Mere Exposure and the Cognitive-Affective Debate Revisited

ABSTRACT - There currently are two interpretations of how affect is generated in the mere exposure effect. Key evidence for and against each interpretation is reviewed and found questionable. A new approach to study the phenomenon is suggested, building on the concept of "perceptual fluency". This approach may give new insights into how and under which conditions low-involvement exposures to stimuli in a consumer behavior context may lead to more positive affect for those stimuli.

Citation:

Marc Vanhuele (1994) ,"Mere Exposure and the Cognitive-Affective Debate Revisited", in NA - Advances in Consumer Research Volume 21, eds. Chris T. Allen and Deborah Roedder John, Provo, UT : Association for Consumer Research, Pages: 264-269.

Advances in Consumer Research Volume 21, 1994      Pages 264-269

MERE EXPOSURE AND THE COGNITIVE-AFFECTIVE DEBATE REVISITED

Marc Vanhuele, Groupe HEC

ABSTRACT -

There currently are two interpretations of how affect is generated in the mere exposure effect. Key evidence for and against each interpretation is reviewed and found questionable. A new approach to study the phenomenon is suggested, building on the concept of "perceptual fluency". This approach may give new insights into how and under which conditions low-involvement exposures to stimuli in a consumer behavior context may lead to more positive affect for those stimuli.

INTRODUCTION

The mere exposure effect is introduced in many recent consumer behavior and advertising handbooks as one way in which low-involvement exposures to marketing stimuli can generate more positive affect for those stimuli. Important is that this enhancement has been observed even when those stimuli are not recognized. Although mere exposure has become part of the jargon of our discipline, its implications for consumer behavior are actually unclear. Despite all the research effort psychologists and consumer behavior researchers have devoted to the phenomenon, some authors have questioned its relevance for understanding repetition effects of marketing stimuli (e.g. Cohen and Areni 1991; Petty, Unnava, and Strathman 1991). Two decades of systematic investigations of the effect by psychologists (see Bornstein 1989 for a review) have indeed produced few insights into how and under which conditions it may operate in consumer behavior. The main reason for this knowledge gap may be that we do not have a well-accepted theory of how mere exposure can influence affect. Such a theory would help us transcend the specific manipulations in the existing studies and could lead to more systematic applied research. Unfortunately, the dynamics behind the mere exposure effect are still a matter of debate. The present paper analyzes the strongest evidence advanced by the two camps in this debate. It is argued that this evidence is inconclusive as to what the driving factors behind the mere exposure effect are. In the final section of the paper a possible new direction for future research is suggested that avoids the problems with these previous studies. The main objective of this paper, however, is to activate the debate on the antecedents of affect in the mere exposure effect. A first section introduces the antagonists and their position, and the second section gives a critique of their interpretation of the empirical evidence. The third and final section develops a new interpretation of mere exposure, based on a recent theory of recognition memory.

THE DEBATE: ONE OR TWO EVALUATION SYSTEMS

Zajonc (1968), who introduced the mere exposure effect and the associated research methodology to mainstream psychology, used it as an anchor to develop his theory about the independence of cognition and affect. He demonstrates that exposure to a stimulus is a sufficient condition for the enhancement of affect, and that this enhancement is independent of recognition (Kunst-Wilson and Zajonc 1980). This result is taken as evidence that affect and cognition are under the control of two separate, parallel, and partially independent systems. Moreover, these systems not only function independently but also attend to different stimulus features, labeled "preferanda" and "discriminanda" respectively (Zajonc 1980).

Other researchers challenge this view with the argument that the absence of recognition does not imply the absence of cognition: some other form of cognitive analysis that is not reflected in recognition and cannot be detected with existing methodologies may precede and influence the affective judgments (Lazarus 1984; Tsal 1985). Zajonc replies that this argument cannot be falsified and that therefore "such a theory would have no constraints at all and ... would be quite useless" (Zajonc, Pietromonaco, and Bargh 1982, p. 213). On the other hand, Zajonc from his part fails to specify the circumstances under which affect would be a product of the affective system alone, independently of cognition (like in mere exposure). Another issue is how changes in affect that are accompanied by recognition can be explained: can affect also in this case be the product of the affective system alone, or does the presence of recognition imply that affect is a product of (the interaction of) both systems? And if both systems play a role, what determines their respective contributions?

Zajonc's position will be referred to here as the independence hypothesis. This hypothesis gained a lot of attention because it challenges the universality of cognitive mediation theories (Obermiller 1985). The competing cognitive-affective hypothesis maintains that even in the case of mere exposure, the affective response is the last step in a series of cognitive processes (Anand, Holbrook, and Stephens 1988; Anand and Sternthal 1991).

REVIEW OF EMPIRICAL EVIDENCE

The review and discussion focus on seven studies that are representative for the two positions on the formation of the mere exposure effect. Table 1 gives a brief characterization of the seven studies, which are all seven variations on the same experimental paradigm.

Subjects are exposed to the target stimuli in the first phase of the experiment. Visual stimuli can be presented for a couple of seconds, that is in the case of words long enough to be read and pronounced, or for only a couple of milliseconds, such that subjects have the impression that they saw something without being able to identify what they saw. Auditory stimuli are in three studies presented during a dichotic listening task in which other messages are presented simultaneously with the target stimuli, each group of stimuli being presented to a different ear. In the second phase of the experiment, subjects are asked for affective and recognition judgments for the target stimuli and interspersed distractor stimuli. Only experiments where each subject gave both judgments for all the stimuli are included here, because the relation between these judgments is the issue of the debate.

The target and distractor items are in two studies presented in pairs and subjects are asked to select that item that has been presented previously and that item they like most (the order of these two pairwise choices being counterbalanced across subjects). AnotherCand for our discussion more interestingCprocedure consists of the individual presentation of targets and distractors with a recognition test of the yes/no type and affect measured on a rating scale. The data from this procedure can be analyzed by comparing the average affect ratings in the cells of table 2. One possible comparison is that of the averages for the rows or for the columns of Table 2: one can contrast the ratings of target stimuli to those of distractors, a dimension that is usually referred to as objective familiarity (Obermiller 1985), or focus on subjective familiarity, that is the difference between stimuli that subjects claim they recognize and those they claim are new. In addition to these contrasts of the marginals of Table 2, the cell values themselves can be compared. Signal-detection terminology is used here to identify those cells (cf., Anand and Sternthal 1991).

TABLE 1

DESCRIPTION OF REVIEWED EXPERIMENTS

TABLE 2

FAMILIARITY MATRIX (CLASSIFICATION OF AFFECT BY RECOGNITION CONDITION)

Evidence favoring the independence hypothesis

According to Zajonc, the firmest evidence of the independence of affect and cognition in the formation of the mere exposure effect is the finding that increased affect for a stimulus can occur as a result of a previous exposure even when recognition is only at a chance level (Zajonc, 1980). He refers to the results of Wilson (1979), where subjects' recognition for target melodies was at a chance level because their attention was focused on a different message during a dichotic listening task, and those of Kunst-Wilson and Zajonc (1980), where a presentation time of only one millisecond for visual stimuli led to the same level of recognition. There are two important problems with the way these date are interpreted, the first with the interpretation of chance recognition, the second with the use of recognition as an indicator of cognition.

Chance Recognition. The results of Zajonc and Wilson are usually cited as strong evidence for the independence position because, unlike in other demonstrations, recognition is not controlled for by a statistical procedure but by an experimental manipulation that apparently removes its effect altogether. The next step in the (implicit) argument is that because recognition is eliminated, it cannot have caused the observed changes in affect for the target stimuli, and those changes must be due to objective familiarity only. In reality, recognition can even in this case be an important determinant of affect. Chance recognition only means that recognition accuracy is at 50 percent, and "perfect" chance recognition means that there was a hit for every miss and a false alarm for every correct rejection. Even with "perfect" chance recognition every item that is a hit can be liked more than a miss, and every false alarm can be liked more than a correct rejection. Thus, even "perfect" chance recognition does not at all exclude the possibility of a "perfect" dependence of affect on recognition.

Recognition and cognition. At the time of Zajonc's and Wilson's experiments, recognition may have been the most sensitive well-accepted indicator of cognitive processing, but no longer. It has been shown, for instance, that items perceived at one point in time facilitate tasks like perceptual identification of semantically related items, lexical decisions, and word-fragment completion, and that these enhancements in cognitive processing can occur without recognition of the previously presented items. These tasks are part of a larger set of so-called indirect memory tests and measure a form of memory that has been labeled implicit, in contrast with explicit memory which is measured with the traditional recognition and recall tests (cf. Richardson-Klavehn and Bjork 1988; Roediger 1990). The results of indirect tests suggest that cognitive processing must have taken place during the exposure to the stimulus and during the indirect memory task, although there is no trace of this in terms of recognition memory. Applied to our discussion, the identification of cognition with recognition, one of the bases of Zajonc's argument, no longer seems justified in light of this recent research. Thus, it is possible that cognitive processing is responsible for the mere exposure effect.

Evidence favoring the cognitive-affective hypothesis

Anand and Sternthal (1991) recast the debate in terms of the objective-subjective familiarity distinction. They argue that in Zajonc's view affective judgments occur "automatically as a function of actual stimulus repetition or objective familiarity", while according to the cognitive model, these judgments depend on the subjective familiarity of the stimulus (p. 294). Translating these ideas into signal-detection terminology, they hypothesize that if objective familiarity is the dominant factor (which they present as Zajonc's position) the following relations should be found (see Table 2): Affect for hits > Affect for false alarms and Affect for misses > Affect for correct rejections. On the other hand, they consider a finding of Affect for hits > Affect for misses and Affect for false alarms > Affect for correct rejections as evidence for the dominance of subjective familiarity. In addition, they propose that these four tests can be replaced by one more direct test, namely a comparison of false alarms and misses, that is of stimuli that are subjectively but not objectively familiar and stimuli that are objectively but not subjectively familiar. Higher affect for false alarms would in their opinion support the cognitive-affective hypothesis. This is indeed what they find, which makes them conclude that "these observations imply that affect without recognition involves a cognitive process..." (p. 299).

Anand, Holbrook, and Stephens (1988) also present an explicit comparison of the two positions on the relation between affect and cognition, and also conclude that their "results tend to advance the cognitive-affective model over the independence hypothesis as an explanation for the formation of affective judgments" (p. 399). This conclusion is based on the finding of a significant interaction between objective and subjective familiarity (Figure 1.a), in that affect is more positive for correctly recognized items. This is interpreted as a mediating (or moderating) effect of cognition.

A closer look at these two studies suggests that the evidence in favor of the cognitive-affective model may not be as strong as the authors report in their conclusions.

Noncompeting alternative hypotheses. A first problem with both studies is that the way the hypotheses are operationalized makes them no longer mutually exclusive: accepting one of the competing hypotheses does not lead to rejection of the alternative hypothesis. This is most apparent in Anand and Sternthal's study. They start out with the four inequalities inferred from their basic hypotheses but, instead of evaluating those, they propose another supposedly more direct test. A review of the data in their Table 1, however, indicates that all four inequalities hold (see Figure 2.a) and in all likelihood are statistically significant. Following their rationale, this means that both the independence and cognitive-affective models would be consistent with the data. With respect to the second study, it is obvious from Figure 1.a that there is an interaction between objective and subjective familiarity, which indeed suggests a moderating role for cognition in the generation of affect. This figure presents the data in terms of objective and subjective familiarity. Looking at these results from a different perspective, however, brings the main effect of objective familiarity to the foreground. The two factors in Figure 2.b are objective familiarity and recognition accuracy. The latter, clearly a cognitive factor, does explain the difference between the top and bottom two data points, but leaves the main effect of objective familiarity unexplained. Thus also in this study, both the independence and cognitive-affective hypothesesCat least in the formulation given by these authorsCare confirmed.

Mere exposure and subjective familiarity. The previous discussion about the noncompeting hypotheses raises the more general issue of the value of separating the effects of objective and subjective familiarity to compare the independence and cognitive-affective models. Matlin (1971) was the first to analyze the exposure-liking relationships along the two familiarity dimensions (Figures 3a and 3b). She reports the role of subjective familiarity but the important finding from Zajonc's perspective was that of the simple main effects of objective familiarity (Zajonc 1980). It should be stressed that Zajonc acknowledged from the start the importance of subjective familiarity in the creation of affect, even under conditions of limited cognitive processing (like presentation durations of only a millisecond). Thus, the independence hypothesis was not developed to explain the complete pattern of results in the familiarity matrix, but to explain the effect of objective familiarity when subjective familiarity is controlled for. The two studies that defend the cognitive-affective model therefore cannot be considered as evidence against the independence hypothesis. Rather, they show only that subjective familiarity also plays a role. Moreover, the results of those studies actually replicate the finding that subjective familiarity does not really help in explaining the "pure" mere exposure effect. An illustration of those two points is the comparison of the results of Wilson (1979) and Anand and Sternthal (1991). The basic pattern in Figures 2 and 4 is almost the same, which is not surprising because Anand and Sternthal followed Wilson's design as closely as possible (in order to obtain a chance-level recognition). Thus, ironically the same data pattern that is usually considered as the strongest demonstration of the independence of affect and cognition also is interpreted as evidence that even affect without recognition involves a cognitive process. The four interpretation problems raised in the present paper explain this inconsistency.

A comparison of the data patterns across the different studies (see the previous Figures) raises one additional question. Although its experimental design is similar to other studies in this review, the interaction effect found by Anand et al. (Figure 1) is very distinct. How can this be explained? Obermiller (1985) notes that a subject's impression of the accuracy of a recognition response is most likely associated with positive affective feelings, which then may be transferred to the recognized stimulus. In Obermiller's case subjects had higher affect for correctly recognized objectively familiar stimuli only. Anand et al. in addition find higher affect for objectively unfamiliar stimuli that are correctly rejected. The main difference between these studies is that the stimuli in Obermiller's experiment were randomly generated sequences of tones, that probably were difficult to recognize, while Anand et al. analyzed affect for actual musical passages. In addition, Anand et al. pooled the affect ratings for those melodies and the text passages presented to the other ear in the dichotic listening task. As a result, subjects in the latter experiment probably could get a better assessment of their accuracy not only when they claimed they recognized a stimulus, but also when they claimed they did not recognize the stimulus. This may explain the exceptionally high affect for correct rejections.

PERCEPTUAL FLUENCY: A DIRECTION FOR FUTURE RESEARCH

Theories of recognition memory distinguish between two bases of recognition: familiarity and recollection (Mandler 1980). A stimulus can be recognized because it merely looks (or sounds, tastes, smells, or feels) familiar, but this impression can also be confirmed by recollecting details about the context in which it was presented, the thoughts that came to mind during its presentation, the attitudes that were formed etc. Jacoby (Jacoby, Kelley, and Dywan 1989; Kelley and Jacoby 1990) developed a theory of the antecedents of familiarity that in his opinion also may apply to the mere exposure effect (although he did not test this notion). In short, a previously presented stimulus is easier to perceive than a similar new one, and people are able to make an unconscious assessment of how fluent their perception is. In recognition, they attribute this enhanced perceptual fluency to a previous experience with the stimulus, and thus interpret it as familiarity. Jacoby suggests that the same impression of perceptual fluency may also be interpreted as positive affect and be attributed to some attractive characteristic of the stimulus. For the current discussion, this means that mere exposure would not be a sufficient condition for the enhancement of affect, as Zajonc claims, and that a certain level and form of cognitive processing are required. This cognitive processing is apparently more or less unconscious and automatic, much like how affective processing is supposed to operate in Zajonc's framework. In contrast to Zajonc, the same stimulus features may be used for recognition and affective judgments: the difference in these responses is not in what they are based on, but in what fluency is attributed to.

FIGURES 1A-4A

AFFECTIVE RATINGS SCORED BY OBJECTIVE (OLD VS. NEW) AND SUBJECTIVE ("OLD" VS. "NEW") FAMILIARITY

FIGURES 1B-4B

AFFECTIVE RATINGS SCORED BY OBJECTIVE FAMILIARITY AND RECOGNITION ACCURACY (CORRECT VS. WRONG)

Jacoby's theory may give a cognitive explanation for the creation of affect that applies to low-involvement exposures. In its present formulation it does, however, not account for the fact that affective discrimination of new and old stimuli can be better than discrimination by recognition. This difference in performance is most obvious from the simple main effect of objective familiarity for stimuli that are not subjectively familiar, which was referred to above as a "pure" mere exposure effect (see any of the figures), and from Seamon et al's (1983) comparisons. These authors used paired-choice measures for recognition and affect and observed a "better" performance for affective judgments, in that previously presented stimuli were more often selected as liked than as recognized. Zajonc explains these differences as resulting from a higher sensitivity of the affective system. Another explanation, more in line with Jacoby's reasoning, is that people use their assessments of perceptual fluency differently depending on the type of judgment they have to make. Assume that subjects act as if they set some criterion level of fluency that has to be exceeded for a positive response. Because a recognition response is evaluated on accuracy, it would be natural to use a more stringent criterion than for affect where, from the perspective of the subject, no mistakes can be made. Thus, the difference between recognition and affect would be in the level of the decision criterion used to interpret perceptual fluency.

From this discussion it appears that the main difference between Zajonc's independence framework and the cognitive-affective model based on Jacoby lies in their account of the different discrimination performance for recognition and affect. For Zajonc, this is also a matter of different sensitivity of the two independent systems involved, while for the framework based on Jacoby, this is a result of two different decision criteria, one for affect and one for recognition, used by the cognitive system. Signal-detection analysis was developed to measure sensitivity of and decision criteria used by detection systems, and it may therefore be possible to develop statistical tests that make a more direct comparison of the independence and cognitive-affective hypotheses than the ones reviewed in the previous sections.

IMPLICATIONS FOR CONSUMER BEHAVIOR AND MARKETING

The framework derived from Jacoby's theory about the attribution of perceptual fluency not only introduces a new perspective on the generation of affect and relationship of affect and recognition, but also can be directly contrasted with Zajonc's conceptualization. Moreover, signal-detection analysis brings new methodological leverage to the debate because it is based on a process model of judgments, which has proven its value in many different contexts.

Demonstrating that the perceptual fluency framework can account for the mere exposure effect not only is theoretically relevant but also has practical utility. The previous section introduced the distinction between recognition by familiarity and recognition by recollection. The perceptual fluency framework implies that conditions that enhance the former should also enhance the mere exposure effect. In contrast, conditions that make the latter type of recognition more likely may make it less likely to observe a mere exposure effect. The argument here is that although mere exposure is a robust phenomenon, it is relatively small compared to the effect of retrieved attitudes, and the contribution of mere exposure to affect formation is therefore probably minor when attitude retrieval is possible. All this means that existing research on the two bases of recognition may help in identifying the conditions that make a mere exposure effect more likely. Interestingly, there seems to be considerable agreement on which factors affect each of the types of recognition (e.g., Gardiner and Java 1991; Jacoby and Dallas 1981). These factors should be examined in a consumer behavior or advertising context.

In conclusion, even if perceptual fluency may not be the full answer to the cognitive-affective debateCmaybe no theoretical argument or empirical test can resolve the semantical and philosophical disagreements surrounding this discussionCit at least provides a research agenda for applied research on the mere exposure effect.

REFERENCES

Anand, Punam, Morris B. Holbrook and Debra Stephens (1988), "The Formation of Affective Judgments: The Cognitive-Affective Model Versus the Independence Hypothesis," in Journal of Consumer Research, 15 (December), 386-391.

Anand, Punam and Brian Sternthal (1991), "Perceptual Fluency and Affect Without Recognition," in Memory & Cognition, 19 (3), 293-300.

Bornstein, Robert F. (1989), "Exposure and Affect: Overview and Meta-analysis of Research, 1968-1987," in Psychological Bulletin, 106 (September), 265-289.

Cohen, Joel B. and Charles S. Areni (1991), "Affect and Consumer Behavior," in Handbook of Consumer Behavior, eds. Thomas S. Robertson and Harold H. Kassarjian, Englewood Cliffs, New Jersey: Prentice-Hall, 188-240.

Gardiner, John M. and Rosalind I. Java (1991), "Forgetting in Recognition Memory With and Without Recollective Experience," in Memory & Cognition, 19 (6), 617-623.

Jacoby, Larry L. and Mark Dallas (1981), "On the Relationship Between Autobiographical Memory and Perceptual Learning," in Journal of Experimental Psychology: General, Vol.110, No.3, 306-340.

Jacoby, Larry L., Colleen M. Kelley and Jane Dywan (1989), "Memory Attributions," in Varieties of Memory and Consciousness: Essays in Honour of Endel Tulving, eds. H.L.Roediger, III and F.I.M. Craik, Hillsdale, NJ: Erlbaum.

Kelley, Colleen M. and Larry L. Jacoby (1990), "The Construction of Subjective Experience: Memory Attributions," in Mind & Language, Vol.5, No.1, 49-68.

Kunst-Wilson, William R. and Robert B. Zajonc (1980), "Affective Discrimination of Stimuli That Can not be Recognized," in Science, 207 (February), 557-558.

Lazarus, Richard S. (1984), "On the Primacy of Cognition," in American Psychologist, 39 (2), 124-129.

Mandler, George (1980), "Recognizing: The Judgment of Previous Occurrence," in Psychological Review, Vol.87, No.3, 252-271.

Matlin, Margaret W. (1971), "Response Competition, Recognition, and Affect", in Journal of Personality and Social Psychology, 19, 3, 295-300.

Obermiller, Carl (1985), "Varieties of Mere Exposure: The Effects of Processing Style and Repetition on Affective Response," in Journal of Consumer Research, 12 (June), 17-30.

Petty, Richard E., Rao Unnava and Alan J. Strathman (1991), "Theories of Attitude Change," in Handbook of Consumer Behavior, eds. Thomas S. Robertson and Harold H. Kassarjian, Englewood Cliffs, New Jersey: Prentice-Hall, 241-280.

Richardson-Klavehn, Alan and Robert A. Bjork (1988), "Measures of Memory", Annual Review of Psychology, 39, 475-543.

Roediger, Henry L., III (1990), "Implicit Memory. Retention Without Remembering," in American Psychologist, Vol. 45, No. 9, 1043-1056.

Seamon, John G., Nathan Brody, and David M. Kauff (1983), "Affective Discrimination of Stimuli That Are Not Recognized: Effects of Shadowing, Masking, and Cerebral Laterality," in Journal of Experimental Psychology: Learning, Memory, and Cognition, Vol.9, No.3, 544-555.

Tsal, Yehoshua (1985), "On the Relationship Between Cognitive and Affective Processes: A Critique of Zajonc and Markus," in Journal of Consumer Research, 12 (December), 358-362.

Wilson, W.R. (1979), "Feeling More than We Can Know: Exposure Effects without Learning", in Journal of Personality and Social Psychology, 37, 811-821.

Zajonc, Robert B. (1968), "The Attitudinal Effects of Mere Exposure", Journal of Personality and Social Psychology Monograph, 9, (No. 2, Part 2).

Zajonc, Robert B. (1980), "Feeling and Thinking: Preferences Need No Inferences," in American Psychologist, 35 (2), 151-175.

Zajonc, Robert B., Paula Pietromonaco, and John Bargh (1982), "Independence and Interaction of Affect and Cognition", in Affect and Cognition: the Seventeenth Annual Carnegie Symposium on Cognition, eds. Margaret S. Clark and Susan T. Fiske, Hillsdale, NJ: Erlbaum, 211-227.

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