The Effects of Message Valence on Inferential Processes

Lydia J. Price, INSEAD
[ to cite ]:
Lydia J. Price (1992) ,"The Effects of Message Valence on Inferential Processes", in NA - Advances in Consumer Research Volume 19, eds. John F. Sherry, Jr. and Brian Sternthal, Provo, UT : Association for Consumer Research, Pages: 359-365.

Advances in Consumer Research Volume 19, 1992      Pages 359-365


Lydia J. Price, INSEAD

Although inferential belief formation and inferential belief change processes have frequently been investigated, there remains a gap in our knowledge about the naturally occurring conditions that may lead to their activation. In this paper we explore the effects of message valence on inferencing behavior. Empirical evidence suggests that negative information may be effective in stimulating inferential reasoning. Results concerning the effects of positive information cues were inconclusive.

In recent years a number of researchers have examined inferential processes by which consumers expand upon given product information in order to form new beliefs (Ford and Smith 1987; Gardial and Biehal 1987; Huber and McCann 1982; Johnson and Levin 1985; Simmons and Leonard 1990) or modify existing beliefs (Kaplan 1972; Lutz 1975a, 1975b; Yi 1990a, 1990b) about attributes that are unknown or are unmentioned in a communication. Although there is mounting evidence that inferencing can have a strong effect on consumer judgments, it is clear that inferential processing does not always occur spontaneously (Lim, Olshavsky, and Kim 1988; Simmons and Lynch 1991). Investigators thus have begun to search for environmental factors that may provide an impetus for inferential reasoning. In the research reported here, we examine the effects of message valence on inferential belief change processes. Empirical evidence suggests that negative messages may be effective in stimulating inferential thinking.


In a communications context, inferencing occurs when consumers expand upon the attribute information provided in a message so as to form new beliefs or update their existing beliefs about attributes that are not mentioned in the message. Inferred attribute values are likely to be either probabilistically (i.e., correlationally) or evaluatively consistent with the attribute claims made in the communication (Fishbein and Ajzen 1975). Although both probabilistic and evaluative inferencing strategies have been observed empirically, it has been suggested that the former are more common, particularly when attribute correlations are high (Fishbein and Ajzen 1975; Gardial and Biehal 1987).

On the basis of this suggestion, it seems reasonable to expect the likelihood of inferencing behavior to be positively related to the strength of the correlations among mentioned and unmentioned product attributes (Huber and McCann 1982). Empirical tests of this proposition, however, indicate that high correlations are not a sufficient condition for the onset of inferential reasoning. Although the magnitude of inferred attribute values has been found to covary with the strength of attribute correlations when probabilistic inferencing is observed (Ford and Smith 1987; Yi 1990b), several researchers have failed to find evidence of spontaneous inferencing behavior in a context where attribute correlations were strong (Lim et al. 1988; Simmons and Lynch 1991).

Others have reasoned that environmental factors must lead consumers to think about unmentioned product attributes or about attribute correlations in order to stimulate inferential processing (Hansen and Zinkham 1984; Huber and McCann 1982). In accordance with this view, Yi (1990b) found significant effects of visual advertising cues in prompting inferential belief change. Apparently, visual cues were effective in leading message recipients to think about the unmentioned attributes and to update their beliefs accordingly. Similarly, Kardes and Strahle (1986) found evidence of inferential belief formation when the perceived technicality of product information was moderate, but not when it was high or low. In this case it seems that subjects were motivated to process the given information deeply, thus prompting inferential thinking, when the information offered new insights which subjects felt capable of understanding. When information was uninformative or overly complex, processing motivation was reduced.

On the other hand, there have been studies where proposed environmental factors have failed to elicit inferencing behavior. Simmons and Lynch (1991) found very little incidence of inferencing activity in a study where competitor communications were used to increase the salience of a target brand's unmentioned attributes. Zwick (1988) similarly failed to find systematic effects of information format (brand dominant, attribute dominant, or matrix format) on the way that subjects accessed information items and subsequently used them to guide inferential reasoning. Moreover, a substantial proportion of Zwick's research subjects (14 out of 37 who were not prompted) showed no signs at all of spontaneous inferencing activity.

In light of the mixed evidence regarding specific environmental factors Dick, Chakravarti, and Biehal (1990) have argued that consumers will engage in inferencing activity only when 1) the unmentioned attribute information is noticed and is sufficiently relevant to motivate efforts to impute its value, 2) an inferencing rule and the necessary supporting information are readily accessible in memory, 3) the inferencing rule is perceived to be capable of generating diagnostic inputs for the decision task at hand, and 4) other diagnostic inputs to the decision are not as readily accessible or available from external sources (Feldman and Lynch 1988; Lynch, Marmorstein, and Weigold 1988). Dick et al. support this view with the results of an experiment where inferencing instructions were varied and information accessibility was manipulated. When subjects were not explicitly instructed to infer missing attribute values, inferencing behavior was not uniformly observed. Inferencing effort was greatest, however, when prior brand evaluations were favorable rather than unfavorable, and attribute information was highly accessible from memory. Unfavorable prior evaluations apparently decreased subjects' interest in a given choice alternative, thus reducing the relevance of an inferencing strategy for subsequent choice deliberations. Low attribute accessibility apparently decreased the perceived diagnosticity of an inferencing rule.

The Dick et al. (1990) framework sheds light on previous failures to stimulate inferencing activity in the laboratory. For example, many subjects in the Simmons and Lynch (1991) study failed to note the absence of target attribute information. Moreover, of those who did notice, some probably lacked access to an inferencing rule linking the predominantly tangible product features used as stimulus attributes. Simmons and Lynch (1991) and Lim et al. (1988) note that inferences pertaining to tangible product features are less common than those pertaining to a price/quality relationship. This is not surprising given that price/quality relationships are observed in many product categories, leading to a highly accessible rule linking the two characteristics. For tangible product features, on the other hand, less frequent observations of attribute associations render inferencing rules less readily accessible. When subjects in the Simmons and Lynch study were induced to repeatedly make similar product evaluation decisions, their awareness of missing information and attribute correlations was likely increased. Consistent with the Dick et al. (1990) framework, inferencing behavior also increased (Experiment 3).

Simmons and Leonard (1990) further point out that, of the inferences made by subjects in the Simmons and Lynch (1991) study, a probabilistic consistency rule tended to apply when attribute correlations were expected to be high. When no attribute correlations were expected, probabilistic inferences were rare. Thus, despite the low overall incidence of inferencing behavior, subjects seemed most willing to use a probabilistic inferencing rule when it was expected to yield diagnostic conclusions.

The Effects Of Information Valence: Following the Dick et al. (1990) framework one would expect that any factor which increases the relevance, the accessibility and/or the diagnosticity of an inferencing rule will increase the likelihood that inferential processing is observed. One such factor might be information valence. The disproportionate influence of negative information on consumer judgment tasks has been documented repeatedly (see Kanouse 1984 for a review of the negativity literature), and there is evidence that inferencing activity may be sensitive to negativity effects. Lutz (1975a), in an early study, found that negative messages stimulated stronger changes than positive messages in an index of total cognitive structure. It appears that inferential belief change may have been stronger in response to the negative stimuli, although it is difficult to verify this conclusion as individual belief changes were not reported in the paper. Mizerski (1982) later found that inferential belief formation was stronger in response to negative rather than positive information in a context where inferences were prompted explicitly by the researcher. Finally, Wansink (1989) found that negative information about the source of a message led subjects to make negative attributions as to why the source omitted attribute information from its advertisements. These negative attributions had an adverse effect on product evaluations, particularly when the unmentioned attributes were covariant with those actually mentioned in the ad. The significant covariance effect suggests that inferencing behavior might have mediated the source effects on product evaluations.

It is possible that negative information would stimulate inferencing behavior because of the way that it is processed by message recipients. Fiske (1980) has argued that negative information is allocated greater attention and is processed more deeply than is positive information. There are many reasons why this might be so. First, negative information may be more rare or atypical than positive information (Boucher and Osgood 1968; Fiske 1980; Zajonc 1968), or it might be less consistent with prior product beliefs. Alternatively, negative product aspects may interfere with the enjoyment of positive product aspects, or individuals may find it risky to ignore the implications of negative claims (Kanouse and Hanson 1972). Although these conditions are not likely to hold in all situations (e.g., when negative information is consistent with prior beliefs because it refers to an already disliked brand, or when negative information is typical, as might be the case for comparative advertisements), they can reasonably be expected in many consumer communication environments.

Although there are differences among these conditions, each represents a motivating reason for the individual to review his beliefs and knowledge about the object of the communication. In the first set of conditions, a review of prior beliefs might serve to check the veracity of the new information and the prior beliefs. In the second set of conditions, a review of prior beliefs might serve to weigh the implications of those beliefs against that of the new information. In either event, negative information increases the likelihood that message recipients will compare the new product claim to previously learned brand information.

One outcome of such a comparison procedure is likely to be an increase in the accessibility of brand attribute information. A second outcome is likely to be heightened recognition of the extent to which the new information deviates from prior product beliefs. In the event that the negative information is judged to be valid, deviations from prior beliefs would create pressure for the individual to change those prior beliefs. Negative information thus may increase the relevance to message recipients of an inferential belief change strategy for modifying product evaluations.

Given the diagnosticity of a probabilistic inferencing rule for updating attribute beliefs, recipients of negative messages may be encouraged to seek the correlation information which, when combined with accessible brand attribute information, would enable them to make probabilistic inferences. Following Dick et al. (1990) we would expect such a search for correlation information to lead negative message recipients to perceive stronger correlations among the attributes of the target product category. As positive messages are more likely to be subject to cursory processing effort (Fiske 1980), message recipients may be less likely to note deviations between positive messages and prior beliefs. Consequently, the perceived relevance of an inferential belief change strategy may be lower following positive information exposure, and recipients may be less likely to seek correlation information.


On the basis of the preceding discussion, two research hypotheses were formed. The first follows from the argument that negative information increases the relevance of an inferential belief change strategy for the modification of brand evaluations. Given high perceived diagnosticity of a probabilistic inferencing rule, message recipients are expected to search for correlation information which would facilitate implementation of such a rule.

H1: Recipients of negative information about a given product attribute will perceive stronger correlations between that attribute and others of the target product category than will recipients of positive attribute information.

As a result of the search for correlation information, the informational inputs to a probabilistic inferencing rule should be highly accessible to the message recipient. The likelihood of observing probabilistic inferencing behavior in this situation is high.

H2a: Individuals exposed to negative information will be more likely than those exposed to positive information to exhibit inferencing behavior.

H2b: Inferential processing in response to negative information will conform to a probabilistic consistency rule.


Subjects and Procedure

An experiment was conducted to test the above hypotheses in a context where attribute information was provided for previously known brands. The focus of the study was on inferential belief change processes rather than inferential belief formation. Subjects were 56 undergraduate and graduate students at two U.S. universities who were recruited from announcements placed on campus. Each person was paid $7 for participating in the study, and his/her name was entered into a lottery for a larger cash prize.

The research design was a two-group pretest-posttest in which groups were exposed to stimulus messages differing in valence. Subjects were told that the purpose of the study was to examine the effects of independent laboratory testing reports on consumer perceptions and opinions. Their first task was to rate a set of 11 adult cereals (one of which was the target brand for the experimental manipulation) on 7-point semantic differential scales labelled "very low/very high in sugar", "very low/very high in calories", "not at all/very nutritious", and "not at all/very pleasant in taste".

After completing a filler task unrelated to the study's purposes, subjects were asked to examine a rating form which, they were told, had appeared in a laboratory testing report. This form, which presented the stimulus message, rated the target brand as being either very low or very high in sugar content. Following their review of the stimulus materials, subjects were asked a series of questions concerning the perceived credibility of the testing laboratory and of the laboratory report. Next, they were asked once again to rate the target brand in terms of sugar content, calories, nutrition, and taste using the 7-point scales described above. Finally, following a format described by Ford and Smith (1987), subjects were asked to indicate (on a 10 point scale) how likely it is that they could predict the level of the other cereal attributes given knowledge of a brand's sugar content. Responses to this question were divided by 10 to serve as post-message measurements of the strength of perceived correlations between sugar and the other cereal attributes.

Stimulus Materials

Pretesting revealed that the target cereal brand was perceived to be moderate in sugar content (3.36 on a 5-point scale). Furthermore, pretest subjects perceived sugar to be strongly correlated with calories, moderately correlated with taste and weakly correlated with nutrition. This pattern of correlations allows for differences in the magnitude of belief change across the three attributes following the application of a probabilistic inferencing rule. In this event, the relative magnitudes of observed belief changes would be expected to correspond to the relative magnitudes of perceived correlations with sugar (Ford and Smith 1987). That is, observed belief changes would be strongest for calories and weakest for nutrition. If inferencing followed an evaluative consistency rule, there would be no differences in the magnitudes of the observed changes across the three attributes. On the basis of this difference in the hypothesized effects of the two inferencing rules, sugar was considered to be an appropriate attribute for testing hypothesis 2b. Accordingly, sugar was selected as the stimulus attribute for experimental messages, and belief changes pertaining to calories, taste, and nutrition were measured as dependent variables.

After rescaling, the target brand was found to be rated 4.5 on a 7-point scale of sugar content. Equal polarity of positive and negative stimulus messages was attained by rating the target brand of cereal either as 2 (very low in sugar) or as 7 (very high in sugar) on a 7-point rating scale. As pretest subjects had indicated that the ideal level of sugar in a cereal was 2.98 on a 7-point scale, a rating of 2 for the target brand was expected to be seen as a positive message whereas a rating of 7 was expected to be seen as negative.


Manipulation Check

Subjects' perceptions about attribute correlations were checked to be sure that they were similar to those of the pretest sample. Correlations were calculated from subjects' brand by attribute ratings of the 11 adult cereals. The average correlation between sugar and calories was 0.67; that between sugar and taste was 0.24, and that between sugar and nutrition was -0.15. This pattern matches that of the pretest sample, suggesting that perceptions of attribute correlations are reasonably stable for the adult cereal category.

A second check sought to verify that the stimulus messages had changed subjects' beliefs about the sugar content of the target brand. Subjects receiving the negative (high sugar) message significantly increased their sugar ratings of the target brand (average increase 1.21, t=4.93, p<.001), indicating a successful belief change manipulation. Subjects receiving the positive (low sugar) message, on the other hand, failed to significantly change their beliefs about the brand's sugar content (average decrease 0.07, t=-0.23, p<.82).

This unexpected failure of the positive manipulation weakens our planned tests of hypotheses 1 and 2a. The failure suggests either that subjects processed the positive message but subsequently rejected it as inaccurate or untrue, or that they simply failed to process the message very deeply. The first possibility seems unlikely in light of the result that subjects perceived the positive message to be as credible as its negative counterpart (4.63 and 4.45, respectively, on a 7-point scale; t54=0.45, p=0.65). The second possibility seems more likely. Fiske (1980) has argued that positive information attracts less attention than negative information, and that it is correspondingly subject to more shallow processing mechanisms. In this case, processing of the positive stimulus message may have been especially low. Why this would be so is unclear. It is possible that subjects in this study classified cereal brands into wide categories of sugary/non-sugary, with the result that the target brand was seen as a low sugar brand, despite its moderate sugar ratings. In this situation, any low sugar message might have matched subjects' expectations to such an extent that it would be subject to little attention.

Limited processing of the positive claim would make a subsequent search for correlation information unlikely. This in turn would render inferencing behavior unlikely. The hypothesized effects of a weakly processed positive message thus are no different from those hypothesized for a stronger positive claim, and the planned tests can still be conducted. In this case, however, the tests will be conceptually similar to a one-sided test which compares a negative message to a null situation where no targeted belief change activity is observed. It will be impossible to draw conclusions about inferencing behavior under conditions of positive change in a targeted attribute belief.

Test of H1

The first hypothesis states that subjects exposed to a negative stimulus message will perceive stronger correlations among the cereal attributes than will those exposed to a positive stimulus message. In contrast to the hypothesis, however, differences in post-message ratings of attribute relationships across the two groups were directionally correct for each attribute, but none of the effects attained statistical significance. Perceived correlations between sugar and calories were 0.81 and 0.77 for the negative and positive groups, respectively (t54=0.69, p<0.50), those between sugar and taste were 0.72 and 0.69, respectively (t54=0.49, p<0.63), and those between sugar and nutrition were 0.62 and 0.58, respectively (t54=0.49, p<0.63). The first hypothesis thus is not supported by these data.

Perceived Correlations. It was observed in the test of the first hypothesis that subjects' stated perceptions about correlations between sugar and nutrition and between sugar and taste were much higher than those calculated from their ratings of the 11 cereal brands. Although this result was unexpected, it makes sense with a bit of reflection. It appears that subjects saw little covariance among the attributes in the sample of adult cereals tested, but they did expect a relationship for cereal brands in general. As questioning did not specify that subjects should think of the adult cereal sub-category, their answers evidently pertain to the cereal category as a whole.

This result is not problematic for testing the first hypothesis unless it is believed that a post-message search for correlation information would be restricted to the adult cereal sub-category. This seems unlikely in light of spreading activation theories of memory processes (e.g., Anderson 1983) which suggest that, even if sub-category perceptions of attribute associations were activated initially, links between perceptions at various product levels would eventually lead to activation of them all. In the experiment reported here, ample time had elapsed between delivery of the stimulus message and measurement of perceived correlations to allow for activation of the broader network of information.

A more interesting question arises with respect to testing the second set of hypotheses. If message recipients have ready access to information about attribute associations at both the category and the sub-category levels, either set of information could be used as an input to an inferencing rule. In the present case, it might be expected that subjects' category-level perceptions of strong associations between sugar and taste and between sugar and nutrition would lead to strong inferential changes in their beliefs about these attributes in response to a strong sugar message. Alternatively, it might be expected that perceptions of weaker relationships at the sub-category level would lead to weaker inferential changes. Although we are unaware of any prior studies that have looked explicitly at this question, it would seem that sub-category perceptions would lead to more diagnostic inferences, and thus would be preferred as an input to decision making. In that case, we would expect relatively weak inferencing behavior with respect to the nutrition and taste attributes.



Tests of H2a and H2b

The second hypothesis requires that inferential belief change be observed in the negative, but not the positive, message condition. Subjects exposed to positive messages thus should exhibit no significant changes in their beliefs about the level of calories, taste, or nutrition exhibited by the target brand. Additionally, H2b requires that inferencing behavior in the negative message condition follow a probabilistic consistency rule. The relative magnitudes of the belief changes observed in the negative message condition thus should correspond to the relative magnitudes of the correlations among the respective attributes.

As sugar and calories consistently were perceived to be strongly related it was expected that subjects in the negative message condition would exhibit large, significant changes in their beliefs about the caloric content of the target brand. For the taste and nutrition attributes, predictions were made on the basis of sub-category perceptions of attribute correlations since these were expected to be more diagnostic than those at the category level. It thus was expected that observed belief changes would be moderate for taste and small for nutrition. The weak sugar/nutrition correlation suggests that the magnitude of the change in nutrition beliefs will fail to reach statistical significance. It is more difficult, however, to predict whether belief change will be significant for taste. No specific predictions were made for this question since greater insight was expected to come from comparing the relative magnitudes of the belief changes rather than their significance levels.

Belief changes were calculated for each attribute by subtracting pre-message ratings of the target brand's attribute levels from their post-message equivalents. The table shows the magnitude of each observed change for both of the experimental groups. As expected, no significant inferencing was detected among subjects in the positive information condition. In contrast, subjects in the negative information condition significantly changed their beliefs about the levels of calories and taste exhibited by the target brand. The results are supportive of the second hypothesis in demonstrating that negative information prompted significant, systematic changes in non-targeted attribute beliefs whereas the positive message prompted no significant effects. Although the failure of the positive manipulation renders the test inconclusive regarding the effects of a positive targeted belief change on inferencing behavior, the lack of significant effects in the positive condition increases our confidence that the inferencing observed in the negative condition was more than a random event.

As seen in the table, the relative magnitudes of the observed belief changes in the negative message condition conform to the relative magnitudes of the perceived correlations among the attributes of the 11 adult cereal brands. These results are strongly supportive of part two of the second hypothesis. If an evaluative consistency rule had been followed, all beliefs would have changed in an unfavorable direction. Further, the magnitude of observed changes would have been similar across the attributes. The observed differences in magnitude and the more favorable ratings of the target brand's taste make it unlikely that an evaluative inferencing rule was followed.

The observed differences in inferencing magnitude also make it unlikely that category level perceptions of attribute correlations were used as an input to the inferencing process. The use of sub-category correlations is consistent with the Dick et al. (1990) argument that the choice of an inferencing rule will be influenced by considerations of rule diagnosticity. In this case, inferences drawn from the adult cereal sub-category would be more diagnostic than those drawn from the cereal category considered as a whole.


The experiment provides evidence that negative information may stimulate inferential processing, although the reasons for the effect remain unclear. Subjects who saw negative brand information did not perceive stronger attribute correlations than those who saw positive brand information. There thus was no evidence that negative message recipients were more likely than positive message recipients to actively seek correlation information. This lack of support for the first hypothesis is especially noteworthy in light of the failure of the positive manipulation. If negative information fails to increase the strength of perceived correlations over the level realized by a weakly processed positive message, it is unlikely to be more effective in relation to a stronger positive message which is subject to heavier processing.

On the other hand, tests of the second hypothesis suggest that perceived correlations were accessible to subjects in the negative message condition. Without access to this information it is unlikely that subjects would have followed the hypothesized probabilistic inferencing strategy. Combined, then, the two hypothesis tests suggest that attribute correlations were accessible to subjects in both information conditions. It is possible that in questioning the positive message recipients about attribute correlations, the experimental procedure led them to seek correlation information. The similarity of responses across the two groups thus may reflect testing effects rather than an error in the logic supporting the first hypothesis. Future research should seek to reexamine this question using less problematic measures of the mechanisms underlying the inferencing process.

An additional question arose during hypothesis testing pertaining to whether inferential reasoning would be based on perceived attribute associations within the entire product category, or on those within a subset of the category to which the target brand belonged. In the present case, the latter appears to have held. The result is consistent with the argument made by Dick et al. (1990) that the inferencing rule which produces the most diagnostic conclusions will be selected. In addition to bolstering support for the Dick et al. framework, our finding should serve as a warning to others who work in this area that the appropriate category level must be used to make inferencing predictions.


The failure of the positive manipulation to change a targeted attribute belief made comparison of the negative and positive message conditions less informative. Conclusions about the effects of message valence on inferencing behavior thus must be made with care. In this research only negative information stimulated inferencing activity. It is possible, however, that the frequency and nature of inferencing behavior would have been the same in the positive information condition if the manipulation had realized greater success.

The study was further limited to product information delivered in the form of a laboratory testing report about a single product category. It may be that negative information from other sources or concerning other product categories would be less effective than that presented here. The laboratory report was judged to be reasonably credible by both groups of research subjects. Other sources clearly would be less credible as providers of both positive and negative information. In such an event, the interaction of source credibility and information valence might yield outcomes that differ from those reported here.

Another question to be explored in future research is that of how inferencing behavior would change if subjects' expectations about message valence were violated. If positive messages were delivered by unlikely sources (such as competitors) or positive information were very extreme, the atypicality of the situation might make positive information a sufficient stimulus for inferencing behavior. Alternatively, if negative information were delivered in a context where it was expected (such as in comparative advertisements), inferencing might not result. These unanswered questions highlight the need for additional research into the mechanisms that underlie inferencing behavior.


Anderson, John R. (1983), "A Spreading Activation Theory of Memory," Journal of Verbal Learning and Verbal Behavior, 22, 261-295.

Boucher, Jerry and Charles E. Osgood (1968), "The Polyanna Hypothesis," Journal of Verbal Learning and Verbal Behavior, 8 (1), 1-8.

Dick, Alan, Dipankar Chakravarti and Gabriel Biehal (1990), "Memory-based Inferences During Consumer Choice," Journal of Consumer Research, 17 (June), 82-93.

Feldman, Jack M. and John G. Lynch, Jr. (1988), "Self-Generated Validity and Other Effects of Measurement on Belief, Attitude, Intention, and Behavior," Journal of Applied Psychology, 73 (August), 421-435.

Fishbein, Martin and Icek Ajzen (1975), Belief, Attitude, Intention, and Behavior: An Introduction to Theory and Research, Reading MA: Addison-Wesley.

Fiske, Susan T. (1980), "Attention and Weight in Person Perception: The Impact of Negative and Extreme Behavior," Journal of Personality and Social Psychology, 38 (6), 889-906.

Ford, Gary T. and Ruth Ann Smith (1987), "Inferential Beliefs in Consumer Evaluations: An Assessment of Alternative Processing Strategies," Journal of Consumer Research, 14 (December), 363-371.

Gardial, Sarah and Gabriel Biehal (1987), "Measuring Consumers' Inferential Processing in Choice," in Advances in Consumer Research, Vol. 14, eds. Melanie Wallendorf and Paul F. Anderson, Provo, UT: Association for Consumer Research, 101-105.

Hansen, Chris J. and George M. Zinkham (1984), "When Do Consumers Infer Product Attribute Values?" in Advances in Consumer Research, Vol. 11, ed. Thomas C. Kinnear, Provo, UT: Association for Consumer Research, 187-192.

Huber, Joel and John McCann (1982), "The Impact of Inferential Beliefs on Product Evaluations," Journal of Marketing Research, 19 (August), 324-333.

Johnson, Richard D. and Irwin P. Levin (1985), "More Than Meets the Eye: The Effect of Missing Information on Purchase Evaluations," Journal of Consumer Research, 12 (September), 169-177.

Kanouse, David E. (1984), "Explaining Negativity Biases in Evaluation and Choice Behavior: Theory and Research," in Advances in Consumer Research, Vol. 11, ed. Thomas C. Kinnear, Provo, UT: Association for Consumer Research, 703-708.

Kanouse, David E. and L. Reid Hanson, Jr. (1972), "Negativity in Evaluations," in Attribution: Perceiving the Causes of Behavior, eds. Edward E. Jones et al., Morristown, NJ: General Learning Press, 47-62.

Kaplan, Kalman J. (1972), "From Attitude Formation to Attitude Change: Acceptance and Impact as Cognitive Mediators," Sociometry, 35 (3), 448-467.

Kardes, Frank R. and William M. Strahle (1986), "Positivity and Negativity Effects in Inferences About Products," in Advances in Consumer Research, Vol 13, ed. Richard J. Lutz, Provo, UT: Association for Consumer Research, 23-26.

Lim, Jeen-Su, Richard W. Olshavsky and John Kim (1988), "The Impact of Inferences on Product Evaluations: Replication and Extension," Journal of Marketing Research, 25 (August), 308-316.

Lutz, Richard J. (1975a), "Changing Brand Attitudes Through Modification of Cognitive Structures," Journal of Consumer Research, 1 (March), 49-59.

Lutz, Richard J. (1975b), "First-order and Second-order Cognitive Effects in Attitude Change," Communication Research, 2 (July), 289-299.

Lynch, John G., Jr., Howard Marmorstein, and Michael F. Weigold (1988), "Choices From Sets Including Remembered Brands: Use of Recalled Attributes and Prior Overall Evaluations," Journal of Consumer Research, 15 (September), 169-184.

Mizerski, Richard W. (1982), "An Attribution Explanation of the Disproportionate Influence of Unfavorable Information," Journal of Consumer Research, 9 (December), 301-310.

Simmons, Carolyn J. and Nancy H. Leonard (1990), "Inferences About Missing Attributes: Contingencies Affecting the Use of Alternative Information Sources," in Advances in Consumer Research, Vol 17, eds. Marvin E. Goldberg et al., Provo, UT: Association for Consumer Research, 266-274.

Simmons, Carolyn J. and John G. Lynch, Jr. (1991), "Inference Effects Without Inference Making? Effects of Missing Information on Discounting and Use of Presented Information," Journal of Consumer Research, 17 (March), 477-491.

Wansink, Brian (1989), "The Impact of Source Reputation on Inferences About Unadvertised Attributes," in Advances in Consumer Research, Vol 16, ed. Thomas K. Srull, Provo, UT: Association for Consumer Research, 399-406.

Yi, Youjae (1990a), "Direct and Indirect Approaches to Advertising Persuasion," Journal of Business Research, 20, 279-291.

Yi, Youjae (1990b), "The Indirect Effects of Advertisements Designed to Change Product Attribute Beliefs," Psychology and Marketing, 7 (Spring), 47-63.

Zajonc, Robert B. (1968), "Attitudinal Effects of Mere Exposure," Journal of Personality and Social Psychology, 9 (June), 1-29.

Zwick, Rami (1988), "Consumers' Use of Same- and Other-Brands Information Sources in the Formation of Inferential Beliefs about Partially Described Multiattribute Alternatives," Unpublished manuscript, The Pennsylvania State University, University Park.