Assimilation and Contrast Effects in Attitude Measurement: an Inclusion/Exclusion Model

Norbert Schwarz, ZUMA
Herbert Bless, University of Mannheim
ABSTRACT - We present a theoretical model of context effects in attitude measurement that predicts (a) the conditions under which context effects are likely to emerge; (b) their direction (i.e., assimilation or contrast); (c) their size; (d) their generalization across related items; and (f) their dependency on the mode of data collection used.
[ to cite ]:
Norbert Schwarz and Herbert Bless (1992) ,"Assimilation and Contrast Effects in Attitude Measurement: an Inclusion/Exclusion Model", in NA - Advances in Consumer Research Volume 19, eds. John F. Sherry, Jr. and Brian Sternthal, Provo, UT : Association for Consumer Research, Pages: 72-77.

Advances in Consumer Research Volume 19, 1992      Pages 72-77

ASSIMILATION AND CONTRAST EFFECTS IN ATTITUDE MEASUREMENT: AN INCLUSION/EXCLUSION MODEL

Norbert Schwarz, ZUMA

Herbert Bless, University of Mannheim

ABSTRACT -

We present a theoretical model of context effects in attitude measurement that predicts (a) the conditions under which context effects are likely to emerge; (b) their direction (i.e., assimilation or contrast); (c) their size; (d) their generalization across related items; and (f) their dependency on the mode of data collection used.

Market researchers have long been aware that attitude measurement is context dependent, and that preceding questions may influence the responses given to subsequent ones. However, the conditions under which context effects may emerge are not well understood -- and when they emerge, it has typically been difficult to predict their direction. In the present paper, we present a process model (Schwarz & Bless, in press) that draws on current theorizing in social cognition (Barsalou, 1989; Kahneman & Miller, 1986) to identify variables that give rise to question context effects and to specify the conditions under which answering a preceding question results in assimilation or contrast effects on subsequent judgments.

ASSIMILATION AND CONTRAST EFFECTS

In short, we assume that individuals who are asked to form a judgment about some target stimulus first need to retrieve some cognitive representation of it. In addition, they need to determine some standard of comparison to evaluate the stimulus. Both, the representation of the target stimulus and the representation of the standard are, in part, context dependent. Individuals do not retrieve all knowledge that may bear on the stimulus, nor do they retrieve and use all knowledge that may potentially be relevant to constructing its alternative. Rather, they rely on the subset of potentially relevant information that is most accessible at the time of judgment. Accordingly, their temporary representation of the target stimulus, as well as their construction of a standard of comparison, includes information that is chronically accessible, and hence context independent, as well as information that is only temporarily accessible, due to contextual influences.

Whereas differences in the chronic accessibility of information reflect respondent characteristics, differences in the temporary accessibility of information are primarily due to questionnaire variables. Most importantly, information that has been used for answering a preceding question is particularly likely to come to mind when respondents are later asked a related question, to which it may be relevant.

Assimilation Effects

How the information that comes to mind influences the judgment, depends on how it is categorized. Information that is included in the temporary representation that individuals form of the target category will result in assimilation effects. This reflects that the judgment is based on the information that is included in the representation used. Accordingly, the addition of some positive information results in a more positive judgment, whereas the addition of some negative information results in a more negative judgment.

Suppose, for example, that you are asked to report your opinion about the politicians of the Christian Democratic Party that governs Germany. To do so, you will presumably recall some relevant information from memory. This information makes up the "cognitive representation" of politicians of the Christian Democratic Party. Assuming some basic familiarity with German politics, the chronically accessible information that comes to mind may include that the CDU is a conservative party, that Chancellor Kohl is a member of it, and so on.

Other features of the target, however, will only come to mind on some occasions. These features are therefore said to be "context dependent". For example, if you were just asked a question about Richard von WeizsScker, a particularly well respected member of the Christian Democrats, information related to Richard von WeizsScker may come to mind later on. If you include this information in the cognitive representation that you construct of the Christian Democrats, you should evaluate politicians of this party in general more positively than if no question about Richard von WeizsScker was asked.

Empirically, this is the case. In one of our experiments we asked subjects to recall the party of which "Richard von WeizsScker has been a member for more than 20 years". Answering "Christian Democratic Party" should include Richard von WeizsScker in the temporary representation that respondents form of politicians of the Christian Democrats. As shown in the first row of Table 1, this manipulation resulted in a more favorable evaluation of the party in general.

Contrast Effects

However, according to our model, the same piece of information may also result in a contrast effect. We propose that this is the case when the information is excluded from, rather than included in, your cognitive representation of the target.

For example, Richard von WeizsScker has not only been a member of the Christian Democratic Party for several decades, but he also serves as President of the Federal Republic of Germany -- and the office of President requires that he no longer actively participates in party politics. The President as a representative figure-head of the Federal Republic is supposed to take a neutral stand on party issues, much as the Queen in the UK. This rendered him particularly suitable for the present experiment because it allowed us to ask other respondents, "Do you happen to know which office Richard von WeizsScker holds, that sets him aside from party politics?".

TABLE 1

EVALUATION OF POLITICAL PARTIES AS A FUNCTION OF THE INCLUSION OR EXCLUSION OF A HIGHLY RESPECTED POLITICIAN

Answering this question should exclude Richard von WeizsScker from the representation that respondents form of Christian Democratic Party politicians. If so, it should result in lower evaluations of Christian Democratic Party politicians in general, although it may do so for different reasons, as discussed below. Empirically, this was again the case, as shown in the first row of Table 1.

This contrast effect may reflect either of two processes. On the one hand, Richard von WeizsScker and his party membership may have been chronically accessible for some subjects who were not asked a question about him. If so, some of our control group subjects may have spontaneously included him in their representation of the party. In this case, the assimilation effect would reflect that the party membership question increased the number of subjects who did so. By the same token, the contrast effect would indicate that the presidency question decreased the number of subjects who did so. Thus, the obtained contrast effect would simply reflect that these subjects "subtracted" Richard von WeizsScker from the data-base that they used in forming a judgment.

On the other hand, evaluations do not only require a representation of the target stimulus, but also a standard against which the target is evaluated. We assume that this standard is just as context dependent as the representation of the target itself, and includes chronically as well as temporarily accessible information. Accordingly, our respondents may not only have excluded Richard von WeizsScker from their representation of his party, but may also have used him in constructing a standard of comparison against which his party was evaluated. Given that WeizsScker sets a high standard, such a comparison process would also result in a lower evaluation.

Thus, our model provides two different mechanisms for the emergence of a contrast effect, which are not mutually exclusive: Namely the subtraction of information from the representation of the target, and the possible use of this information in the construction of a standard or scale anchor. How can we distinguish these two processes? The most straightforward way to distinguish them is to assess the generalization of the obtained contrast effect across different stimuli to which the standard of comparison may be relevant. If the obtained contrast effect solely reflects the subtraction of Richard von WeizsScker from the representation of the target category "politicians of the Christian Democratic Party", the presidency question should only affect the evaluation of politicians of this party. If respondents used Richard von WeizsScker in constructing a standard of comparison or a scale anchor, on the other hand, the obtained contrast effect should generalize to the evaluation of politicians of other parties, such as the Social Democratic Party, to which this standard is also applicable.

To explore this possibility, we exposed other respondents to the same questions but asked them to provide a general evaluation of politicians of the Social Democratic Party. As shown in the second row of Table 1, neither of the questions about Richard von WeizsScker affected subjects' evaluation of politicians of the Social Democratic Party. This suggests that the contrast effect obtained on the evaluation of Christian Democratic politicians reflects a subtraction effect, rather than a change in the standard of comparison or scale anchor.

The Impact of Category Width: Political Scandal Study

Our next study (Schwarz & Bless, in press) extends the inclusion/exclusion logic by demonstrating changes in the standard of comparison. Moreover, it bears on a different variable that may determine the inclusion or exclusion of information, namely the inclusiveness of the question asked.

Suppose, for example, that respondents are induced to think about politicians who were involved in a political scandal, and are subsequently asked to evaluate the trustworthiness of politicians in general. According to our model, the politicians involved in the scandal are members of the general category "politicians" and are therefore likely to be included in the temporary representation that respondents form of that category. If so, their evaluation of the trustworthiness of politicians in general should decrease, reflecting an assimilation effect.

TABLE 2

EVALUATION OF THE TRUSTWORTHINESS OF POLITICIANS IN GENERAL AND OF THREE EXEMPLARS AS A FUNCTION OF THINKING ABOUT A SCANDAL

Suppose, however, that respondents are not asked to evaluate the trustworthiness of politicians in general, but the trustworthiness of a specific politician, Mr. Joe Doe, who was not involved in the scandal. We assume that in evaluating a specific person, this person makes up a category by him- or herself. If so, the politicians who were involved in the scandal should not be included in subjects' representation of Joe Doe. Rather, these politicians may be used in constructing the standard or scale anchor against which Joe Doe is evaluated. If so, Joe Doe may seem particularly trustworthy by comparison, reflecting a contrast effect. Thus, the model predicts that thinking about politicians who were involved in a scandal decreases judgments of the trustworthiness of politicians in general, but increases judgments of the trustworthiness of specific politicians, provided that they were not involved in the scandal.

To test this implication, we asked subjects to recall the names of some politicians who were involved in a recent political scandal in West Germany (the so called "Barschel Scandal", which bears some resemblance to the Watergate scandal), either before or after they answered the dependent variables. As shown in Table 2, thinking about the Barschel scandal resulted in decreased judgments of the trustworthiness of German politicians in general. This assimilation effect reflects that subjects included the politicians who were involved in the scandal in their representation of German politicians in general.

Other subjects, however, were asked to evaluate the trustworthiness of three specific politicians, who pretests had shown to be not particularly trustworthy to begin with, although they were not involved in the scandal under study. In this case, thinking about the Barschel scandal increased judgments of trustworthiness of these specific politicians. This contrast effect reflects that subjects used the easily accessible politicians who were involved in the scandal in constructing a standard of comparison or scale anchor. Note that this contrast effect cannot be accounted for on the basis of a subtraction assumption. The information that was primed by the scandal questions was presumably never part of subjects' representation of the specific politicians they had to evaluate. Hence, the contrast effect obtained here reflects the use of the recalled politicians in constructing a relevant standard of comparison or scale anchor. This information could only be used in constructing the standard, however, when it was not perceived to bear on the respective target category in the first place.

These findings indicate that the same information may affect related judgments in opposite directions, depending on whether the respective target category invites the inclusion or the exclusion of the information that comes to mind. In general, our model predicts that assimilation effects are the more likely to emerge, the more inclusive the target category is. Accordingly, general questions that assess respondents' opinion about a wide target category, that allows for the inclusion of a variety of different information, should be most likely to show assimilation effects. On the other hand, specific questions, that assess respondents' opinion about a narrowly defined target, should be more likely to show contrast effects. This reflects that it is more likely that the information that comes to mind can be included in one's representation of a global rather than of a specific target.

On the substantive side, we would obviously draw very different conclusions about the impact of scandals on politicians' trustworthiness, depending on whether we asked a general or a specific question. In the light of these data, it also comes as no surprise that political scandals are typically accompanied by attempts to channel the public's categorization of scandal related information. To the extent that individual politicians, or groups of politicians, can dissociate themselves from the scandal, they may actually benefit from the misdemeanor of their peers, although the impact on the perception of the profession as a whole is likely to be negative.

TABLE 3

CONTRAST EFFECTS AS A FUNCTION OF THE DIMENSION TAPPED BY PRECEDING QUESTIONS CONTEXT STIMULUS

The Role of Salient Dimensions

So far, we have seen evidence for the emergence of subtraction based contrast effects in the WeizsScker study, and for comparison based contrast effects in the scandal study. Under which conditions, however, are we likely to obtain one or the other? We propose that thinking about some stimulus does only evoke comparison processes in evaluating subsequent stimuli if the stimulus is linked to the relevant judgmental dimension. If the stimulus is thought about with regard to some other dimension, it is unlikely to be used as a standard or as a scale anchor for a later unrelated judgment. With regard to the preceding studies, this suggests that thinking about a political scandal was likely to bring the dimension of trustworthiness to mind. On the other hand, thinking about Richard von WeizsScker's party membership or office may have been less likely to bring the evaluative dimension to mind that was relevant to subsequent judgments of the Christian Democrats.

In line with this assumption, one of our studies demonstrated that comparison effects do only emerge when the context dependent information is linked to the dimension of judgment (Schwarz, MHnkel, & Hippler, 1990). Specifically, we asked subjects to rate how "typically German" a number of different beverages is, namely wine, coffee, and milk. Before they made this judgment, some subjects were asked to estimate the caloric content of a glass of vodka, or of a glass of beer, respectively. Other subjects, however, were asked to estimate how frequently Germans drink vodka or beer.

Both questions should make vodka or beer more accessible in memory. However, only the frequency of consumption question is related to the typicality dimension, whereas the caloric content question is not. If it is sufficient that an extreme stimulus comes to mind, both questions should result in contrast effects on subsequent typicality ratings. If the emergence of comparison effects requires that the extreme stimulus is linked to the relevant judgmental dimension, on the other hand, contrast effects should only emerge when subjects estimated the frequency of consumption.

Empirically, this was the case. When subjects estimated the frequency of consumption, they rated all beverages as more typically German after thinking about vodka than after thinking about beer, as shown in Table 3. Estimating the caloric content of vodka or beer, on the other hand, did not affect their ratings. Accordingly, we conclude that the emergence of comparison or anchoring effects requires that the context dependent information is linked to the relevant dimension of judgment.

This suggests that we may only see comparison or anchoring based contrast effects when the excluded information is thought about with regard to the relevant dimension of judgment. In that case, it may be used as a standard of comparison or as a scale anchor, resulting in contrast effects that generalize across various target categories. If the excluded information is thought about with regard to some other dimension, it may still result in contrast effects, but only by means of a subtraction process. Accordingly, the contrast effects that emerge under this condition should be limited to the evaluation of the category from which the information was excluded in the first place, as was the case in the WeizsScker study.

THE INCLUSION/EXCLUSION MODEL

In summary, when asked to form an opinion about some target, respondents will search memory for relevant information. Some information is likely to come to mind under all circumstances, and is hence called context independent. Other information will only come to mind because it has been used recently, for example, in answering a preceding question. How the information that comes to mind influences the judgment, depends on what respondents do with it. On the one hand, they may use it in constructing a cognitive representation of the target. On the other hand, they may use it in constructing a standard of comparison or scale anchor, against which the target is evaluated.

For which purpose the information is used, depends on a number of categorization decisions that respondents have to make. In the present paper, we have only discussed one of the three decisions shown in Figure 1. This decision reflects if the information "belongs" to the target category or not. This decision is influenced by the content of preceding questions, as in the case of the WeizsScker study, and by the width of the target category, as in the political scandal study, as well as related variables (cf. Schwarz & Bless, in press). If the answer to the "does-it- belong?" decision is YES, the information is included in the representation formed of the target category. This results in an assimilation effect. The size of this assimilation effect increases with the amount and extremity of context dependent information, and decreases with the amount and extremity of context independent information.

FIGURE 1

INCLUSION/EXCLUSION AND THE EMERGENCE OF ASSIMILATION AND CONTRAST EFFECTS

If the answer to the "does-it-belong" decision is NO, the information is excluded from the representation formed of the target category. This results in a contrast effect that may or may not generalize to judgments about other targets. If the information has NOT been thought about with regard to the underlying dimension of judgment, it is only excluded from the representation of the target category. This results in a localized contrast effect that does not influence related judgments of other targets, as we have seen in the WeizsScker study. We call this type of contrast effect a subtraction effect. The size of subtraction effects again depends on the amount and extremity of the subtracted information, and on the amount and extremity of other information that is used in constructing the representation. If the excluded information HAS been thought about with regard to the underlying dimension of judgment, however, it is likely to be used in constructing a standard, which may be applied to other targets as well. This results in generalized contrast effects, as we have seen in scandal study and the study on drinks. We call this type of context effect a comparison effect. Its size depends on the amount and extremity of the context dependent and independent information used in constructing the standard.

In combination, the model predicts the direction as well as the size and the generalization of context effects. Moreover, it provides a framework for the conceptualization of the role of questionnaire variables, which influence the temporary accessibility of information and respondents' categorization decisions, as well as respondent characteristics, which influence chronic accessibility.

As shown in Figure 1, other decisions bear on whether the information comes to mind due to some external influence (e.g., Lombardi, Higgins, & Bargh, 1987), and on whether conversational norms do or do not prohibit its use (e. g., Schwarz, Strack, & Mai, 1991). We provide a detailed discussion of these steps, and a review of relevant research, elsewhere (Schwarz and Bless, in press). In general, these decision steps allow us to conceptualize the impact of introductions to blocks of questions, the presence of filler questions, and other aspects of questionnaire design, including the graphical lay-out of mail questionnaires. All of these variables have been shown to influence inclusion/exclusion decisions and hence the direction of context effects in some of our studies.

One additional point, however, deserves attention. Figure 1 seems to suggest that respondents are engaged in a lot of decision making in a very orderly fashion. Probably, this is not quite so. We assume that the default option is to include what comes to mind and that decisions that may result in exclusion processes need to be triggered by some salient feature of the questioning process. This suggests that we should be more likely to see assimilation rather than contrast effects in survey measurement, reflecting that assimilation effects are based on the default option, namely the inclusion of what comes to mind. Moreover, exclusion processes require additional cognitive work. Accordingly, we should only see them when respondents are motivated and able to engage in this work (see Martin, Seta, & Crela, 1989). This suggests that contrast effects should be more likely to emerge the more respondents have a chance to engage in the necessary cognitive work, as is for example, the case under the leisurely conditions of responding to a self-administered questionnaire, but less under the time pressure of a telephone interview (cf. Schwarz, Strack, Hippler, & Bishop, 1991). Hence, the model predicts that the emergence of context effects may depend on the mode of data collection, although relevant data are not yet available.

[The reported research was supported by grant SWF0044-6 from the Bundesminister fnr Forschung und Technologie of the Federal Republic of Germany to N. Schwarz.]

REFERENCES

Barsalou, L. W. (1989). Intraconcept similarity and its implications for interconcept similarity. In S. Vosniadou & A. Ortony (Eds.), Similarity and analogical reasoning (pp. 76-121). Cambridge: Cambridge University Press.

Lombardi, W. J., Higgins, E. T., & Bargh, J. A. (1987). The role of consciousness in priming effects on categorization: Assimilation and contrast as a function of awareness of the priming task. Personality and Social Psychology Bulletin, 13, 411-429.

Kahneman, D. & Miller, D. (1986). Norm theory: Comparing reality to its alternatives. Psychological Review, 93, 136-153.

Martin, L. L., Seta, J. J., & Crelia, R. A. (1990). Assimilation and contrast as a function of people's willingness to expend effort in forming an impression. Journal of Personality and Social Psychology, 59, 27-37.

Schwarz, N., & Bless, H. (in press). Constructing reality and its alternatives: Assimilation and contrast effects in social judgment. In L.L. Martin & A. Tesser (Eds.), The construction of social judgment. Hillsdale, NJ: Erlbaum.

Schwarz, N., Mnnkel, T., & Hippler, H.J. (1990). What determines a "perspective"? Contrast effects as a function of the dimension tapped by preceding questions. European Journal of Social Psychology, 20, 357-361.

Schwarz, N., Strack, F., Hippler, H.J., & Bishop, G. (1991). The impact of administration mode on response effects in survey measurement. Applied Cognitive Psychology, 5, 193-212.

Schwarz, N., Strack, F., & Mai, H.P. (1991). Assimilation and contrast effects in part-whole question sequences: A conversational logic analysis. Public Opinion Quarterly, 55, 3-23.

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