This form of advertising has been around for a very long time. Clearly, advertising executives must think it works, since much effort and expense is devoted to these forms of highly-visible product placement. This leaves us with two questions: Does it really work? If so, why? For answers to these questions, we turn to psychological research on the "mere-exposure effect." As we shall see, this research gives us evidence that, within certain limitations, mere exposure to a stimulus can indeed lead to a more positive evaluation of that stimulus. Thus, seeing the Coca-Cola can over and over, may lead to a more positive assessment of Coke, and, perhaps, a greater inclination to buy Coke. However, there are no data to show how well this effect applies to the advertising industry. To find out, we need to take a closer look at the underlying nature of the mere-exposure effect.
In his monograph, Zajonc (1968) launched the mere-exposure line of research with a series of studies upon which almost all successive mere-exposure experiments have been based. The first two experiments exposed subjects to novel stimuli various numbers of times, ranging from 0 to 25 times. In one case, the stimuli were nonsense words made from English letters, which subjects were told were Turkish adjectives, and in the other case the stimuli were simulated Chinese ideographs. After the exposure, subjects were tested on the stimuli they had just seen:
Their next task would be to guess what [the stimuli] meant. The experimenter told the subject how nearly impossible this task was, and he therefore did not require him to guess the word meanings exactly. Instead, it would suffice if the subject indicated on a 7-point (0 to 6) good-bad scale whether each word meant something good or something bad and to what extent, because these Turkish adjectives all meant something good or bad. (Zajonc, 1968, p. 14)
Both experiments found that subjects reliably attributed more positive meanings to the more frequently presented stimuli. From zero to 25 exposures, subjects' ratings increased by over one whole point on the 7-point scale.
Zajonc believed that these indirect evaluations accurately represented subjects' liking for these stimuli, but he wanted to be sure. In even stronger tests, he went on to report similar studies that asked directly for subjects' opinions of the stimuli, asking subjects how much they actually liked the stimuli. He also performed a study that monitored subjects' physical arousal, in response to the stimuli, through skin conductance. In all cases affective ratings -- both self-reported and physiologically-measured -- increased with exposure.
In his meta-analysis of two decades of experimentation on this effect, Bornstein (1989) examined several aspects of the body of mere-exposure research. He reported that many factors influence the mere- exposure effect and, in what follows, I offer a partial catalog of these factors. This should convey a broad profile of this phenomenon, and will also serve as a data-base, when we turn, later on, to the task of evaluating candidate explanations for this phenomenon. The stimulus variables that Bornstein examined were stimulus type, and stimulus complexity. Presentation variables include the number of exposures, exposure sequence, exposure duration, and stimulus recognition. Measurement variables include type of affect measure, which he found to make no difference on the mere-exposure effect, and delay between exposure and rating.
This aspect of the exposure effect has always been a great source of controversy for theorists. Every theory that has been proposed has been centered around the nature of this curve. We will return to this effect a little later when we discuss the theories that have been proposed to explain the mere-exposure effect.
Bornstein, however, found a clear result in his meta- analysis. He compared studies that had used an heterogeneous exposure sequence with those that used an homogeneous exposure sequence. What he uncovered was that studies that used heterogeneous presentation sequences produced moderate exposure effect sizes. Yet, those that used homogenous presentation sequences produced no reliable effect at all. Thus he concluded that, in spite of the mixed results of the few studies that directly compared sequence, heterogeneous exposure was much more successful at eliciting the mere- exposure effect.
To examine exposure duration in more detail, Bornstein's meta-analysis compared studies that exposed subjects in one of five categories: less than 1 second, 1 - 5 s, 6 - 10 s, 11 - 60 s, and over 60 s. He found that exposure durations less than 1 second produce a strong mere-exposure effect size. Exposure durations longer than that produce small effect sizes.
For his meta-analysis of this factor, Bornstein divided experiments into four categories: affect rating immediately follows each exposure; affect ratings immediately follow all exposures; affect ratings follow exposures after a delay; and naturalistic studies where the affect ratings were made from stimuli exposed in everyday life. Experiments in the first category produced the weakest overall exposure effect size. Experiments in the second and third categories produced increasingly stronger effects, and the strongest exposure effect was obtained in experiments that used a naturalistic design.
A classic example is the response competition theory. Research had previously shown that novel stimuli often provoke some sort of exploratory behavior (Berlyne, 1954). In order to relate this result to the mere-exposure effect, Harrison (1968) proposed the idea of response competition. In this theory, novel stimuli would provoke a generalized response, based on all of our previous exposure to similar stimuli. However, the potential specific responses generated could be in conflict with each other. This internal conflict would lead to a state of negative affect. Continued exposure would strengthen some responses and weaken others. This would result in less conflict, and thereby a decrease in negative affect, which we would detect as the mere-exposure effect. At the same time, our desire to reduce this internal conflict is what would cause us to want to explore this novel stimulus.
As more research was done on the effect, deficits in this theory appeared. This theory implies that the mere-exposure effect works by turning initially-negative stimuli into neutral ones, rather than neutral-stimuli becoming positive. Yet, most mere-exposure experiments have found that subjects' ratings start in the neutral to mildly positive range, and increase to the moderately positive (again, see Bornstein, 1989).
This sets the pattern for most theories in this domain. While the theories were being proposed, the body of research was continuously growing, and more and more results had to be explained. In consequence, each theory first gained, then eventually lost support among researchers. For example, some early researchers believed that the exposure effect was the result of demand characteristics of the experimental situation. However, altering the form of the experiment to hide the experimental demands from the subjects showed that this was not the case (for a discussion, see Eagly & Chaiken, 1993; Wilson, 1979). Likewise, some theorists have tried to show that the effect is the result of persuasive processes (Cacioppo & Petty, 1979). But again, later research found that the effect could occur even without conscious recollection of the stimuli (e.g. Kunst-Wilson & Zajonc, 1980).
One of the most widely accepted early theories was the two- factor account (Berlyne, 1970; Bornstein, Kale, & Cornell, 1990), which hypothesizes that attitudes towards a stimulus initially vary due to the interaction of positive habituation (uncertainty or conflict reduction) and boredom. During the beginning exposure phase, positive habituation predominates, and initially neutral stimuli are liked better. With continued exposure, however, boredom with the stimulus will counteract the positive habituation. As this counteraction occurs, we see a decrease in the effectiveness of exposure to increase affect ratings. This theory actually fit the data quite well for a long time, and some modern theorists have proposed a modified version of this theory, which we shall discuss shortly.
There is some controversy about this research, however. Some experiments have failed to replicate the mere-exposure effect in circumstances that decreased stimulus recognition (Fox & Burns, 1993; Stading & Thompson, 1990). These researchers claimed that the results originally published showing that mere-exposure effect could occur without stimulus recognition were actually the exception, not the rule. They believed that many other researchers had found that the mere-exposure effect did not occur when subjects could not recognize the stimuli, but, as this is considered a null result, the studies had not been published.
In the last few years, however, the exposure effect without recognition experiments have been more consistently replicated (Bornstein, 1992; Bornstein & D'Agostino, 1992). These successful replications have shifted the dispute from whether to accept the possibility of the mere-exposure effect without recognition to how this result should be interpreted (Todman, 1989). Several new theories have developed attempting to explain the mere-exposure effect while accounting for the lack of stimulus recognition and the possibility of subliminal presentations. Yet, even these newer theories often miss explaining some of the data.
When the first subliminal mere-exposure experiment was conducted (Kunst-Wilson & Zajonc, 1980), Zajonc (1980) proposed a new hypothesis for the effect. He argued that there are two separate streams of processing in humans at all times: the affective and the cognitive. The idea was that these streams worked separately from each other, thereby allowing for the possibility that the mere-exposure effect was the result of processing of incoming stimuli by the affective system, and thus could occur independently of explicit recognition of the stimuli.
Other theorists have criticized this approach, mainly on the grounds that there is a body of literature that shows that cognitive processes can also occur automatically (see Todman, 1989). As well, some researchers have empirically demonstrated that the mere-exposure effect behaves in a manner consistent with various forms of unconscious cognitive learning. For example, Gordon and Holyoak (1983, Experiment 1) exposed subjects to stimuli that conformed to a specific grammatical system, then tested their liking of novel grammatical and nongrammatical stimuli. They found that subjects reliably liked the grammatical stimuli better than the nongrammatical. Thus, the mere-exposure effect shows a generalization gradient guided by cognitive variables (in this case, a rule system), just as other instances of unconscious learning do.
Other theorists have attempted to modify older theories to account for the newer data. Bornstein (1989) has proposed that the two-factor theory discussed briefly above can be modified to include a form of "unconscious learning." This model builds on theories about the contrast between preconscious and conscious processing (Kihlstrom, 1987).
We will not pursue this idea here, because of a conspicuous concern about the theory. Bornstein relies very heavily on the subliminal exposure-time research. As he argues, it is plausible that these subliminal exposure tasks do interfere with subjects' ability to recall explicitly the exposed stimuli. In this situation, bypassing these automatic processes increases the mere-exposure effect. Clearly, however, this interference occurs at the time of exposure. However, there are alternative ways of accomplishing this interference. If we look at a different task, a distraction such as a verbal shadowing task (for example, Wilson, 1979) at the time of test, we see a similar result, but with what must be a different basis. In this case, there is no difference in exposure between this experiment and a typical mere-exposure experiment, in which subjects get a chance to recognize the exposed stimuli. Thus, this experiment could not block these automatic processes. Instead, this paradigm disrupts conscious retrieval of the memory at a later time. Yet, this situation has been shown effectively to enhance the mere- exposure effect.
Most recently, some researchers have suggested that "perceptual fluency" (Jacoby & Dallas, 1981) is the basis for the mere-exposure effect (Bornstein, 1992). In this theory, previous exposure to a stimulus makes subsequent perceptions easier. This ease of perception is attributed by the perceiver as a liking for the stimulus, resulting in the mere-exposure effect. We will deal with the idea of perceptual fluency in terms of implicit memory in our discussion of implicit memory below. At this point, suffice to say that no theorist has adequately empirically explored this relation between perceptual fluency and the mere-exposure effect.
This thesis seeks to examine the mere-exposure phenomenon from a different point of view. Drawing on claims about implicit memory, this work seeks to connect implicit-memory effects with the mere-exposure effect. From this new perspective, we may be able to glean a common explanation for both mere exposure and implicit memory, giving us a better understanding of the processes behind both of these effects. Before we attempt to discover what this connection is, however, we need to have an understanding of the memory effects we are talking about.
As with the mere-exposure effect, no consensual theory of implicit memory has yet been devised. However, numerous ideas have been proposed. Reisberg (1995, ch. 5) presents an overview of these ideas, and a unification of them, focusing on the concept of processing fluency.
The idea of processing fluency builds on the well- documented fact of repetition priming: previously seen stimuli are reliably perceived more easily and more quickly the second time around. This is often described by the assertion that the initial presentation produced fluency that then benefits subsequent processing. The exact nature of this fluency is not specified, but one can imagine pathways gaining efficiency through exercise, or neurons recruiting other neurons, or the like.
Implicit memory effects seem sometimes to come directly from this processing fluency. For example, in the word completion task above, subjects will be more likely to correctly fill in E_E_H_N_ if they have previously seen "ELEPHANT." In other cases, though, the effects of fluency come into play only via an "attribution step." Subjects seem able to detect when the processing has been fluent, and when not. This sense of ease is not detected by people as "ease," however. Instead, there is a general sense of specialness about the processing. In English, phrases like "that rings a bell" describe what this specialness actually feels like.
In any event, this specialness stands out to us, and we generally want to know why. Thus, the specialness triggers an `attribution' process, and, often, this specialness will be correctly attributed to the prior encounter. But, sometimes it isn't, which leads to misattribution of the specialness to some other quality of the stimulus, such as attractiveness. Interestingly, this is essentially what we may see as the mere-exposure effect!
What we see is that exposure to a particular stimulus will ease subsequent perception of the same stimulus. Moreover, exposure to a certain chain of thoughts will ease subsequent thinking along that chain of thoughts. The key is that this entire process works independently of active, conscious memory.
Evidence for this broad proposal comes from a number of ingenious experiments. Imagine a subject, for example, who has recently encountered the name Allen Smith. Although a subject may have explicitly forgotten seeing the name before, the pathway for the perception of "Allen Smith" will have been primed. Therefore, when subsequently asked if Allen Smith is a famous person, the subject will register some vague sense of distinctiveness about the name. The subject, however, may in many circumstances be unlikely to attribute this specialness to the specific prior encounter. If, for example, the prior exposure was one day earlier, then the subject may have explicitly forgotten this exposure. Therefore, she is likely to misattribute this specialness in a fashion guided by the experimental context, which is, in this case, a question about fame. As a result, a "false fame" effect is produced in which subjects mistakenly think that a name is famous, simply because of their prior experience with it. This example is what happened in one of the classic implicit-memory experiments (Jacoby, Kelly, Brown, & Jasechko, 1989).
The implications of this research for the mere-exposure effect should be clear. If subjects can misattribute the specialness to a belief that the name is famous, they might also be able to misattribute the specialness to other stimulus attributes in other circumstances. In particular, subjects could be asked to express affective evaluations of the stimuli. In this case, they again attribute their sense of specialness in a fashion guided by this experimental context, and show an increase in affective rating. This could produce the mere-exposure effect, with exposure producing fluency, and fluency attributed to attractiveness.
Much research has been conducted on the nature of implicit memory, and one important aspect of these memories is their specificity. If you see the word "elephant" in a list of words, you will not be primed for "banana." The idea is that this fluency comes from practice, and to be primed for "banana" one needs to practice with "banana." Indeed, this practice may be hyperspecific. Evidence indicates that, if you earlier saw "elephant," you will probably not even be primed for "ELEPHANT," (with modified type face) simply because it looks different (Graf & Ryan, 1990).
As well as this stimulus specificity, this priming is task specific. In other words, if you have seen "elephant" you will do better on a test that works by perception, such as word completion (Name a word that starts with "ele"). However, if you were primed with "Dumbo is one of these," you are much more likely to respond correctly on a task that requires thinking about elephants (What animal was used by the Mongols for warfare?) (see Blaxton, 1989). This, too, fits with the fluency idea on the argument that a specific chain of processing has been primed, making subsequent use of that processing path easier. Clearly what has been primed is the task of thinking about an elephant, not the task of perceiving the word "elephant."
The mere-exposure paradigm is conceptually very similar to an implicit-memory test, inasmuch as both ask for evaluations of a stimulus, as opposed to a direct memory query. Also, both, quite obviously, reflect effects of prior experience on current thoughts and actions, but neither, typically, is experienced by the subject as a "memory." Exposure to a stimulus creates an implicit memory for the stimulus. This memory, as stated above, is noticed by the perceiver as a sort of specialness about the stimulus. The key step, then, is that the mere-exposure effect could be the result of subjects' misattribution of this sense of specialness to liking for the stimulus.
There is one last curve we need to work around. Imagine that we designed an experiment in which subjects were asked how much they disliked the stimuli. With the idea of specialness and attribution, we might expect subjects to attribute the specialness to the distastefulness of the stimuli. This would give the opposite effect from the mere-exposure effect. However, this is not the result we see. Instead, even with the "dislike" question, we still see affect increasing with exposure (Bornstein, 1989). Clearly we need an extra assumption: this specialness or fluency is not merely distinctive and in need of attribution. Instead, the fluency somehow feels good to the subject. Thus, when asked to attribute the specialness to affect, the subject will do so, but only in the positive direction.
There are circumstances that make the subject more likely to form an explicit memory for the stimulus, such as continued exposures, or longer exposure durations. In these situations, if a subject can explicitly remember a stimulus, the source of the implicit memory is more likely to be correctly identified as the previously exposed stimulus. This will lead to a decrease in misattribution, and a lessening of the mere-exposure effect.
This leads to many predictions. As we have seen, for example, the mere-exposure experiment generally shows subjects' affective ratings initially climbing, with more and more exposures, but then reaching a plateau. This result might be explained by virtue of the fact that almost all of the basic mere-exposure effect experimental designs make complete recognition difficult, by nature of their design. For example, it is difficult to memorize novel Chinese ideographs after only six presentations of two seconds each. This will make explicit memory for these stimuli unlikely, and, as we have noted, these incomplete explicit memories can lead to misattribution of the implicit memory for the ideographs, potentially leading to an attribution to attractiveness. As the number of exposures increases, however, the strength of both implicit and explicit memories will increase. At first, when explicit memories may still be easily lost, this will produce an increase in affect ratings. However, after a while, explicit memories will become easier for the subject to maintain. This will lead to a decrease in misattribution, and the decrease in the mere- exposure effect.
Similar logic can be used to explain many of the results we have already reviewed. In the following sections, we return to these results, and examine how memory could be at work in them.
The experiments reported here are an attempt to look for similarity between the mere-exposure effect and the effect of implicit memory. These experiments are designed around the traditional mere-exposure design, with subjects first exposed to different stimuli a number of times, and then asked to rate the stimuli on affective grounds (i.e., for how much they like them). These experiments add an additional factor, however. In some conditions, the stimuli that subjects will see in the test phase will be slightly altered from how they looked in the exposure phase. As noted, these slight changes can have drastic effects on performance in implicit-memory experiments. Therefore, what we are looking for is a similar change in the mere-exposure effect due to these changes.
For example, Graf and Ryan (1990, Experiment 3) exposed subjects to words in either an outline font or a solid font. Then, they tested priming, using a word identification task, with words in either the same or the opposite font. Their data showed a statistically reliable main effect on their implicit memory test, in which subjects showed more priming for words when they were in the same font in test as they had been in the initial presentation.
The present experiment followed the Graf and Ryan (1990, Experiment 3) design, but with stimuli and a dependent measure adapted from the mere-exposure literature (Moreland & Zajonc, 1976). The stimuli were solid black Chinese calligraphy characters on a white background (taken from a poem in Yee, 1938 ). Some of the stimuli were presented to the subject during the exposure phase. Then, in the test phase, subjects rated all stimuli with regard to their aesthetic evaluation (i.e., how much they liked them). We know from Moreland and Zajonc that Chinese ideographs show a mere-exposure effect -- i.e., that earlier presented ideographs are preferred over novel ones. The question is whether this effect would be diminished by a change in font between exposure and test.
The test phase, therefore, compared four basic conditions. The first two were the conditions needed to assess the mere- exposure effect itself: previously seen stimuli vs. novel (but otherwise matched) stimuli. The second two conditions allowed us to probe the generalization of this effect: previously seen but modified (an outline of the character rather than a solid character) stimuli vs. novel modified stimuli.
The first two conditions were expected to give the standard mere-exposure effect. Previously viewed stimuli should be rated more highly than novel stimuli. Further, if mere- exposure is a form of perceptual fluency, the modification of the stimuli should destroy the exposure effect. Subjects should rate the modified form of a previously-presented stimulus at about the same level that they rate novel stimuli.
Design. The experiment was a 2 x 2 factorial design with all comparisons within-subjects. Stimuli were either previously exposed or novel, and presented as a solid black character on a white field, or as only a black outline. This resulted in four test conditions: PreExposed Normal (PN), PreExposed Outline (PO), Novel Normal (NN), and Novel Outline (NO). Assignment of stimuli to conditions was counterbalanced.
Stimuli. The stimuli were digitized Chinese characters (see the Appendix). Characters were scanned at 300 dpi from a Chinese calligraphy poem (Yee, 1938). They were chosen to be relatively complex, and recognizable as Chinese characters, yet not too distinctive (if the characters were too easily discriminated, then subjects would presumably remember them explicitly as well as implicitly). Each character was roughly 350 x 350 pixels. Nine stimuli per condition were used to follow the Moreland and Zajonc (1976) design.
Procedure. The experiment consisted of two phases: exposure and test. In the exposure phase, the stimuli were presented on the screen of a Macintosh II computer using the PsyScope experiment scripting program (Cohen, MacWhinney, Flatt, & Provost, 1993). Nine characters were displayed 3 times each for a 2 seconds duration, each followed by a three second pause. The durations used were derived from Moreland and Zajonc (1976). Stimuli were presented in a random sequence with the constraint that no successive trials showed the same stimulus. Subjects were told simply to look at the characters on the screen.
After the exposure phase was complete, subjects read the instructions for the test phase, and then the test phase immediately began. Subjects made nine aesthetic evaluations (one per stimulus) in each of four conditions (PN, PO, NN, and NO). In all conditions, each test trial consisted of a two second presentation of the stimulus, followed by the presentation of the question "Please rate how you liked the last picture on a scale from 1 - 7" on the screen, with the 1 - 7 scale displayed anchored with the terms "dislike" at 1 and "like" at 7. The order that conditions occurred was randomized, with the constraint that subjects could never get three trials in a row from a single condition. Subjects were not told about any relationship between the first and the second parts of the experiment.
After the test phase, subjects were debriefed, and the experimenter answered any questions they had.
Figure 1: Graph of means from Experiment 1.
Curious about this lack of interaction, we approached the problem with a new perspective. Because of the length of the experiment, and its simple and repetitive nature, it may be possible that viewing the stimuli in early trials in the test affects subjects' ratings in later trials. Perhaps subjects grow fatigued or impatient with the task, and therefore shift their criteria for responding as the trials proceed. Or perhaps stimuli early in the sequence somehow prime the later stimuli (some sort of "generalized" exposure effect). It is unclear what we should make of these suggestions, but, in the interests of caution, we re- analyzed the data, including presentation order in a 2 (exposure) x 2 (format) x 9 (order) ANOVA. The order factor simply tracked the serial position of the judgment -- and thus looked at the first judgment for each subject, the second, and so on. Given the random sequencing of test stimuli, different subjects viewed different stimuli in each of these positions. Therefore, this analysis averages across stimuli. A main effect for order was found to be statistically reliable (F(8,128) = 3.087, p = 0.003). The means for each serial position are shown in Figure 2. However, none of the interactions between order and the other factors were reliable.
Figure 2: Mean affect rating by serial position.
However, this claim does rest on a null result -- namely, the null interaction. As with all null results, we must examine this one with care. Most importantly, we need to note that there is ground for worrying about the statistical power of the result, given that there are not many subjects.
Three other factors also cloud the interpretation of this study. First, the exposure could have led to a strong explicit memory for the stimuli, which would help mask the implicit memory effects. Second, as we have already discussed, it is possible that through the course of the test phase, characters that appeared later in the test were primed by their counterparts that may have appeared earlier. In other words, there may have been priming from outline figures to normal and vice versa, making "novel" figures late in the test not novel any longer. Third, it is possible that the subjects got bored with the experiment. The duration of the exposure to the stimuli was relatively long, and the delay between each stimulus was also long.
Evidence supporting these possibilities comes from several sources. During debriefings, most subjects reported that they easily remembered in the test phase which items they had seen in the exposure phase. Several subjects reported feeling boredom with the procedure. In addition, the second analysis we conducted showed an effect for order of presentation. The order did not directly interact with either exposure or format, which means we must take this result with a grain of salt. However it does show an overall increase in affect ratings over the course of the experiment. This gives some credence to the second possibility mentioned above.
Fortunately, these three possibilities are easy to remove as concerns, by modifying the experiment's parameters. This leads directly to Experiment 2, which was designed to remove or reduce these concerns. Shorter durations were employed to diminish the likelihood of explicit memories. Fewer stimuli were used to decrease the possibility of early test stimuli priming later test stimuli. These two together should decrease potential subject boredom.
In addition, to enhance the priming ability of the exposure phase stimuli, each picture appeared more often. This change was made because the basic mere-exposure effect in Experiment 1 did not quite reach statistical reliability. More exposures should, hopefully, increase the priming power of the stimulus, thereby enhancing the mere-exposure effect.
Finally, to help reduce the possibility of explicit remembering, filler pictures were presented in the exposure phase that were not seen in the test phase. This modification should help diminish the chance of subjects explicitly remembering any specific stimulus, due to the increase in overall memory load, thereby enhancing the mere- exposure effect.
Design and Procedure. The design was the same as in Experiment 1, with the following changes. In the exposure phase, 8 stimuli were presented 6 times each. The stimuli were from the same set of Chinese ideographs scanned for the first experiment (see the Appendix). Each stimulus was presented for 250 ms, followed by a 750 ms pause. Four of these stimuli were experimental stimuli, to be tested later, and four were fillers. Order of the stimuli was randomized across subjects with the constraint that no stimulus appeared twice in a row. Subjects were told simply to look at the images.
As in Experiment 1, the test involved four types of stimuli: PreExposed Normal (PN), PreExposed Outline (PO), Novel Normal (NN), and Novel Outline (NO). Subjects made aesthetic evaluations of only four stimuli from each group. The picture flashed on the screen for 1500 ms, and was immediately followed by the question, with a repetition of the scale to be used in making the evaluations. Subjects were not told of any relationship between the two parts of the experiment.
Figure 3: Graph of means from Experiment 2.
Subjects' responses from this experiment were analyzed in a 2 (exposure) x 2 (format) ANOVA. The results are shown in Figure 3. Only the main effect for format was statistically reliable (F(1,16) = 10.231, p < 0.01), with subjects preferring the solid (normal) characters to the outlined ones. The main effect for exposure was not reliable (F(1,16) = 1.083) and there was no hint of an interaction (F(1,16) = 0.038). Interestingly, the data tend to tip away from a mere-exposure effect, since novel stimuli were slightly preferred.
This experiment used similar durations to the first experiment, while cutting down the actual number of stimuli, as in the second. However, the key to this experiment is the addition of a distracter task between the exposure and test phases. This distracter should help eliminate any possible explicit memories that may have formed for the previously exposed stimuli.
The distracter task should help diminish an explicit-memory contribution. This may happen for two reasons. First, the addition of a task between exposure and test adds a delay. Explicit memories seem to fade faster than implicit memories. Second, explicit memories can be maintained by rehearsal. An intervening distracter task guarantees an interruption in that rehearsal. Thus, a distracter task should increase the chances of subjects having an implicit memory for the stimuli without an explicit memory.
One additional concern with the previous experiments is that subjects were not told why they were seeing the images in the first part. In this situation, it is plausible that subjects assumed a memory test would be coming, and therefore worked to form explicit memories for the exposed stimuli. Therefore, we decided to make a change in this experiment that would give subjects an alternative reason to be looking at the images. Subjects were thus told that the study was examining the correlation between physiological responses and conscious aesthetic responses. To make this story more credible, a heart-rate monitor was attached to the subject. This setup presumably encouraged subjects to pay attention to the stimuli, without trying to memorize them.
Finally, this experiment added an exclusion instruction. This means that subjects were asked to report if they felt any familiarity with the stimulus. If they did, that stimulus was excluded. This final change was implemented to catch any explicit memories left over from the exposure phase. Similar exclusion procedures have been used before to separate implicit and explicit memories (Jacoby & Kelly, 1992).
The problem with this exclusion instruction occurs when a subject has both an implicit and explicit memory for a stimulus. In this situation, the subject's evaluation will be excluded, eliminating another data point. This means that the procedure will exclude a certain number of implicit memories, leading to an underestimate of the magnitude of the implicit memory.
Design and Procedure. The design of this experiment was, again, similar to that used in the previous studies. However, in an attempt to eliminate any traces of explicit memories, a distracter task was added between the exposure and the test. This distracter task was designed to lengthen the delay between exposure and test, and to keep subjects from performing any sort of elaborate rehearsal for the stimuli.
Prior to the experiment's start, a heart monitor was attached to their ear, with a wire connecting it to the computer. Subjects were then told that they were going to see some pictures and the computer would record their heart rate, which is correlated to how much they like the pictures.
During the exposure phase, 8 of the Chinese characters (see the Appendix) were presented on the screen 6 times each. Each stimulus was presented for 2 seconds, followed by a 3 second pause. Four of these stimuli were experimental stimuli, to be tested later, and four were fillers. Order of the stimuli was randomized across subjects.
After the "physiological" phase, subjects were told to return for the rest of the experiment after they participated in another student's experiment. Subjects were told that the delay was needed to avoid any potential physiological confounds. Actually, this delay was used to help keep subjects from forming explicit memories for the images. This delay involved one of several procedures: some subjects participated in another student's thesis experiment, an hour long perceptive experiment; some played Tetris for fifteen minutes; and some worked with their email for fifteen minutes.
As before, the test involved four types of stimuli: PreExposed Normal (PN), PreExposed Outline (PO), Novel Normal (NN), and Novel Outline (NO). Subjects were informed that they would be seeing more images, but this time, they were to respond on a scale from 1 - 7 on how well they liked each image. If the character seemed familiar to the subject, however, they were asked to respond with a zero instead. This was the exclusion instruction mentioned above.
Each image was flashed on the screen for 1500 ms, and was immediately followed by the question, the scale to be used for the evaluations, and a reminder to press 0 if the image seemed familiar.
Analysis was first done using only subjects who had zero responses on two or fewer stimuli out of the four stimuli in each condition. However, only 25% of the subjects met this condition. A 2 (exposure) x 2 (format) ANOVA was conducted on these data. None of the results from this analysis were statistically reliable.
As an attempt at exploring the pattern a bit more sensitively, an alternate analysis was conducted. If a subject reported subjective familiarity with a stimulus (by responding with a zero), that data point was replaced with the average response for that subject in the condition. This increased data set was then analyzed in a 2 x 2 ANOVA. The results are shown in Figure 4. The data show a reliable main effect for exposure, with subjects liking the novel stimuli better than the previously exposed stimuli (F(1,19) = 7.821, p = 0.0115). There was also a reliable main effect for format, where subjects tended to prefer the solid (normal) ideographs better than the outlined ones (F(1,19) = 5.623, p = 0.0285). The interaction was not significant (F = 0.008).
Figure 4: Graph of the means from Experiment 3.
Finally, to make sure that the time and task variations did not affect the results, these between subject variables were analyzed. Neither task nor time was found to make a difference (F(1,7) = 0.800 for task and F(1,12) = 0.019 for time).
A good reason to be cautious of this result is the number of trials that were excluded. Out of 80 total trials in the PN condition, only 24 were not excluded. Out of 20 total subjects, only 5 could be included in the analysis. These numbers severely limit the power of the statistics.
Amazingly, subjects quite often reported familiarity with the stimuli during the test phase, in spite of the intervening task. Only 25% of the subjects remembered 2 or fewer of the stimuli from each category, in spite of the attempts made in this experiment to limit subjects' explicit memories.
Figure 5: Size of the mere-exposure effect in each experiment.
To test the effectiveness of our general experimental paradigm, an analysis was conducted across experiments. The 3 (experiment) x 2 (exposure) x 2 (format) ANOVA found a few interesting results. First, there was a main effect for format, with solid figures uniformly preferred over outline figures (m = 4.6 for solids, and m = 4.1 for outlines, F(1,53) = 19.004, p = 0.0001). Second, an interaction between experiment and exposure was found. Each experiment found a successively lower relationship between the previously exposed and the novel ideographs (see figure 5).
This latter result is especially surprising. Each of the changes made in these experiments should have made the mere- exposure effect stronger, not weaker. One note that should be made is that the first two experiments were essentially the same, with only minor modifications to the exposure durations and sequence. The third experiment, on the other hand, introduced a new instruction, and significantly changed the subjects' understanding of the procedure. Therefore, this result is not necessarily telling us that our changes progressively lessened our ability to detect the mere-exposure effect. Instead, it simply tells us that Experiments 2 and 3 were entirely unsuccessful.
It is possible that the basic design of these experiments was fundamentally flawed, although the identity of that flaw is uncertain. However, we must take this idea with a grain of salt. Bornstein's (1989) literature review and meta-analysis clearly showed that the mere-exposure effect is not very sensitive to the details of experimental design. For the parameters that do matter, such as stimulus complexity, presentation sequence, and timing, the present experiments were matched to several others in the literature. Thus, we should be wary of blaming the experimental design.
A different, more attractive, possibility involves an often disregarded aspect of experimentation: the subjects. An interesting effect that has been found at Reed College is that students here seem to be much better at many common memory tests.[1] In classes, senior theses, and faculty research, Reed student subjects have shown themselves to be good at remembering things, even when an experiment is set up to attempt to make this difficult, as in our experiments above.
How can this improved memory account for the lack of experimental success? As discussed above, the existence of explicit memories weakens implicit memory tests. In Experiment 2, the changes made were clearly not successful at reducing the subjects' explicit memory for the stimuli. With the explicit memories around, subjects may have been less likely to misattribute the implicit memory to liking for the stimuli. This scenario would result in the null effect seen.
In Experiment 3, it is plausible that for many of the trials in which a subject had an explicit memory, the subject also had an implicit memory, possibly due to this extraordinary memory ability seen in studies using Reed students. However, because the explicit memory existed, the trial was removed, thanks to the exclusion instruction added. Therefore, the mere-exposure effect could have been wiped out by the experiment eliminating many of the trials for which a subject had an implicit memory. The "counter" mere-exposure effect that was seen may be due to implicit memories for the novel characters, possibly from exposure early in the test. These trials would not have been excluded because the subject had no explicit memory for the stimulus. However, we should be very wary of the effect in this experiment at all, for reasons noted above.
These results leave us wondering, then, about the connection between implicit memory and the mere-exposure effect. Clearly, our original experimental design was not successful in testing our hypothesis. However, the experimental design was also not successful at reproducing the mere-exposure effect itself. Obtaining this effect is essentially a precondition for interpreting the experiments. Because the experiments failed to reproduce this effect, we should not draw any firm conclusions from the studies themselves.
Clearly, this potential connection needs further study. Continued research on this connection should be wary of two main factors. First, the experimental design needs to be very carefully tuned. Pilot studies should be conducted to gauge the efficiency of the paradigm in getting the mere- exposure effect itself. In other words, previously exposed stimuli should be liked more with repeated exposure. Getting this effect is an important control condition of any research designed to test a theory of mere-exposure.
Secondly, researchers need to be aware of their subject pool, and design their experiment accordingly. In this case, researchers may have to stray a little further away from the standard memory testing situations in a place such as Reed, where students seem to have such a high ability to maintain explicit memories. Again, this comes down to clever and careful experimental design.