Why Familiar Stimuli Are Better Liked. a Study on the Cognitive Dynamics Linking Recognition and the Mere Exposure Effect

Marc Vanhuele, Groupe HEC
ABSTRACT - If a brand that was advertised is encountered at a later point in time, the original presentation(s) to and evaluation of the advertisementCif ever formedCmay not be retrievable from memory, but the brand itself may still look familiar, which may lead to recognition and liking. The cognitive dynamics underlying these effects are examined using a probabilistic model that describes the processes generating the recognition and liking responses.
[ to cite ]:
Marc Vanhuele (1995) ,"Why Familiar Stimuli Are Better Liked. a Study on the Cognitive Dynamics Linking Recognition and the Mere Exposure Effect", in NA - Advances in Consumer Research Volume 22, eds. Frank R. Kardes and Mita Sujan, Provo, UT : Association for Consumer Research, Pages: 171-175.

Advances in Consumer Research Volume 22, 1995      Pages 171-175

WHY FAMILIAR STIMULI ARE BETTER LIKED. A STUDY ON THE COGNITIVE DYNAMICS LINKING RECOGNITION AND THE MERE EXPOSURE EFFECT

Marc Vanhuele, Groupe HEC

ABSTRACT -

If a brand that was advertised is encountered at a later point in time, the original presentation(s) to and evaluation of the advertisementCif ever formedCmay not be retrievable from memory, but the brand itself may still look familiar, which may lead to recognition and liking. The cognitive dynamics underlying these effects are examined using a probabilistic model that describes the processes generating the recognition and liking responses.

INTRODUCTION

In the typical ad pretest design with recognition, all subjects are first exposed to the ad and after some retention interval receive a portfolio with a set of ads from which they have to select the one(s) they saw during the first phase of the test. In a recognition-based campaign evaluation, people are contacted on a random basis, presented with one or a set of ads, and asked whether they recognize the target ad(s). These two tests involve different types of recognition that may partly be affected by different factors because they depend on different cognitive processes. For that reason they may also have a different impact on evaluation and choice.

In the first type of recognition test the recognition question explicitly or implicitly refers to the episode in which the ad was originally presented. The respondent may then for instance recall the thoughts and evaluations she had during the initial presentation and based on this information confirm that she saw the ad during the first phase of the test. In the second type of test, usually no reference is made to a specific exposure. The recognition decision is, therefore, more likely to be based on a mere feeling that the stimulus is familiar. The respondent then infers from that feeling that she must have seen the ad before. Dual-process models of recognition make the distinction between the two bases of recognition illustrated here and label them retrieval and familiarity (Mandler 1980). Crucial for the present project is that these bases are independent pieces of information. Familiarity is not simply based on a "weaker" representation of the information in memory: the two types of recognition rely on different qualities of the memory representation.

The distinction between the two types of recognition memory has not yet been examined in our domain (cf., Taschian, White, and Pak 1988 and Alba and Hutchinson 1987 for perspectives on recognition). Our current recognition tests of ad and brand awareness, for instance, are almost exclusively focused on retrieval-based recognition. On the one hand, it may be argued that involvement in the typical advertising experiment and pretest is higher that in a natural context (cf., Hawkins and Hoch 1992) and that the frequency of retrieval-based recognition is therefore overstated. Consumers may in reality not be able or not be willing to find out why they recognize a brand. On the other hand, this focus on retrieval-based recognition may understate the impact of advertising (Pluzinski 1992).

Familiarity is evaluated positively, and this positive affect may be transferred to the familiar marketing stimulus (Janiszewski 1993), a phenomenon that is usually referred to as the mere exposure effect. It has been argued that familiarity and liking (in the mere exposure effect) in general covary because they share a common antecedent and are generated by similar cognitive processes (Anand and Sternthal 1991; Jacoby and Kelley 1987, Janiszewski 1993). It will be argued here that retrieval-based recognition not only is not necessary for the effect, but even may counteract it.

The present project examines both the distinction between the two types of recognition memory and the link between familiarity-based recognition and the mere exposure effect. Even low-involvement advertising exposures may be sufficient to generate the effects studied here. To examine these relatively automatic (and maybe even unconscious) responses to advertising, a methodology is used that relies to a minimal extent on introspection and verbalization of the process that generates the responses. This methodology instead infers what the cognitive process must have been like to produce a certain set of responses.

TWO BASES OF RECOGNITION

Theories of recognition memory distinguish between two bases of recognition, familiarity and retrieval, that can contribute independently and additively to a recognition judgment (Mandler 1980). Noticing a given brand on the store shelves, for instance a brand in an infrequently purchased product category, may trigger a feeling of knowing that brand. This response is relatively automatic and effortless: the brand seems to "jump out" and present itself as familiar. As a result, we may decide that we recognize the brand. We may also, in addition, realize that we recently saw a commercial for that brand, and thus understand why the brand looks familiar. Retrieving scenes from that commercial then confirms our feeling of knowing the brand.

The two bases of recognition depend on different qualities of the memory representation. In Mandler's terminology, familiarity mainly results from intraitem integration, while retrieval is a function of interitem elaboration. The latter refers to the associations in long-term memory: when there are more and stronger associations to other stimuli and events, it is easier to recognize a target stimulus on the basis of retrieval. Intraitem integration is the extent to which the elements or features of a stimulus are integrated. Sheer exposure and repetition make the different elements of an item seem more coherent and integrated, and thereby make it look more familiar.

PERCEPTUAL FLUENCY: AN ANTECEDENT OF FAMILIARITY AND LIKING

Mandler leaves the issue of the psychological processes that generate feelings of familiarity unaddressed. Larry Jacoby and his colleagues (Kelley and Jacoby 1990; Jacoby, Toth, Lindsay, and Debner 1992) advance the notion that both familiarity and recognition should be conceptualized as inferences. Familiarity and the results of possible retrieval are among the cues that inform a recognition inference. A stimulus is judged old when it generates a sufficiently strong feeling of familiarity and/or when enough details about a previous occurrence can be retrieved. Familiarity itself is, at least partly, based on what Jacoby et al. define as "perceptual fluency", that is the ease with which a stimulus is processed perceptually. They also argue that affective evaluations may as well, at least in part, be based on perceptual fluency. This common antecedent may explain the link between familiarity and liking in the mere exposure effect. The argument is basically that perceptual fluency is a vague unspecific sensation that something is different about an item. This feeling may then be attributed to a range of qualities of the item, for instance that it is true, that it is fitting in a certain context, that it has been encountered before, or that it is likeable (cf., Anand and Sternthal 1991; Hawkins and Hoch 1992; Janiszewski 1993).

THE INVARIANCE OF FAMILIARITY-BASED RECOGNITION

The low-involvement nature of many advertising exposures and purchasing occasions may inhibit recognition by retrieval because there is an insufficient quantity and quality of processing, either of the original exposures to the advertisements, or of the later exposure to the product when an evaluation has to be made. The following review of factors that influence the two types of recognition confirms this idea but also suggests that the typical low-involvement ad exposure may still generate enough perceptual fluency for familiarity and a mere exposure effect.

Recent studies have found that quite a few factors influence recognition by retrieval but not recognition by familiarity. Gardiner and his co-workers found that recognition based on familiarity was not affected by levels of processing, dividing attention during study, passive reading versus actively generating the target information, and time delays of up to one week (see the summary of this work in Gardiner and Java 1991). On the other hand, each of those factors had a significant influence on recognition with retrieval: lower levels of processing, divided attention, passive reading, and a test delay lead to weaker recognition performance by retrieval. Jacoby and his colleagues (Jacoby et al. 1992) found that manipulating attention to the target stimuli during the recognition test and inducing different levels of processing affected recognition based on retrieval, but had no influence on familiarity-based recognition.

In summary, recognition by retrieval apparently depends on a minimum amount and certain quality of processing at exposure and test that recognition without retrieval does not require. Within a certain time frame, the latter type of recognition also shows little deterioration compared to the first type. Thus, the low-involvement conditions that characterize many advertising exposures may on the one hand lead to the inhibition of more conscious, deliberate attitude formation and retrieval, but on the other hand still generate familiarity, and a mere exposure effect. This also means that the conditions under which familiarity, as opposed to retrieval, becomes the dominant base in recognition should make the mere exposure effect a more important contributor to affect. With recognition based on retrieval, evaluations that were formed at the time of exposure are retrieved or new evaluations are formed based on retrieved information. These evaluations probably overwhelm the contribution of the mere exposure effect to affect. This hypothesis can be tested on any of the factors that have been shown to affect retrieval and not familiarity. In the present study one factor, the passage of time, is examined. A time delay between exposure and test should enhance the mere exposure effect, because such a delay has relatively little effect on familiarity and reduces the ability to retrieve. The test proposed in the next section not only tracks the size of the mere exposure effect over time, but also links the effect to the bases of recognition.

A MODEL OF DUAL-PROCESS RECOGNITION AND MERE EXPOSURE

Description

Assume that a person is presented with a set of previously encountered marketing stimuli and is asked to retrieve from memory her initial evaluation of those stimuli. The framework described in the previous section predicts that stimuli that were liked initially can be categorized again as liked on the test for three reasons:

- the initial evaluation is retrieved,

- the same evaluation is formed again, or

- the stimuli have become perceptually fluent and this fluency is interpreted as a positive evaluation (i.e., the mere exposure effect).

Stimuli that were not liked before may be categorized as liked as a result of the mere exposure effect. Detecting the impact of mere exposure on affect depends on two factors:

- perceptual fluency remains stable in the time frame consideredCwhich is what we expect from research on the invariance of familiarityC, and

- retrieval is suppressed over time.

Assume now that the retrieval-of-evaluation test just discussed is combined with a "yes/no" recognition test. New and old stimuli are presented and subjects are instructed that when they give a positive recognition response, they also have to tell if they initially liked or disliked the stimulus. Figure 1 gives the processing trees that are hypothesized to underlie the response behavior for this task. For each type of stimulus (initially liked, initially disliked, and those that have not been seen before) there is a processing tree. The labels on the branches of the trees are the parameters representing the probability of the cognitive events described above. For each type of stimulus there are three possible responses ("Liked", "Disliked", and "Not Seen"). These responses are the result of the different cognitive events. R and F are the probabilities that retrieval and familiarity are sufficient for recognition responses, and PF is the probability that perceptual fluency is sufficient for a mere exposure effect. In the absence of retrieval and sufficient fluency, people have to resort to guessing to produce a response. In the first place they have to guess the exposure status and, in case they guess that the stimulus is old (which they do with probability gs), also have to guess their initial affective reaction (they will guess it was favorable with probability gl).

The model (and corresponding processing tree) represent how each response flows from a (series of) cognitive event(s). For instance, a "like" response for a previously disliked stimulus can result from two strings of events. In any case, a "like" response indicates that the initial negative evaluation was not retrieved from memory (1-R). The response can be either based on fluency or, in case there is insufficient fluency for a feeling of familiarity (1-F), on guessing. In the first case, the liking response would only be given if the item looks familiar (F) and generates sufficient perceptual fluency for a mere exposure effect (PF). In the second case, the response is the result of two guesses, first that the item has been seen before (gs), second that it was liked (gl). In mathematical notation:

P("L"|NL)=(1-R)F*PF + (1-R)(1-F)gsgl

This equation also makes clear that the model (unlike what the tree representation may suggest) does not make any assumptions about the sequence of the different cognitive events. Similar descriptions and equations can be formed for each stimulus/response combination. This gives nine equations relating responses to parameters that represent hypothetical cognitive events.

As explained, conditions that have been shown to affect retrieval negatively but that leave familiarity intact should produce a stronger mere exposure effect. This prediction can be evaluated on the basis of a comparison of the estimates for the parameters R, F, and PF, for conditions that are and are not expected to reduce retrieval.

FIGURE 1

COGNITIVE PROCESSING TREE FOR EXPERIMENT

Parameter Estimation

Multinomial modeling is used to estimate the parameters (Batchelder and Riefer 1990). In short, the responses in the experiment described above are the result of a series of dichotomous events, and the 5 parameters are the probabilities of each event occurring. If responses for each stimulus type are independent and identically distributed across the three response types, the joint distribution of data for a given stimulus type is given by the trinomial model. Each p-value can itself be written in terms of the cognitive parameters (cf., Eq. 1). If in addition there is independence between stimulus types, then the complete data pattern can be described by the product of the three trinomial distributions. This product gives the likelihood equation that can be maximized to get the parameter estimates. Multinomial modeling, thus, basically infers the probabilities of a series of cognitive events from estimates of the response probabilities across a large number of stimuli. Comparing the corresponding parameters, especially R, F and PF, across different delay conditions that are expected to influence the likelihood of observing mere exposure effect is the objective of the following pilot study.

DESIGN OVERVIEW

Subjects were 42 undergraduate students. The study was presented to them as a search for the best-liked product logos. They received a folder with unknown trademarks and were asked to indicate which trademarks they liked or disliked. In order to enhance the statistical power of the analyses, they were asked for a roughly equal number of likes and dislikes. They were instructed to familiarize themselves with the full set first and to only then write down their evaluations. This gave each trademark two exposures during this phase of the experiment. In case they really could not make up their mind, they could check off a third, neutral, response alternative. Stimuli receiving this response were eliminated from the analysis. The evaluation task took about 7 minutes.

TABLE 1

MAXIMUM-LIKELIHOOD ESTIMATES (STANDARD ERRORS BETWEEN PARENTHESES)

After a retention interval, subjects saw a folder in which the previously evaluated trademarks were mixed with new ones. They were asked for their recognition and for their initial evaluation in case they indicated they recognized the trademark. This phase took between 4 and 6 minutes.

Stimuli

A set of 81 unknown trademarks were collected from different published sources. To counterbalance item-specific effects, four different selections of 54 stimuli were made for the first phase of the experiment. The remaining 27 functioned as distractors in the folder used in the second phase.

Response Measures

In the first phase, responses were collected on a two-page list with 54 3-alternative scales ("liked", "?", "disliked"). In the second phase the experimenter checked off the responses to the 81 stimuli from a list. Subjects could answer with "liked", "did not like" or "did not see". This way of response collection prevented subjects from adjusting previous responses to end up with the desired number of responses in each category (they knew that of the recognized logos about half received a liking response in the first phase).

Independent Variable

The retention interval between the two phases was either about 30 minutes or 1 day. In both cases the recognition test came unanticipated to minimize the possibility of subjects memorizing the material and responses. Following the first phase, subjects were asked to make a choice between continuing the study right away (they anticipated a duration of 50 minutes) or returning the next day for a session of about 5 minutes. If they chose the second option, they would also enter a lottery for $175. Of the 42 subjects, 13 chose the second option.

Results

The parameter estimates are given in Table 1. Differences between the parameter estimates for the two retention intervals were tested by comparing the difference in likelihood-ratio statistics G2 between the unrestricted model and a model that restricts the parameter of interest to be the same across the retention intervals. The values of the guessing parameters are not essential to the hypothesis, but have face validity and, thus, support the model itself. In both conditions the absence of retrieval and familiarity apparently is a reliable cue for the recognition judgment and when both are absent people rarely guess that a stimulus is old (gs=0.10). Although the maximum-likelihood estimate is higher for the one-day delay (0.13 vs. 0.11), the difference is not significant (DG21=2.01, p<.15). This result may indicate that even at the longer delay in this experiment, the presence or absence of familiarity and retrieval are still a strong informative cue in the recognition judgments. In case people guess they saw the stimulus before, the theory would predict that because there is no familiarity information, the level of perceptual fluency is also not sufficiently high to be a basis for a liking judgment. People are therefore expected to randomly pick a liking response. The parameter gl is indeed close to 0.5 in the short-delay condition and that for the long-delay condition is not significantly different from the first (DG21=0.56, p<.45). As hypothesized, the retention interval of one day considerably inhibits the possibility of retrieval (0.25 vs. 0.57, DG21=74.04, p<.0001), compared to the 30-minute retention interval. Contrary to the predictions, however, the probabilities that perceptual fluency exceeds the decision criteria for familiarity and liking do not remain stable but increase (for F: DG21=16.1 p<.0001; for PF: DG21=4.52 p<.04). One possible explanation for these results is the small number of subjects in the long-delay condition. To check to what extent the responses of specific individuals (possibly outliers) drive the results, a jackknife-like estimation was carried out. The results were, however, confirmed. A second explanation is that familiarity was not always used as a cue in the short-delay condition (even though it would have been an informative cue) because retrieval was relatively easy. The familiarity (and mere exposure) parameter may therefore have been underestimated for this condition. A third explanation is that the specific model formulation adopted here drives the results. More research is needed on this explanation. Applying the procedure used by Jacoby (1992), however, led to the same substantive findings.

To conclude, the results of this pilot study certainly call for a replication study, and for experiments on the effect of other processing factors on the contributions of familiarity and retrieval to liking. More work is also needed on the role of the specific model used to estimate the results, and on the assumptions on which the model is based. Some of these assumptions may be testable independently of the model developed here.

CONCLUSION

There is almost no research in our discipline on the effects of low-involvement advertising exposures. Even the low-involvement conditions in the typical advertising response experiment "still seem reasonably involving compared with the natural conditions under which consumers inattentively process advertising information" (Hawkins and Hoch 1992, p. 14). In addition, most of our research methods and instruments are geared toward introspective verbal descriptions of cognitive and affective responses to advertising, and these may be less appropriate for studying low-involvement processes.

The theories presented here suggest that low-involvement exposures are conducive to a mere exposure effect, because even this type of presentation has been shown to enhance familiarity-based recognition, and because this type of presentation is expected to inhibit retrieval-based recognition. In addition, the passage of time makes it more likely to observe a mere exposure effect because retrieval, unlike familiarity, is suppressed over time. The present study presents an experimental procedure that not only tests the prediction for the effect of time on the mere exposure effect, but that also, and more importantly, allows to verify if and to what extent the effect is produced by the inhibition of retrieval and stability of familiarity over time. This procedure, which can be applied to examine a wide range of possible mediators of the mere exposure effect, has two important advantages. A problem with the explanatory variables in this research is that they usually have parallel effects. Familiarity and retrieval both lead to a positive recognition response, and the mere exposure effect and retrieval of a previous positive evaluation both lead to a liking response. A first advantage of the procedure developed in this paper is its power to separate the contribution of these different parallel factors to a given response. Second, because familiarity and mere exposure involve relatively automatic, and thus relatively unconscious processes, people may not be able to report on them. The procedure proposed here addresses this issue by inferring the probability that a response originated from a certain factor, instead of directly asking for the source of the response. The main contribution of the present study is the development of the combination of an experimental procedure and a multinomial model that allows us to study the dynamics underlying relatively automatic cognitive processes.

REFERENCES

Alba, Joseph W. and J. Wesley Hutchinson (1987), "Dimensions of Consumer Expertise," Journal of Consumer Research, 13 (March), 411-454.

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

Batchelder, William H. and David M. Riefer, "Multinomial Processing Models of Source Monitoring," in Psychological Review, Vol.9, No.4, 548-564.

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

Hawkins, Scott A. and Stephen J. Hoch (1992), "Low-Involvement Learning: Memory without Evaluation," in Journal of Consumer Research, 19 (September), 212-225.

Jacoby, Larry L. and Colleen M. Kelley (1987), "Unconscious Influences of Memory for a Prior Event," in Personality and Social Psychology Bulletin, Vol.13, No.3 (September), 314-336.

Jacoby, Larry L., Jeffrey P. Toth, D. Stephen Lindsay and James A. Debner (1992), "Lectures for a Layperson: Methods for Revealing Unconscious Processes," in Perception without Awareness. Cognitive, Clinical, and Social Perspectives, eds. Robert Bornstein and Thane S. Pittman, New York: The Guildford Press.

Janiszewski, Chris (1993), "Preattentive Mere-Exposure Effects," in Journal of Consumer Research, December, 376-392.

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

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

Pluzinski, Carol (1992), The Partial Activation of Inaccessible Brand Names in Consumer Memory, Working Paper Mark-92-9, Department of Marketing, Stern School of Business, New York University, New York, NY 10003.

Tashchian, Armen, J. Dennis White and Sukgoo Pak (1988), "Signal Detection Analysis and Advertising Recognition: An Introduction to Measurement and Interpretation Issues," in Journal of Marketing Research, Vol.15 (November), 397-404.

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