Dissociating mere exposure and repetition priming as a function of word type.

Abstract

The mere exposure effect is defined as enhanced attitude toward a stimulus that has been repeatedly exposed. Repetition priming is defined as facilitated processing of a previously exposed stimulus. We conducted a direct comparison between the two phenomena to test the assumption that the mere exposure effect represents an example of repetition priming. In two experiments, having studied a set of words or nonwords, participants were given a repetition priming task (perceptual identification) or one of two mere exposure (affective liking or preference judgment) tasks. Repetition priming was obtained for both words and nonwords, but only nonwords produced a mere exposure effect. This demonstrates a key boundary for observing the mere exposure effect, one not readily accommodated by a perceptual representation systems (Tulving & Schacter, 1990) account, which assumes that both phenomena should show some sensitivity to nonwords and words.

Full text

Page 1

Memory & Cognition
2004, 32 (5), 759-767
Zajonc (1968) posited that “the mere repeated exposure
of the individual to a stimulus is a sufficient condition for
the enhancement of his attitude toward it” (p. 1). Over the
last 35years, there have been many demonstrations of this
simple phenomenon, termed the mere exposure effect
(see Bornstein, 1989, for a review), using a broad range
of procedures and materials. For example, recent studies
have employed stimuli as diverse as advertisements
(Janiszewski & Meyvis, 2001), Turkish words (Marie
et al., 2001), photographs of people (Zárate, Sanders, &
Garza, 2000), Chinese ideographs (Monahan, Murphy, &
Zajonc, 2000), and polygons (Seamon, McKenna, &
Binder, 1998). Despite the interest that the phenomenon
has generated, however, a widely accepted theoretical in-
terpretation remains elusive. Of those that have been pro-
posed over the years (e.g., Berlyne, 1974; Harrison, 1977),
the primacy of affect (Zajonc, 1980, 2000) and the attribu-
tion of fluency (Bornstein & D’Agostino, 1994; Seamon,
Brody, & Kauff, 1983) theories have been most influential.
Increasingly, though, the mere exposure effect has
been described as an example of implicit memory (e.g.,
Bornstein & D’Agostino, 1992; Jacoby & Kelley, 1987;
Schacter, 1987; Squire, 1992). Implicit memory is de-
fined as the nonintentional, nonconscious retrieval of
previously acquired information and is demonstrated on
tasks that do not require conscious, intentional recollec-
tion of past experiences (Graf & Schacter, 1985). By de-
finition, the fact that the mere exposure effect can be ob-
tained without reference to the initial learning episode
means that it would qualify as an example of implicit
memory in its broadest sense (as might implicit learning,
for example). However, tacit in this is the assumption that
the mere exposure effect belongs to a particular implicit
memory subgroup—namely, repetition priming (as mea-
sured by tasks such as perceptual identification and word
fragment completion; see Butler & Berry, 2001b; Fleisch-
man & Gabrieli, 1998; Seamon et al., 1995). Repetition
priming occurs when prior exposure to a stimulus biases
or facilitates the processing of that stimulus on subsequent
exposures. Although this link between repetition priming
and mere exposure is intuitively appealing, not least be-
cause of the opportunity it accords to accommodate the
latter within an established theoretical framework (e.g.,
Roediger, Weldon, & Challis, 1989; Tulving & Schacter,
1990), there have been few attempts to explore it directly.
Seamon and colleagues (1995; Seamon et al., 1997)
have made some progress in this respect, reporting a
number of parallels between mere exposure and repeti-
tion priming. For example, Seamon et al. (1997) found
that study-to-test changes in retinal size and left–right
orientation produced the same pattern of results using a
mere exposure task, as had been previously reported with
an object decision task (Cooper, Schacter, Ballesteros, &
Moore, 1992). However, although compelling, not all ev-
idence shows such a convergence. In a recent review,
Butler and Berry (in press) highlighted what appear to be
a number of important differences between the two phe-
nomena. For example, whereas increasing the number of
times a stimulus is exposed at study appears to have lit-
759Copyright 2004 Psychonomic Society, Inc.
This research was supported by an Economic and Social Research
Council (ESRC) grant (R00233329) awarded to L.T.B. and D.C.B. We
thank Paula Brown for her assistance in data collection. Correspondence
should be addressed to L.T. Butler, School of Psychology, University of
Reading, Earley Gate, Whiteknights, Reading RG6 6AL, England (e-mail:
l.t.butler@reading.ac.uk).
Dissociating mere exposure and repetition
priming as a function of word type
LAURIE T. BUTLER and DIANNE C. BERRY
University of Reading, Reading, England
and
SHAUN HELMAN
University of Wales Institute, Cardiff, Wales
The mere exposure effectis defined as enhanced attitude toward a stimulus that has been repeatedly
exposed. Repetition priming is defined as facilitated processing of a previously exposed stimulus. We
conducted a direct comparison between the two phenomena to test the assumption that the mere ex-
posure effect represents an example of repetition priming. In two experiments, having studied a set of
words or nonwords, participants were given a repetition priming task (perceptual identification) or one
of two mere exposure (affective liking or preference judgment) tasks. Repetition priming was obtained
for both words and nonwords, but only nonwords produced a mere exposure effect. This demonstrates
a key boundary for observing the mere exposure effect, one not readily accommodated by a percep-
tual representation systems (Tulving & Schacter, 1990) account, which assumes that both phenomena
should show some sensitivity to nonwords and words.

Page 2

760BUTLER, BERRY, AND HELMAN
tle effect on performance on perceptual repetition prim-
ing tasks (e.g., Jacoby & Dallas, 1981), the same variable
often increases the size of the mere exposure effect (e.g.,
Bornstein & D’Agostino, 1992). One caveat with such
comparisons is that they have tended to reflect the out-
put of independent studies that have examined either one
or the other of the phenomena, but not both together. This
makes interpretation difficult, given that these studies
typically differ in other ways as well, such as in terms of
procedural details and materials employed. Therefore,
the aim of the present article is to provide a direct com-
parison of the two phenomena. In so doing, we report
new experimental data that provide further evidence that
the relation between mere exposure and repetition prim-
ing should be accorded closer scrutiny.
In accounting for their data, Seamon etal. (1997) applied
a multiple memory systems framework (e.g., Schacter,
1994; Tulving & Schacter, 1990) to the mere exposure
effect. Specifically, they suggested that affective prefer-
ence judgments are mediated by a structural description
system, a subsystem of the perceptual representation
system (PRS) described by Tulving and Schacter. The
structural description system, which encodes the struc-
tural relations among parts of an object, has previously
been implicated in nonverbal repetition priming tasks
such as the object decision task (Schacter, 1994; Schac-
ter, Cooper, & Delaney, 1990). By this view, then, both
mere exposure and repetition priming for objects reflect
the operation of a common representation system. Al-
though not discussed by Seamon et al. (1995), presum-
ably a similar relation might exist in the case of verbal
stimuli. As part of the architecture of the PRS, Tulving
and Schacter proposed that a visual word form system
subserves repetition priming for verbal materials. This
system is characterized as encoding information about
the visual and orthographic form of words and, in creat-
ing novel representations, is thought to support priming
for both nonwords and familiar words (see Schacter,
1994). Therefore, following Seamon et al.’s (1997) lead,
a visual word form system might also mediate mere ex-
posure effects for verbal materials. If this is correct, then
a seemingly straightforward prediction would be that
mere exposure and repetition priming should display
some sensitivity to both words and nonwords.
The existing literature on nonwords indicates that
such materials do produce reliable repetition priming
(e.g., Bowers, 1996; Dorfman, 1998) and mere exposure
effects (Dorfman, 1994; Marie etal., 2001). Furthermore,
the repetition priming literature contains numerous ex-
amples of priming effects for words (see Roediger &
McDermott, 1993, for a review). However, the picture with
regard to mere exposure and words is less clear. At first
glance, the evidence would seem to complement that for
repetition priming, with Bornstein (1989), in an influen-
tial and widely cited meta-analytical review, reporting
that meaningful words produce some of the largest ef-
fects of all. A closer inspection of individual studies,
though, reveals that nearly all the reported mere exposure
effects for real words actually reflect correlational data
between word frequency counts and liking ratings rather
than experimental manipulations of exposure per se (see,
e.g., Harrison, 1969; Sluckin, Colman, & Hargreaves,
1980). Indeed both of the experimental studies of which
we are aware (Amster & Glasman, 1966; Grush, 1976)
failed to obtain a significant mere exposure effect for
positive real words.
Therefore, in contrast to what might be predicted on
the basis of access to a visual word form system (Tulving
& Schacter, 1990), the available evidence suggests that
mere exposure effects are not observed for real words. This
is all the more striking given that much of the repetition
priming literature is based upon observations using such
stimuli. However, this apparent difference must be treated
cautiously given that the evidence is drawn from largely
independent literatures employing different methodolo-
gies. In this article, for the first time, we directly compare
the two phenomena using a set of real words and also a
carefully matched set of nonwords. If mere exposure and
repetition priming do reflect the operation of a common
underlying representation system (visual word form sys-
tem), then both phenomena should be obtained for words
and nonwords.
In Experiment1, we compare performance on common
measures of repetition priming and mere exposure—
namely, perceptual identification and affective liking. In
addition to being used extensively in their respective lit-
eratures, the tasks are well matched in that stimuli are
presented in the same format at test (i.e., complete sin-
gle items), albeit only briefly in the case of perceptual
identification. In Experiment 2, by way of extension, we
introduce a second commonly used measure of mere ex-
posure (preference judgment). Additionally, we include
a measure of explicit recognition to address an issue raised
in Experiment 1.
EXPERIMENT 1
Method
Participants. Ninety-four undergraduate students from the Uni-
versity of Reading participated in the experiment, for which they re-
ceived either course credit or a small cash payment. Forty-four par-
ticipants were randomly assigned to a repetition priming condition
(22 each for words and nonwords) and 50 participants to a mere ex-
posure condition (25 each for words and nonwords).
Design. The experiment employed a 2 (test type: affective liking
vs. perceptual identification) ? 2 (stimulus type: words vs. non-
words)?2 (exposure: studied vs. nonstudied) mixed factor design,
with repeated measures on the latter factor.
Materials. A key requirement for the stimuli was that they could
be used in conjunction not only with affective liking and perceptual
identification tasks, but also with a preference judgment task (see
Experiment 2). To meet this, 240 word pairs and 120 nonword pairs
were initially generated. Words were selected from a pool formed
by searching the MRC psycholinguistic database for two-syllable
words with a Thorndike–Lorge frequency count of less than or
equal to 25. The frequencies selected fell between the ranges de-
fined by Thorndike and Lorge (1944) as low (1–9) and medium
(19–49). Pronounceable nonwords were created by combining two
morphemes (prefix, suffix, or base morphemes) from a corpus de-
vised by Ljung (1974). Words and nonwords within a pair were
matched for number of letters, and had at least the first two letters

Page 3

MERE EXPOSURE AND REPETITION PRIMING761
in common (e.g., Dorfman, 1994). Each stimulus was then rated for
liking (stimuli presented individually) and for preference (stimuli
presented in pairs). Due to the large volume of stimuli, each partic-
ipant in the norming study (total n ? 180) rated a subset of the
available stimuli. Overall, this meant that each stimulus was rated
for liking and separately for preference (as part of a pair) by 20 par-
ticipants. Liking ratings were made on the following 6-point scale:
(1) really dislike, (2) dislike, (3) sort of dislike, (4) sort of like,
(5) like, (6) really like. For preference ratings, participants simply
circled the stimulus that they preferred within each pair.
From the resulting norming data, 32 word pairs and 32 nonword
pairs (see Appendix) were selected. For words and nonwords sepa-
rately, four lists each containing 16 items were created. These were
arranged so that Lists A1 and A2 and Lists B1 and B2 were paired.
The lists were closely matched; those stimuli within each pair (e.g.,
List A1 and List A2 items) were matched for liking and preference.
Furthermore, mean ratings were matched across lists (e.g., between
Lists A1 and B1). Finally, the lists were constructed so that both
words and nonwords were matched overall for mean liking (3.10
and 2.99, respectively).
At study, participants were presented with one of the eight pos-
sible lists (Lists A1, A2, B1, or B2 for words or nonwords). The
lists were rotated and counterbalanced across participants. At test,
for affective liking, perceptual identification and recognition (Ex-
periment2) tasks, participants received the studied list (e.g., ListA1)
in conjunction with the corresponding nonstudied list (e.g., ListB1).
For preference judgment only (Experiment 2), participants viewed
stimulus pairs comprising targets from the studied list (e.g., ListA1)
and distractors from the paired nonstudied list (e.g., List A2). Pairs
from the remaining lists (e.g., ListsB1 and B2) served as additional
distractors in this case. The position of targets and distractors (i.e.,
left and right) was randomized and fully counterbalanced for each
participant. Overall, arranging the stimuli in this way meant that, re-
gardless of test condition, all participants received 16 stimuli at
study and 32 at test (either single items or pairs). Presentation orders
were fully randomized at both study and test for each participant.
A filler task was also created for the purposes of the experiment.
The task was a short questionnaire that contained a series of 30 sim-
ple mathematical problems (e.g., 4 ? 16). The questionnaire was
designed to take no more than 2 min and, as is common in implicit
memory research, was administered between study and test phases.
Procedure. All participants were initially informed that they
would be taking part in a study to establish the ease with which peo-
ple read words (or nonwords). Having been given this cover task,
participants were instructed that a series of words (or nonwords)
would appear centrally on the computer screen that they should read
silently as accurately (but also as quickly) as possible. Participants
then saw 16 study items (either word or nonword Lists A1, A2, B1,
or B2), presented one at a time for 1,000 msec each. All items were
presented in bold Arial size 26 font and were preceded by 4 prac-
tice stimuli. Each trial was initiated by means of the space bar, and
there was a 1,500-msec delay between trials.
At the end of the study phase, participants were given 2 min to
complete the mathematics filler task. Subsequently, they were given
instructions for either mere exposure (affective liking) or repetition
priming (perceptual identification) tasks. Participants assigned to
the mere exposure condition were told that they would be provid-
ing normative data on word (nonword) likeability. They were fur-
ther informed that stimuli would appear one at a time on the com-
puter screen, and that they should indicate, on the 6-point scale
provided, how much they liked each stimulus. Participants initially
saw 16 practice items, followed by the 32 test words (nonwords).
Each item was displayed for 1,000msec, after which it was replaced
by a blank screen. Following a keypress (1–6), there was a 1,000msec
delay before the onset of the next item.
Participants assigned to the perceptual identification condition
were told they were providing data on word (nonword) identifica-
tion speed. Participants were further informed that they should sim-
ply try to identify stimuli that were flashed briefly upon the screen.
Participants then initially attempted to identify aloud 16 practice
items displayed in decreasing steps of 100, 80, 60, or 40msec (non-
words) or 80, 60, 40, or 20 msec (words), with 4 items presented at
each duration. Each item was immediately followed by a mask that
consisted of a row of ampersands (&&&&&&&&&). Responses
were recorded by means of a tape recorder and voice key, with par-
ticipants being told to say “pass” only if they had no idea as to the
identity of a particular stimulus. Performance on these practice
items was used to determine an exposure duration for test items for
each participant. This was taken as the point at which participants
identified roughly 50% of stimuli. As for affective liking, 32 test
words (nonwords) were then presented with a 1,000-msec delay be-
fore the next item following a response.
Results
Affective liking. The mean liking ratings for words
and nonwords are displayed in Figure 1. Comparison of
the ratings given to studied and nonstudied items indi-
cates that only nonwords produced a mere exposure ef-
fect. This was confirmed by a 2 (stimulus type: words
vs. nonwords) ? 2 (exposure: studied vs. nonstudied)
mixed analysis of variance (ANOVA). There was a main
effect of exposure [F(1,48) ?5.77, MSe?0.10, p ?.05]
but not of stimulus type [F(1,48) ? 1.06, MSe? 0.23,
p ? .05], but this finding was qualified by a significant
studied
nonstudied
nonwords
words
3.9
3.8
3.7
3.6
3.5
3.4
3.3
3.2
3.1
3
Stimulus type
Mean liking
Figure 1. Mere exposure (affective liking) for words and nonwords. Error
bars represent 1 SE above and below the mean.

Page 4

762BUTLER, BERRY, AND HELMAN
interaction [F(1,48) ? 4.85, MSe?0.10, p ? .05]. t tests
showed that the effect of exposure (studied vs. nonstudied)
was significant for nonwords [t(24) ? 3.16, p ? .005],
but not for words [t(24) ? 0.15, p ? .05].
Perceptual identification. Figure 2 displays the data
for the perceptual identification participants. Words were
associated with a somewhat higher baseline than non-
words, which reflects the variability involved in setting
individual thresholds for participants. However, the im-
portant comparison is between the mean proportion of
studied and nonstudied items correctly identified. As can
be seen, both words and nonwords produced equal, and
large, amounts of priming. An ANOVA confirmed this,
showing significant main effects of exposure [F(1,42) ?
28.80, MSe? 0.02, p ? .0001] and of stimulus type
[F(1,42) ? 17.43, MSe? 0.05, p ? .0001], which indi-
cated that overall a greater proportion of words than non-
words were identified. Crucially, though, the interaction
was not significant [F(1,42) ?1], indicating comparable
amounts of significant priming for both words [t(21) ?
4.86, p ? .0001] and nonwords [t(21) ? 3.22, p ? .001].
Discussion
Repetition priming was observed for both stimulus
types, whereas nonwords but not words produced a mere
exposure effect. Although consistent with earlier obser-
vations (e.g., Amster & Glasman, 1966; Grush, 1976),
the absence of a mere exposure effect for words is in di-
rect contrast to the numerous reports of repetition priming
effects for words (see Roediger & McDermott, 1993).
Furthermore, the finding seems difficult to reconcile
with the proposal of common access to a visual word
form system for the two phenomena.
Given the importance of the null finding, an analysis
was conducted to determine the power of the present ex-
periment to detect a mere exposure effect for words. On
the basis of the mere exposure effect observed for non-
words, a medium to large (see Cohen, 1988) effect size
of .69 was derived. From this, it was calculated that with
n ? 25 and ? ? .05 (one-tailed), the power to detect a
mere exposure effect for words, similar in magnitude to
that obtained for nonwords, was substantial (.78). As fur-
ther support for the reliability of the finding, it should be
noted that recent work carried out in our laboratory,
using a different set of materials, also failed to demon-
strate a mere exposure effect for words. In brief, in one
study, participants rated a series of studied and nonstud-
ied common nouns (e.g., goat, chair) for liking using a
9-point scale. The difference between studied (5.25) and
nonstudied (5.20) ratings was not significant [t(15) ?
0.33, p ? .74]. This pattern was replicated in a second
study, this time using a 6-point scale (3.56 vs. 3.51)
[t(16) ? 0.64, p ? .53].
Notwithstanding this consistent pattern, a second ex-
periment was conducted in the present study to extend
the finding to a different measure of mere exposure. Show-
ing that the null effect for words generalizes across mea-
sures would provide an even more compelling case be-
cause it would dispel any suggestion that the finding was
a function of the idiosyncrasies of the affective liking
task. A second aim of Experiment 2 was to clarify the re-
sults of Experiment 1 by addressing a possible explana-
tion for the absence of a mere exposure effect for words.
Specifically, we wondered whether the words that we
used were simply more recognizable than the nonwords.
According to one popular mere exposure account (Born-
stein & D’Agostino, 1992), participants may be less likely
to attribute the fluency that results from exposure to a
stimulus to liking if they realize that they have seen that
stimulus before. Therefore, if our words were more eas-
ily recognized than our nonwords, it would be unsur-
prising that they generated smaller mere exposure ef-
fects. For this reason, Experiment 2 incorporated an
explicit recognition condition. To provide an appropri-
ate comparison to the affective liking task used in Ex-
periment 1, a yes–no recognition task was employed.
studied
nonstudied
nonwords
words
.8
.7
.6
.5
.4
.3
.2
.1
0
Stimulus type
Proportion identified
Figure 2. Repetition priming (perceptual identification) for words and nonwords.
Error bars represent 1 SE above and below the mean.

Page 5

MERE EXPOSURE AND REPETITION PRIMING763
EXPERIMENT 2
Method
Participants. Seventy-two University of Reading undergradu-
ates took part in the experiment in return for course credit or a small
cash payment. Forty participants were randomly assigned to the
mere exposure condition (20 each for words and nonwords). On the
basis of power analyses performed for Experiment1, it was calcu-
lated that this number of participants would be sufficient to demon-
strate even a small to medium size effect. A further 32 participants
were assigned to the recognition condition (16 each for words and
nonwords). None had participated in Experiment 1.
Design. The experiment employed a 2 (test type: preference
judgment vs. recognition)?2 (stimulus type: words vs. nonwords)?
2 (exposure: studied vs. nonstudied) mixed factor design, with re-
peated measures on the latter factor.
Procedure. The study and filler task phases were the same as for
Experiment 1. At test, participants assigned to the mere exposure
(preference judgment) condition were told that they were providing
normative data on word (nonword) preferences. They were further
informed that pairs of stimuli would appear on screen, one pair at a
time, and that they should indicate by means of a keypress (“Z” for
left and “M” for right) which item in each pair they preferred. Par-
ticipants initially saw 16 pairs of practice items, followed by 32 test
pairs. Each pair was displayed for 1,000msec before being replaced
by a blank screen. Following a keypress, there was a 1,000 msec
delay before the onset of the next pair.
For the recognition condition, participants were told that they
would see a series of words (nonwords), some of which they had
seen before. They were further informed that they should use their
memories to help decide whether they recognized a particular word
(nonword) as having being studied. Following 16 filler items, which
participants treated as practice items, 32 test items were displayed
one at a time. Words (nonwords) were each displayed for 1,000msec
followed by a blank screen, which remained blank until participants
had made a recognition decision (“Z” for yes or “M” for no). There
was then a 1,000-msec delay before the onset of the next item.
Results
Preference judgment. The proportion of studied items
selected over nonstudied ones was much greater for non-
words (0.61, SD?0.14) than for words (0.52, SD?0.13).
A between-participants t test showed that the selection
rates for the two stimulus types were significantly dif-
ferent from each other [t(38) ? 2.25, p ? .05]. Further-
more, single-sample t tests revealed that preference for
studied stimuli was significantly above that which would
be expected by chance (i.e., .5) in the case of nonwords
[t(19) ? 3.67, p ? .01], but not words [t(19) ? 1].
Recognition. Recognition performance is displayed in
Figure3. As can be seen, the uncorrected performance was
slightly higher for nonwords than for words. However,
when we take into account false alarms, performance
(hits ?false alarms) was identical for words (0.57, SD ?
0.20) and nonwords (0.57, SD ? 0.21). This was con-
firmed by a between-participants t test [t(30) ? 1].
Discussion
In Experiment 2, using a preference judgment task,
mere exposure was again observed for nonwords but not
for words. Thus, we have provided a clear replication of
the pattern in Experiment 1. Additionally, we found that
both stimulus types were equally recognizable. Interest-
ingly, this pattern for recognition was very similar to one
recently reported by Mulligan (2002, Experiment 1) for
words and pronounceable nonwords that were initially
read. Therefore, our data suggest that the lack of a mere
exposure effect for words was not due to participants
being more likely to recognize them as being studied.
GENERAL DISCUSSION
Using two different measures, we have shown that
mere exposure effects are not obtained using real words,
under laboratory conditions at least, even though robust
effects are produced for nonwords under exactly the
same conditions. In contrast, repetition priming shows
no such differential sensitivity, with equal amounts of
priming being observed for both words and nonwords.
Although dissociations should be interpreted carefully,
we believe that this one is particularly noteworthy be-
hits
false alarms
nonwords
words
1
.9
.8
.7
.6
.5
.4
.3
.2
.1
0
Stimulus type
Proportion recognized
Figure 3. Yes–no recognition for words and nonwords. Error bars represent 1 SE
above and below the mean.