On the Locus of Range Effects in Judgment and Choice

ABSTRACT - Existing research suggests that the range of alternatives immediately available in choice and numerical judgment tasks can have a systematic effect on the obtained results. In many cases these effects have been shown to be mediated primarily by response language processes. The research reported here specifically tests for effects that are not due to response language. Moreover, the data suggest that the obtained effects do not result from discrimination processes, but from changes in psychological magnitude and they are limited to individuals who have low Product/attribute familiarity.

Citation:

J. Wesley Hutchinson (1983) ,"On the Locus of Range Effects in 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: 305-308.

Advances in Consumer Research Volume 10, 1983      Pages 305-308

ON THE LOCUS OF RANGE EFFECTS IN JUDGMENT AND CHOICE

J. Wesley Hutchinson, University of Florida

ABSTRACT -

Existing research suggests that the range of alternatives immediately available in choice and numerical judgment tasks can have a systematic effect on the obtained results. In many cases these effects have been shown to be mediated primarily by response language processes. The research reported here specifically tests for effects that are not due to response language. Moreover, the data suggest that the obtained effects do not result from discrimination processes, but from changes in psychological magnitude and they are limited to individuals who have low Product/attribute familiarity.

INTRODUCTION

In general, there is widespread agreement that the range of stimulus values may affect performance in judgment and choice tasks. In particular, differences between stimuli with respect to a given attribute are "smaller" when the stimulus range for that attribute is large. However, there is little agreement as to the exact nature of these effects or the mechanism(s) through which they operate. It is a matter of some controversy whether, or not, the stimulus range affects anything other than the rule that maps subjective impressions into the response language or the task at hand. This paper reports research that has been designed to specifically address this issue of "locus of effect."

Background

Over fifty years ago L. L. Thurstone made a point of the fact that the discriminal processes involved in the representation of stimuli may be quite independent of the nature of the psychophysical function (Thurstone, 1927) Simply put, the former was a matter of variance, the latter was a matter of central tendency. Interestingly, two of the principal streams of research on the effects of context in psychophysical judgment may be characterized as having concentrated on one or the other of these two factors. One stream is modern sensory psychophysics (Durlach & Braida, 1969; Gravetter & Lockhead, 1973; Luce, 1977). The impetus for this stream derived mainly from early research on channel capacity (e.g., Miller, 1956). From this perspective, contextual effects are systematically related to the nature of noise in the sensory system. The other stream is that of categorical judgment and magnitude estimation (Parducci, 1965; Birnbaum, 1982; For want or a better term I will refer to this area as numerical judgment theory. These are rather general models of psychological measurement based on racing scale responses. As such they have concentrated on the effects or context on the nature of the psychophysical function. Moreover, they have not limited their investigation to sensory events, but have often dealt with the more abstract stimuli of cognitive and social psychology. The details of these approaches are too elaborate to present here; however, they are implicit in what follows. The framework outlined below attempts to combine the two perspectives in order to address the issue of "where" in the flow of information processing stimulus context exerts its effect.

Most models of Judgment and choice assume, or at least allow, that there are four distinct stages of processing. The corresponding functions or these stages are (1) the encoding or stimuli, (2) the representation of stimuli in terms of relevant attributes, (3) the evaluation of stimuli as some function of attribute values, and finally (4) the selection of a response corresponding to the result of the evaluation process. Sensory psychophysicists have concentrated on the distinction between the first stage and the stages that follow. Most of these researchers have concluded that the stimulus range affects the latter stages. Numerical judgment theorists have concentrated on the distinction between the last stage and chose that precede it. Much of the evidence indicates that context exerts its effects on response selection; however, under some conditions the earlier stages also appear to be affected (Birnbaum, 1982). Thus, there is some consensus that the stimulus range does not affect encoding, that it does affect response selection and that whether, or not, it affects representation and/or evaluation is still an open question.

There is less consensus about the mechanism of these effects. In particular it is not known whether the stimulus range affects discrimination or psychological magnitude, or both. In this paper, discrimination effects refer to chose effects that result from changes in the variance of the internal representation, or from the fact that internal scale values are normalized with respect to the variance (see Figure 1 for an illustration of this latter effect). Psychological magnitude effects refer to those effects that result from changes in scale values in the representation or changes in attribute importance weights that are part of the evaluation function. Thus, the question of locus for range effects includes determining whether the representation and evaluation stages of processing are affected; and, if they are, whether the effect is mediated by discrimination or psychological-magnitude.-

FIGURE 1

ILLUSTRATION OF RANGE EFFECTS

AN EXPERIMENT ON PRICE ESTIMATION AND PURCHASE LIKELIHOODS

Design

A simple approach to this problem would be to observe means and variances of a response scale that is not sensitive to the stimulus range. That is, use a response scale that is absolute, rather than relative. The prices or automobiles, in dollars, would be such a scale. There are two problems that severely reduce the viability of this approach. First, although subjects are told to give responses in real dollars, they may adopt a strategy that all cars fall into a given price range, and then use that price range as if it were a relative scale similar to those of categorical judgment. Second, since most reported occurrences of range effects have required a between subjects design, all individual differences in the response scale mapping will inflate the observed response variance. Preliminary studies indicated that these were serious problems; therefore, an alternative approach was developed.

Instead of the means and variances associated with particular responses, consider the correlation of response values with real prices for each subject. Assume that only items common to all subjects are entered into these computations and the context items that set the range are excluded. This measure, of course, is insensitive to any linear transformation of the response values, including those due to response language and individual differences. On the other hand, this measure is sensitive to increases in.internal variance. Since this is a single attribute task, we can assume, operationally, that no evaluation is involved and any differences in these correlations must be due to discrimination factors affecting representation processes.

Now consider a multi-attribute task like ranking the automobiles according to the likelihood of purchase. How should stimulus range affect the correlation or price estimates and purchase likelihood ranks for the non-context items? It is possible that a discrimination effect could lead to lower correlations when the range is large owing to regression. This assumes that subjects draw independent samples of their price representation for the estimation and the purchase likelihood tasks. If these casks are performed consecutively, this seems unlikely; and, in any case, a range effect for these correlations is predicted only when an effect is also found for the first measure (i.e., correlations of price estimates and real values). On the other hand, if stimulus range affects psychological magnitude or the importance or price in computing purchase likelihood, lower correlations are predicted when the range is large, regardless of whether, or not, there was an effect on the accuracy or the price estimates. This prediction entails the assumption that price is not the only attribute that is considered for determining purchase likelihood and it is not highly correlated with the other attributes under consideration.

Method

A questionnaire was construct a according to the design described above. There were three parts to the questionnaire. First, respondents were instructed to read the names or twenty cars and then give their best estimate of the price of the least expensive version of the model named. Half or the respondents were instructed to give real dollar amounts as responses and half were instructed to use a nine point scale. Both groups were given an end anchor or the lowest priced car in America (i.e., Datsun '10). The stimulus set included fifteen low to moderate priced cars that were common to all conditions and five context cars that were either moderate (i.e., narrow range condition) or expensive (i.e., spread range condition) in price. Thus, overall, there was a 2 x 2 between subjects design with factors of Response Type and Stimulus Range. A list or the stimuli that were used is given in Table 1. After making their price estimates, respondents were instructed to turn the page and rank order the same twenty cars according to the likelihood that they would purchase that car within the next five years. Finally, respondents were asked several demographic and psychographic questions, including sex, Whether, or- not, they had ever purchased a car and whether, or not, they planned to purchase a car in the next year.

TABLE 1

STIMULUS SETS

There were 285 respondents who participated as part of an introductory psychology course requirement. The questionnaire was part of a package of questionnaires, none of which dealt with cars or prices. The questionnaire required about fifteen minutes to complete. Stimulus order was random with the constraint that the first two items were context automobiles. Across subjects there were two such randomizations and each was counterbalanced with respect of order (i.e., four stimulus orders all together).

Results

As a manipulation check, mean price estimates for each common car were computed for each level of Response Type and Stimulus Range. Regression lines for the Spread Range on the Narrow Range were then computed for each Response Type. The results of this analysis are given in Table 9. The typical range effect should manifest itself by a regression slope that is less than 1 and a mean value for common items that is higher in the Narrow Range condition. As expected, this pattern obtained only for Relative Responses.

TABLE 2

REGRESSION PARAMETERS FOR PRICE ESTIMATE MEANS

Interestingly, for Absolute Responses, the mean price estimate was actually significantly greater in the Spread Range condition than in the Narrow Range condition, indicating a "assimilation" effect. This, however, is not particularly relevant to the hypotheses at hand and will not be discussed further.

In order to test the main hypotheses of this experiment, two data points were computed for every respondent: (1) a Fischer z transformation of the Pearson produce-moment correlation of price estimates for the fifteen common cars with their actual prices, and (2) a Fischer z transformation of the correlation of price estimates for the fifteen common cars with the purchase likelihood ranks for chose same cars (i.e., the cars were re-ranked with the context cars excluded). These measures will be referred to as Price and Rank, respectively.

An analysis of variance was performed for the factors of Stimulus Range, Response Type, Sex, Purchase Experience -(i.e., the respondent had previously purchased at least one car, new or used) and Purchase Intent (i.e., the respondent planned to purchase a car in the coming year). The only statistically significant main effects for price were Sex (F = 41.6, p = .0001) and Purchase Experience (F = 13.1, p = .0004). For Rank, Sex (F = 11.4, p = .0008), Purchase Experience (F = 23.3, p = .0001) and Stimulus Range (F = 4.39, p = .037) were significant. Table 3 lists means for all main effects for both dependent measures. There were no statistically significant interactions for Price. For Rank, two interactions were statistically significant. Figure 2 depicts the interaction of Stimulus Range and Sex for Price (F= 0.15) and for Rank (F = 5.02, p = 0.026). Figure 3 depicts the interaction of Stimulus Range, Response Type and Purchase Experience for Price (F = 0.39) and for Rank (F = 3.88, p = 0.050). For ease of interpretation, all reported means are inverse Fischer z transforms of the numbers used in the analysis.

TABLE 3

ANOVA MAIN EFFECTS

Discussion

If we assume that males and people who have previously purchased a car know more about the prices of cars than their counterparts, then it seems safe to conclude that the only discrimination effects that have been shown to occur at the level of representation are those of expertise. Clearly, there are no effects of range on the price estimates that result from discrimination processes; and it is impossible to tell from this data alone whether, or not, the differences in mean price estimates result from effects on psychological magnitude, or response language. The purchase likelihood data, on the other hand, evidence strong range effects in addition to effects of expertise. As was discussed above, these effects cannot be mere response language effects, but must occur at the level of representation or evaluation, or both. Moreover, the absence of discrimination-effects for price estimation implies that these effects are most likely mediated by psychological magnitude.

FIGURE 2

MEANS FOR THE INTERACTION OF STIMULUS RANGE AND SEX

FIGURE 3

MEANS FOR THE INTERACTION OF RESPONSE TYPE, PURCHASE EXPERIENCE AND STIMULUS RANGE

The story does not end here, however. The two-way and three-way interactions suggest a slightly different interpretation. Perhaps the expertise that mediates these effects is not product expertise, per se, but is product/ attribute expertise. In this case, that means knowledgeability about the prices of cars in terms of real dollars In essence, knowledge about real dollar values entails, at least implicitly, a very broad context (i.e., all things that can be purchased). This knowledge "counteracts" the effects of the immediate stimulus context. Thus, people who have recently made absolute judgments, even if those judgments are not veridical, have invoked a context that renders their representation and evaluations of price relatively insensitive to the stimulus range. People who have good knowledge about price are similarly insensitive to stimulus range, regardless of the type of judgment required for their price estimates. Under these assumptions a range effect will be obtained only for people who have low price knowledge and have made only relative judgments. This interpretation predicts a three-way interaction of Price Knowledge, Judgment Type and Stimulus Range, and somewhat smaller two-way interactions of Price Knowledge and Judgment Type with Stimulus Range and a main effect of Stimulus Range. This prediction is consistent with the obtained results, assuming that Sex and Purchase Experience are surrogates for Price Knowledge, with the exception that there is no interaction of Response Type and Stimulus Range. In order to test this interpretation more specifically, Sex and Purchase Experience were combined into a single Price Knowledge factor and an analysis of variance was performed. Only the main effects of Price Knowledge and Stimulus Range and the three-way interaction were statistically significant (F=12.75, F=5.03 and F=4.06, respectively). The two-way interaction of Price Knowledge and Stimulus Range was marginally significant (F=2.54, p < .10). The means from the three-way interaction are shown in Figure 4.

FIGURE 4

MEANS FOR THE INTERACTION OF RESPONSE TYPE, PRICE KNOWLEDGE AND STIMULUS RANGE

It is clear that the only condition for which there is a striking range effect is when low Price Knowledge is combined with Relative Response. If this segment of the data is removed from the analysis, then, outside of Price Knowledge (F=6.84), none of the discussed main effects or interactions are statistically significant.

CONCLUSIONS

The results discussed above indicate that stimulus range can affect stages of processing other than response language. In particular, the impact of differences in price on purchase likelihoods was shown to be less when the price range was large. The exact conditions under which range effects of this sort will be obtained are not completely specifiable from these data. It is possible that there is a global range effect that is complicated by interactions with factors related to familiarity with the product and the response scale. On the other hand, it is possible that the effect only occurs when knowledgeability about the product and the response scale are low. The analyses discussed above suggest that the latter is the case, but strong conclusions must await future experiments.

Since tests for evidence that stimulus range can affect the discriminability of the representation failed, it is likely that the range effects that were demonstrated were mediated by changes in psychological magnitude. Whether these changes occurred at the level of representation or evaluation could not be determined from these data. Future research along these lines should (1) replicate these effects across different attributes and product categories, (2) clarify the role that expertise and response scale play in range effects and (3) develop models and procedural paradigms that allow the magnitude of these effects to be parameterized for each stage of processing.

REFERENCES

Birnbaum, M.H. (1982), "Controversies in Psychological Measurement," in B. Wegener (Ed.) Social Attitudes and Psychophysical Measurement, Hillsdale, N.J.: Lawrence Erlbaum Associates, 1982, 401-485.

Durlach, N.I. & Braida, L.D. (1972), "Intensity Perception. 1. Preliminary Theory of Intensity Perception," Journal of the Acoustical Society of America, 51, 377-383.

Gravetter , F. & Lockhead, G.R. (1973), "Criterial Range as a Frame of Reference for Stimulus Judgment," Psychological Review, 80, 203-216.

Luce, R.D. (1977), "Thurstone's Discriminal Processes Fifty Years Later," Psychometrika, 42, 461-489.

Miller, G.A. (1956),"The Magical Number Seven Plus or Minus Two: Some Limits on Our Capacity to Process Information," Psychological Review, 63, 81-97.

Parducci, A. (1965), "A Range-Frequency Model," Psycho- logical Review, 72, 407-418.

Thurstone, L.L. (1927), "Psychophysical Analysis," American Journal of Psychology, 38, 368-389.

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