two modalities may be well-learned and used to rate the stimuli. This would remove the need to judge the stimuli in their own modalities prior to comparison (cf. Mellers and Birnbaum 1982b, Birnbaum 1982). In either case, the results are quite consistent with the tentative empirical generalization offered previously about when context may or may not be expected to affect psychological representations in multi-dimensional stimulus judgments.

One may conjecture that subjects less knowledgeable about cars would not have been familiar with the tradeoffs required, just as subjects in the Mellers (1982) and Mellers and Birnbaum (1982b) studies were unfamiliar with the trade-offs required for the cross-modality comparison and combinations necessary in their tasks. The subjects may then have had to invent trade-off strategies analogous to those postulated to explain the results of the latter two studies. Though the mechanism responsible must be considered an open question, one may expect that Judgments made by less knowledgeable consumers might actually have shown context induced rank order reversals in the implied preferences of the core set cars in the Chakravarti and Lynch study.

The above, in essence, is exactly what Hutchinson (1983) found. His results showed that context effects of a representational nature affected purchase likelihood judgments only for low experience subjects who had made a context dependent rating scale judgment immediately prior to making the purchase likelihood judgment. In contrast, the correlation between the purchase likelihood judgments and the estimated prices was not affected by context for experienced subjects. The explanation offered here is that the more experienced subjects carried an enduring representation of the prices of the cars in long term memory. These representations were combined with other retrieved information to make the purchase likelihood judgments. In contrast, inexperienced consumers had no stable prior representation of the prices of the cars. Consequently, their price estimates were probably used as though these were the subjective scale values (see Figure 4) and were integrated with other information retrieved from long term memory to make the purchase likelihood judgments. Because rating scale (but not dollar) estimates of the prices of various core cars were more similar in the wide range versus the narrow range context, rated price had lower correlation with the purchase likelihood judgment than in the former condition. This is quite similar to Sherman et al's (1978) findings that context effects, though initially due only to response language changes, may carry over to subsequent--behavior if made salient at the time-of the behavior (purchase intentions, in the case of the Hutchinson study). Hutchinson's results, though, suggest that it is critical whether the person carries a strong prior representation of the object to be judged. If So? the prior representation, rather than the prior rating, is integrated with other information to affect behavior. Only inexperienced consumers will use their prior (response language driven) rating to guide subsequent action.

There is an interesting parallel between this notion and Bem's (1972) self-perception theory that holds that people will infer their attitudes from past behavior, but only when no strong prior representation of attitude exists. In the present case, past behavior (the overt estimate) becomes part of the mental representation of the car, but only if no strong prior representation of the car prices existed in long term memory.

Finally, Huber et al (1982), reported several studies t to explore whether the addition of an asymmetrically dominated alternative to the choice set changed perceptions of other elements in the set or changed some aspect of the choice process. In terms of Figure 1, the present authors believe that their results should not be interpreted as a change in the encoded representations of the alternative products or in the scale values, Si and sj. Rather, the data are perhaps best interpreted in terms of a contextually induced change in the way the context invariant, encoded information was combined to make a choice. In other words, the context effects were probably localized in the combination function, C(si, s ). This interpretation is based on their findings that while extending the range on one dimension (by adding an asymmetrically dominated alternative) decreased the apparent "importance" of that dimension, the effect was not sensitive to the degree to which the range was extended.

SUMMARY AND CONCLUSIONS

This paper has discussed context effects that may influence judgment and choice. First, context could affect the selection of aspects for processing in two ways. It could bias attentional focus on certain external information or cause selective retrieval of information from long term memory, influencing the contents of short term memory. Alternatively, it could influence how information in short term memory is combined to make a judgment or choice. Gilovich (1981) has reported the former type of effect and Huber et al's (1982) results may be interpreted as an instance of the latter- type of effect.

Second, context can influence category judgments by changing the response language consumers use to report their subjective judgments. Such effects are the most commonly observed (e.g., Birnbaum 1974, Chakravarti and Lynch 1982, Mellers and Birnbaum 1982a, Sherman et al 1978, Upshaw 1965, 1978). The effect apparently, however, occurs without a parallel effect on the underlying subjective impression of the stimulus. Hence, such effects are unlikely to be observed in or carried over to other judgments or behavior toward the stimulus.

The third type of context effect, though most frequently discussed (Helson 1964, Parducci 1969, Sherif and Hovland 1961, Sherif et al 1965), is the most difficult to demonstrate. This is a true change in the subjective location of the stimulus along the dimension of judgment, i.e., a change in scale value. Prior reviews (Wyer 1974, Upshaw 1978) have indicated that few studies have demonstrated "perceptual" or representational changes as a function of context

More recent research suggests two exceptions to the above. First, context effects have been shown to carry over to related behaviors (thus demonstrating a representational change) if a response language driven rating is made salient during a subsequent judgment. Apparently, this will occur only when the relative salience of this prior rating is greater than some enduring, context invariant prior representation (Hutchinson 1983). Experienced or knowledgeable judges may bring an extra-experimental and prior mental context to bear on the judgment. This may overwhelm any experimental manipulation of the context (see, e.g., Upshaw 1965). Second, in multidimensional judgments that require judges to make unfamiliar trade-offs, context can influence the effective scale values of the inputs (Torgerson 1965, Mellers 1982, .Mellers and Birnbaum 1982b). This paper has speculated that in such tasks, it is as if subjects implicitly rate the stimuli on each dimension to translate the diverse inputs into a common scale. These implicit ratings, which are susceptible to contextual changes in response language, become the effective "scale values" that are combined to make the overall judgment. Thus, context effects are observed on the judged rank orders of the multi-dimensional stimuli. Other studies (Chakravarti and Lynch 1982, Mellers and Birnbaum 1982a), show that such effects are not observed when the stimulus dimensions are in equivalent units or when the consumer is familiar with making trade-offs among the dimensions described in different units.

This review has highlighted some under-researched areas in the context effects literature. This is an important domain for consumer researchers in view of the growing interest in situational influences on consumer judgment and choice (Bettman 1979, Payne 1982). First, little work has been done on how context governs attentional and memory retrieval processes that determine which aspects of a stimulus will be processed (cf. Gilovich 1981). Second, how prior knowledge influences the sensitivity of judgments to context is of particular importance (Hutchinson 1983). Third, context effects on combination processes (e.g., decision rules and judgment heuristics), deserves careful scrutiny (Huber et al 1982). In particular, context effects on the processing agenda, e.g., the order in which dimensions are processed in choice strategies such as "elimination by aspects" (Tversky 1972) need to be examined. Finally, how context influences the results of measurement methodologies used for gauging consumer perceptions and preferences needs to be researched since these analyses usually serve as the bases for product design and positioning strategies. The Chakravarti and Lynch (1982) studies represent a effort in this direction

REFERENCES

Alba, J.W. and L. Hasher (1982), "Is Memory Schematic?" Psychological Bulletin, in press.

Bem, D.J. (1972b), "Self Perception Theory." in L. Berkowitz ed. Advances in Experimental Social Psychology, 6. New York, Academic Press, 1-62.

Bettman, J.R. (1979), "An Information Processing Theory of Consumer Choice.4>Reading, MA., Addisson Wesley.

Birnbaum, M.H. (1974), "Using Contextual Effects to Derive Psychophysical Scales." Perceptions and Psychophysics, 15, 1, 89-96.

Birnbaum, M.H. (1982), "Controversies in Psychological Measurement." in B. Wegener, ed. Social Attitudes and Psychophysical Measurement. Hillsdale,NJ., Earlbaum, 411-485.

Chakravarti, D. and J.G. Lynch, Jr. (1982), "Context Effects: Implications for Conjoint Analysis of Consumer Preferences." Working Paper, Center for Consumer Research, University of Florida.

Gilovich, T. (1981), "Seeing the Past in the Present: The Effect of Associations to Familiar Events on Judgments and Decisions." Journal of Personality and Social Psychology, 40, 797-808.

Helson, H. (1964),"Adaptation Level Theory." New York, NY. Harper and Row.

Huber, J., J. Payne and C. Puto (1982), "Adding Asymmetrically Dominated Alternatives: Violations of Regularity and the Similarity Hypothesis." Journal of Consumer Research, 9, 1, 90-98.

Hutchinson, J.W. (1983), "On the Locus of Range Effects in Judgment and Choice." in R. Bagozzi and A. Tybout, eds. Advances in Consumer Research, 10, Ann Arbor, MI, Association for Consumer Research. in press.

Kanouse, D.E. and L.R. Hanson (1972), "Negativity in Evaluation." in E.E. Jones et al eds. Attribution: Perceiving the Causes of Behavior, Morristown, NJ. General Learning Press.

Luce, R.D. (1959), "Individual Choice Behavior." New York. NY. John Wiley and Sons.

Manis, M. and G.W. Armstrong (1971), "Contrast Effects in Verbal Output." Journal of Experimental and Social Psychology. 7, 381-388.

McFadden, D. (1980), "Econometric Models for Probabilistic Choice Among Products." Journal of Business, 53, 513-529

Mellers, B.A. (1982), "Equity Judgment: A Revision of Aristotelian Views." Journal of Experimental Psychology, General, 111, 242- 271

Mellers, B.A. and M.H. Birnbaum (1982a), "Contextual Effects in Social Judgment." Journal of Experimental and Social Psychology, in press.

Mellers, B.A. and M.H. Birnbaum (1982b), "Loci of Contextual Effects in Judgment." Journal of Experimental Psychology: Human Perceptions and Performance, 8, 4, 582-601.

Ostrom, T.M. and H.S. Upshaw (1968), "Psychological Perspective and Attitude Change." in A. Greenwald, T. Brock and T. Ostrom eds. Psychological Foundations of Attitude New York NY. Academic Press.

Parducci, A. (1965), "Category Judgment: A Range-Frequency Model." Psychological Review, 72, 407-418.

Parducci, A. (1968), "The Relativism of Absolute Judgment." Scientific American, 219, 84-90.

Parducci, A. and L.F. Perrett (1971), "Category Rating Scales: Effects of Relative Spacing and Frequency of Stimulus Values." Journal of Experimental Psychology Monograph, 89, 427-452.

Payne, J.W. (1982), "Contingent Decision Behavior." Psychological Bulletin, 92, 382-402.

Sherif, M. and C.J. Hovland (1961), "Social Judgment: Assimilation and Contrast Effects in Communication and Attitude Change. New Haven, CT. YaLe University Press.

Sherif, C., M. Sherif and R.E. Nebergall (1965), "Attitude and Attitude Change: The Social Judgment Involvement Approach." Philadelphia, PA. Saunders.

Sherman, S.J., R. Ahlm, L. Berman and S. Lynn (1978), "Contrast Effects and Their Relationship to Subsequent Behavior." Journal of Experimental and Social Psychology, 14, 340-350.

Torgerson, W.S. (1958), "Theory and Methods of Scaling." New York, NY. John Wiley and Sons.

Torgerson, W.S. (1965), "Multi-dimensional Scaling of Similarity." Psychometrika, 30, 379-393.

Tversky, A. (1972), "Elimination by Aspects: A Theory of Choice." Psychological Review, 79, 281-299.

Upshaw, H.S. (1962), "Own Attitude as an Anchor in Equal-Appearing Intervals." Journal of Abnormal and Social Psychology, 64, 85-96.

Upshaw, H.S. (1965), "The Effects of Variable Perspectives on Judgments of Opinion Statements for Thurstone Scales: Equal Appearing Intervals." Journal of Personality and Social Psychology, 2, 60-69.

Upshaw, H.S. (1969), "The Personal Reference Scale: An Approach to Social Judgment." in L. Berkowitz, ed. Advances in Experimental Social Psychology, 4, New York, NY. Academic Press.

Upshaw, H.S. (1978), "Social Influence on Attitudes and on Anchoring of Congeneric Attitude Scales." Journal of Experimental Social PsychoLogy, 14, 327-339.

Wyer, R.S. (1974), "Cognitive Organization and Change: An Information Processing Approach." Potomac, MD. Lawrence Earlbaum Associates.

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