Content uploaded by Christian A. Meissner
Author content
All content in this area was uploaded by Christian A. Meissner on Dec 30, 2017
Content may be subject to copyright.
APPLIED COGNITIVE PSYCHOLOGY
Appl. Cognit. Psychol. 15: 603±616 (2001)
DOI: 10.1002/acp.728
A Meta-analysis of the Verbal Overshadowing
Effect in Face Identi®cation
CHRISTIAN A. MEISSNER
and JOHN C. BRIGHAM
Florida State University, USA
SUMMARY
Recent studies have demonstrated that requesting individuals to produce a verbal description of a
previously seen face can hinder subsequent attempts at identi®cation. This phenomenon, termed
`verbal overshadowing', has been studied rather extensively in the face-identi®cation paradigm;
however, studies have not always replicated the general effect. Based upon both practical and
theoretical interests in the phenomenon, a meta-analysis of 29 effect size comparisons (N2018)
was conducted. Across the sample of studies there was a small, yet signi®cant, negative effect
(Fisher's Z
r
0.12), indicating some degree of verbal impairment or overshadowing. A ®xed-
effects analysis of several moderating variables demonstrated a signi®cant effect of post-description
delay and type of description instruction. The pattern of means indicated that overshadowing effects
were more likely to occur when the identi®cation task immediately followed the description task,
and when participants were given an elaborative, as opposed to a standard (free recall), instruction
during the description task. Inconsistencies in the literature are discussed, as well as various
theoretical and applied issues regarding the verbal overshadowing effect. Copyright #2001 John
Wiley & Sons, Ltd.
Recent research on the susceptibility of eyewitness memory suggests that individuals may
be in¯uenced by a variety of factors involving encoding and retrieval of event-related
information (for reviews, see Ross et al., 1994; Sporer et al., 1996). While this line of
research has had some effect on the legal community's notion of eyewitness memory for
objects and events at the crime scene (Wells, 1993), until recently it was thought that the
procedure of obtaining a verbal description of the suspect had relatively minor con-
sequences for the memory of the witness. For centuries, law enforcement of®cials have
operated on this premise in their daily protocol of investigative procedures. Furthermore,
several studies have previously demonstrated the positive effects of verbal description,
rehearsal, and elaboration on later recognition of faces (Chance and Goldstein, 1976;
McKelvie, 1976; Read, 1979; Wogalter, 1991, 1996).
However, a series of six experiments by Schooler and Engstler-Schooler (1990)
demonstrated apparent deleterious effects of verbal description on subsequent identi®ca-
tion of a target face, an effect they termed verbal overshadowing. In these experiments,
participants were initially presented with a 30 s video presentation of a bank robbery,
including a salient perpetrator, after which they were assigned to either a description
condition or a no-description control condition. Those in the description condition were
instructed to use the next ®ve minutes to describe the facial features of the robber, while
Received 5 October 1999
Copyright #2001 John Wiley & Sons, Ltd. Accepted 25 April 2000
Correspondence to: Christian A. Meissner, Department of Psychology, Florida State University, Tallahassee, FL
32306-1270, USA
control participants were given an innocuous ®ller task. Identi®cation was assessed for
each condition using an 8-photo target-present lineup. Schooler and Engstler-Schooler's
results consistently demonstrated that when participants were asked to provide a post-
exposure description of the target, they were signi®cantly less able to identify him or her
when compared with a control group.
Since their initial set of experiments, a number of studies have further replicated the
overshadowing phenomenon within the facial memory paradigm (Dodson et al., 1997;
Fallshore and Schooler, 1995; Finger and Pezdek, 1999; Read and Schooler, 1994,
presentation at the Third Practical Aspects of Memory Conference, College Park, MD;
Ryan, 1992, unpublished thesis; Ryan and Schooler, 1998; Schooler et al., 1996).
However, several attempts at replicating the effect in other labs have proven unsuccessful
(Lindsay, 1990, personal communication cited in Schooler et al., 1996; Lovett et al., 1992,
unpublished manuscript; Meissner et al., 2001; Memon et al., 1999, unpublished manu-
script; Tunnicliff and Clark, 1999, unpublished raw data; Yu and Geiselman, 1993). In
addition, several shortcomings in the effect have been demonstrated. For example, it
has been shown that the effect may be attenuated on subsequent trials within-subject
(Fallshore and Schooler, 1995; Houser et al., 1997; Melcher and Schooler, 1995; Schooler
et al., 1996), or following re-presentation of the original stimulus prior to identi®cation
(Schooler et al., 1996), or when a signi®cant delay between the description and
identi®cation phases exists (Finger and Pezdek, 1999).
Regarding this latter post-description delay effect, Finger and Pezdek (1999) have
recently demonstrated that when participants were provided a 24-minute delay prior to
identi®cation of the target face, a `release of verbal overshadowing' occurred. In
particular, the delayed description condition outperformed not only the no-delay descrip-
tion condition, but also the no-delay control condition. This increase in accuracy was said
to result from a release of interference created by the description task. Although some
researchers have observed similar effects (cf. Yu and Geiselman, 1993), others have been
unable to replicate these results (Boelter and Reisberg, 1999, presentation at the Society
for Applied Research in Memory and Cognition Conference, Boulder, CO; Meissner et al.,
2001; Schooler and Engstler-Schooler, 1990). We will attempt to further examine this
potential moderating effect in the present meta-analysis.
As a result of various empirical inconsistencies in the literature, theoretical accounts of
the overshadowing phenomenon have often been dif®cult to formulate. In a recent review,
Schooler et al. (1997) concluded that the majority of research appeared to provide support
for both the modality mismatch hypothesis (i.e. the notion of competing representations in
memory: verbal versus non-verbal information) and the availability assumption (i.e. the
notion that the visual representation remains available in memory despite the temporary
verbal impairment). However, Schooler and colleagues noted little empirical support for a
recoding interference hypothesis (i.e. the belief that overshadowing effects are due to a
non-veridical verbal description of the target stimulus), as studies within the facial
memory domain have generally failed to ®nd a relationship between quality of description
and subsequent identi®cation accuracy (Fallshore and Schooler, 1995; Schooler and
Engstler-Schooler, 1990).
In contrast to this conclusion, though, several studies have found evidence of retrieval-
based factors in¯uencing whether or not the overshadowing effect occurs (Finger and
Pezdek, 1999; Meissner et al., 2001). For example, in a recent study, we attempted to test
the possibility that impairment on the identi®cation task depended upon the manner in
which individuals were initially instructed to describe the target from memory. This
604 C. A. Meissner and J. C. Brigham
Copyright #2001 John Wiley & Sons, Ltd. Appl. Cognit. Psychol. 15: 603±616 (2001)
hypothesis is rather consistent with the ®ndings of Finger and Pezdek (1999), who also
observed that variation in the type of description task (i.e. standard interview versus
cognitive interview) in¯uenced the likelihood of a verbal overshadowing effect. In testing
this retrieval-based effect, we found that inducing participants to generate many descrip-
tors by lowering their response criterion resulted in impairment or overshadowing on
a photo lineup task both immediately and 30 minutes following the description task.
However, conditions in which participants were encouraged to adopt a moderate or strong
response criterion failed to yield subsequent impairment. Based upon this evidence, we
concluded that failures to replicate the verbal overshadowing effect may have involved
differences in the response criterion (set perhaps via variations in instructions or degree of
exhortation used by experimenters) that participants used when attempting to describe the
target stimulus. This effect of instruction type will also be assessed in the present meta-
analysis.
Given the current degree of variability in results across studies, we felt that a statistical
synthesis of the literature would be useful and appropriate in order to generalize current
®ndings, clarify theoretical inconsistencies, and identify productive directions for future
research. An additional practical concern is that verbal overshadowing could have an
important in¯uence on the manner in which law-enforcement of®cials obtain information
from the eyewitness. If the impairment due to generating a description were found to be
substantial across studies, it would seem important to inform law-enforcement of®cials of
the potential harm in such a procedure.
The current meta-analysis took the approach advocated by Hedges and Olkin (1985) in
which a mean weighted effect size for the sample of studies was initially calculated,
followed by prediction of effect size based upon several moderating variables (see Johnson
et al., 1995, for a discussion of various approaches). We had three goals in synthesizing the
verbal overshadowing literature. First, we sought to estimate the true effect size for the
verbal overshadowing effect with regard to facial stimuli, including as many published and
unpublished studies as could be obtained. Second, we wanted to examine whether several
moderating variables inherent to the design of most verbal overshadowing studies might
predict when the effect would more likely occur. Finally, we examined several sets of
studies which failed to meet the methodological criteria for inclusion in the overshadow-
ing paradigm, but which involved post-encoding verbalization or recall of facial details.
Our interest here was to investigate whether the overshadowing effect might be found in
other paradigms (i.e. face-recognition paradigm), in other identi®cation tasks (i.e. mug
shot sorting task), and in other forensically relevant recall tasks (i.e. facial composite
reconstruction).
METHOD
Studies
For the primary analysis of verbal overshadowing on face identi®cation, a total of 29 effect
size comparisons described in 15 research articles were located, representing the responses
of 2018 participants. Studies were obtained by several methods, including: (a) searches of
the PsycINFO, Socio®le,Dissertation Abstracts, and First Search databases using the key
words `verbal overshadowing', `face identi®cation', `face description', `eyewitness
memory', and `facial memory'; (b) a search of the Social Sciences Citation Index using
the previous key words and the Schooler and Engstler-Schooler (1990) principal citation;
Verbal overshadowing synthesis 605
Copyright #2001 John Wiley & Sons, Ltd. Appl. Cognit. Psychol. 15: 603±616 (2001)
(c) cross-referencing of the bibliographies of review papers on the topic (Schooler et al.,
1996, 1997); (d) a search of several selected conference programmes (i.e. Psychonomic
Society, APA, APS, and SARMAC) over the past 5 years; and (e) contact with colleagues
in the ®eld who may have had knowledge of fugitive literature which had neither been
published nor presented at a conference. Of the 29 effect size comparisons located,
11 (38%) appeared in unpublished manuscripts or theses/dissertations.
Inclusion/exclusion criteria
To be included in the primary analysis of verbal overshadowing in face identi®cation,
studies must have involved a contrast between participants who were asked to both
describe and identify a target individual (description group), and those asked only to
identify the target individual (no-description control group). The statistical difference
between these two groups is known as the verbal overshadowing effect. Second, given the
potential in¯uence of confusion of multiple target faces on later verbalization, and the
evidence of attenuation of the effect when using multiple trials within-subject (see
Schooler et al., 1996), we included only studies that used a single stimulus face and
identi®cation task for each participant. Finally, participants must have been presented
with a target-present photo lineup identi®cation task, a predominant characteristic of
studies in the overshadowing paradigm.
1
Studies which were excluded due to one or more
criteria (N9) are included in a separate analysis. Several other studies (N8) which
involved the use of an intervening facial composite recall task are also subsequently
evaluated.
Coded variables
2
Four moderator variables, described below, were coded for each study. Moderators were
derived from aspects of the standard verbal overshadowing paradigm and included: (a) the
manner of stimulus presentation (i.e. live/video versus photograph); (b) the presence of
a post-encoding delay; (c) the presence of a post-description delay; and (d) the type of
description instruction given to participants (i.e. standard-free recall versus elaborative-
forced recall).
3
While both stimulus type and description instruction were represented as
dichotomous variables, post-encoding and post-description delay variables were log
1
Two experiments by Fallshore and Schooler (1995) utilized a paradigm in which participants were exposed to a
series of trials in which they were asked to view, describe, and identify a given target face. Because attenuation of
the effect occurred on later trials, only data from the ®rst trial were used in the present analysis. Other studies were
excluded due to various reasons. In particular, Chance and Goldstein (1976, Experiment 3) reported in Dodson
et al. (1997), McClure (1998, unpublished dissertation), and Woglater (1991, 1996) were excluded from the
overall analysis due to the use of multiple target faces at presentation. Mauldin and Laughery (1981) and
Thompson (1979, unpublished dissertation) were excluded due to the use of an alternative identi®cation
procedure (mug shot sorting task). Experiment 1 reported in Westerman and Larsen (1997) was also
excluded from the analysis due to the lack of a no-description control condition resulting from the within-
subjects design of the study.
2
Two independent raters coded each of the studies, after which percent agreement and Kappa () values were
calculated for each moderator variable. All Kappa's were in the excellent range (0.75 ), as suggested by Fleiss
(1981). A third rater was used to resolve any discrepancies in the codings.
3
A fourth moderator variable that might seem appropriate involved whether the target and test stimuli were
identical (`photo recognition'), or whether they varied in pose and context (`face recognition'). However, only
two early studies (Chance and Goldstein, 1976; Mauldin and Laughery, 1981) used identical stimuli at encoding
and recognition. Consequently, this moderator was not included in the analysis.
606 C. A. Meissner and J. C. Brigham
Copyright #2001 John Wiley & Sons, Ltd. Appl. Cognit. Psychol. 15: 603±616 (2001)
transformed and used in their continuous form, as performed when simulating decay/
forgetting in many cognitive models of memory (see Rubin and Wenzel, 1996).
First, the manner of stimulus presentation was recorded as either involving a short video
presentation of a mock witness event or the display of a single photograph of the target
face. A previous meta-analysis on facial identi®cation by Shapiro and Penrod (1986) also
examined the role of stimulus presentation. They hypothesized that performance would
be best when more retrieval cues were available at encoding; thus, a video presentation
should result in a higher rate of correct identi®cations when compared with a photograph of
the target. Their review, however, showed only mixed support for this hypothesis.
Within the verbal overshadowing paradigm, we might expect a similar process to be
working, such that variations in perspective and context might provide more cues for
verbalization, and thus increase identi®cation accuracy when compared with a stimulus
photograph.
Second, the presence of a post-encoding delay was coded. A majority of studies have
used a task to distract participants from rehearsing the target face, with the delay between
encoding and the description task varying between 5 and 20 minutes. However, a few
studies have immediately requested participants to provide a description of the target face.
If the mechanism of verbal overshadowing involves the veridicality of the description,
then one might predict that the presence of a delay prior to verbalization would decrease the
accuracy of the description due to a degradation of the original stimulus face in
memory. Consequently, the likelihood of impairment on later identi®cation could also
increase.
Third, the presence of a post-description delay was coded. While most studies have
assessed participants' memory immediately after the description, a few researchers have
examined the practical implications of inserting a delay between the description and
identi®cation tasks. Such post-description delays have ranged from as little as 3 minutes to
as long as 2 days. As noted previously, Finger and Pezdek (1999) demonstrated that
participants who described the target and then experienced a 24-minute delay prior to
identi®cation outperformed participants in the no-delay control condition. This effect was
assessed across the current sample of studies.
Finally, studies were coded for the type of instruction that participants were presented
during the description task. Given the paucity of information provided in most manuscripts
regarding the description instructions, authors were contacted and asked whether they
used any methods that may have `encouraged participants to verbalize for the entire
description period'. We were particularly interested in whether the authors explicitly
encouraged their participants to go beyond their normal criterion of free recall and to
provide more elaborative descriptions. We contrasted this with standard (free recall)
instructions in which participants were allowed to establish their own recall criterion.
Consistent with our previous ®ndings (Meissner et al., 2001), we hypothesized that studies
using an elaborative instruction would be more likely to demonstrate overshadowing
effects when compared with studies using a standard instruction.
Measure of effect size
Fisher's Z
r
was used as the measure of effect size across the studies. In most cases, a single
degree of freedom (df ) analysis (i.e.
2
) between the description and control conditions
was transformed into the effect size estimate. When studies presented only the proportion
of correct identi®cations (hit rates), a
2
statistic was computed using the number of
Verbal overshadowing synthesis 607
Copyright #2001 John Wiley & Sons, Ltd. Appl. Cognit. Psychol. 15: 603±616 (2001)
correct versus incorrect decisions across the control and description conditions.
2
values
were then converted to r(correlation coef®cient) by way of
rSQRT 2=N1
Finally, to control for skewness in estimating the true population parameter, rwas
transformed to Fisher's Z
r
by way of
Zr0:5loge1r=1r 2
Both the rand Z
r
transformation formulae were obtained from Rosenthal (1994). Because
verbalization is thought to impair subsequent identi®cation (Description±Control <0),
effect sizes demonstrating the overshadowing effect would be in the negative direction.
Conversely, a positive effect size would demonstrate some degree of verbal facilitation.
RESULTS
Effect size analysis
Effect sizes (Z
r
) computed for each study are presented in Table 1, in addition to sample
size, and coded variables. The mean weighted effect size across studies was Z
r
0.12, a
signi®cant effect size, Z5.27, p<0.001, with 95% con®dence intervals of (0.17,
0.08). No signi®cant outliers were noted in the sample of studies. Overall, the verbal
overshadowing effect accounted for 1.44% of the variability across studies, and indicated
that participants who described the target were 1.27 times more likely to misidentify the
target when compared with participants in the no-description control condition.
Moderator variables
A test of the homogeneity of variances across the sample of weighted effect sizes indicated
a signi®cant degree of variability, exceeding that expected on the basis of sampling error
alone,
2
(28) 59.42, p<0.01. Thus, the design moderators discussed earlier were used
to predict the variability across the sample of effect sizes. Due to the small sample of
studies, we utilized a ®xed-effects approach in which a weighted least squares regression
analysis was run on the sample of effect sizes with the four design moderators (stimulus,
post-encoding delay, post-description delay, and description instruction) as predictors (see
Hedges, 1994).
4
Studies were weighted in the analysis as a function of their sample sizes.
Results indicated that both the post-description delay, Z
j
2.00, p<0.05, r
s
0.21, and
description instruction variables, Z
j
2.83, p<0.01, r
s
0.40, signi®cantly in¯uenced the
magnitude of observed effect sizes. However, neither the manner of stimulus presentation,
Z
j
1.69, ns, nor the presence of a post-encoding delay, Z
j
1.19, ns, were signi®cant
predictors. The pattern of weighted means for the post-description delay variable indicated
that verbal overshadowing effects were seen in studies using either an `immediate'
4
A random effects model would have been most appropriate given other considerations; however, it is not as
sensitive an approach and would have provided only inconclusive results given the small number of studies (see
Raudenbush, 1994). In fact, results of a random effects analysis indicated only marginal effects for both post-
description delay, t1.73, p<0.10, and type of description instruction, t1.78, p<0.10.
608 C. A. Meissner and J. C. Brigham
Copyright #2001 John Wiley & Sons, Ltd. Appl. Cognit. Psychol. 15: 603±616 (2001)
Table 1. Sample size (N) effect size (Z
r
), and moderator variables across studies
Post-encoding Post-description Description
Study Exp./condition NFisher's Z
r
Stimulus delay delay instruction
Boelter and Reisberg (1999 unpublished raw data) 36 0.21 Video 48 h Immediate Elaborative
36 0.17 Video Immediate 48 h Elaborative
Dodson et al. (1997) Exp. #1 40 0.17 Video 20 min Immediate Standard
Exp. #2 80 0.13 Photo 5 min Immediate Standard
Fallshore and Schooler (1995) Exp. #1 120 0.34 Photo 5 min Immediate Elaborative
Exp. #3 160 0.05 Photo Immediate Immediate Elaborative
Finger and Pezdek (1999) Exp. #3 54 0.38 Photo 5 min Immediate Elaborative
Lovett et al. (1992, unpublished manuscript) Exp. #1 48 0.05 Video 30 min Immediate Elaborative
Exp. #2 48 0.14 Photo 30 min Immediate Elaborative
Exp. #1 59 0.13 Photo 30 min Immediate Elaborative
Meissner et al. (2001) Exp. #1 60 0.20 Photo 5 min 30 min Standard
60 0.00 Photo 5 min Immediate Standard
Exp. #2 60 0.34 Photo 5 min Immediate Elaborative
Memon et al. (1999, unpublished manuscript) Exp. #3 39 0.16 Video 20 min Immediate Elaborative
Miner and Reisberg (1999 unpublished raw data) Exp. #1 30 0.42 Video 15 min 3 min Elaborative
Exp. #2 24 0.09 Video 15 min 3 min Elaborative
Exp. #3 20 0.00 Video 15 min 3 min Elaborative
Ryan (1992, unpublished thesis) Incidental 229 0.08 Video 2 min Immediate Elaborative
Ryan and Schooler (1998) 165 0.11 Photo 2 min Immediate Elaborative
Schooler and Engslier-Schooler (1990) Exp. #1 88 0.27 Video 20 min Immediate Elaborative
Exp. #2 70 0.33 Video 20 min Immediate Elaborative
Exp. #4 78 0.22 Video Immediate 10 min Elaborative
Exp. #5 67 0.25 Video Immediate 48 h Elaborative
Exp. #6 56 0.31 Photo 5 min Immediate Elaborative
Schooler et al. (1996) Exp. #1 60 0.17 Photo Immediate Immediate Elaborative
Exp. #2 60 0.07 Photo Immediate Immediate Elaborative
Tunnichiff and Clark (1999, unpublished raw data) 62 0.15 Video Immediate 1 week Standard
Westerman and Larsen (1997) Exp. #2 59 0.11 Video 20 min Immediate Standard
Yu and Geiselman (1993) 50 0.33 Video Immediate 48 h Standard
Verbal overshadowing synthesis 609
Copyright #2001 John Wiley & Sons, Ltd. Appl. Cognit. Psychol. 15: 603±616 (2001)
assessment of identi®cation accuracy (Z
r
0.16), or a `short' (10 minutes) delay
between description and identi®cation phases (Z
r
0.13). However, in studies that
employed a `long' delay (30 minutes), differences between the description and control
conditions were marginal (Z
r
0.07), but in the direction of verbal facilitation. The pattern
of weighted means for the description instruction variable indicated that studies using an
elaborative instruction (Z
r
0.16) were more likely to elicit verbal overshadowing
effects when compared with studies using a standard instruction (Z
r
0.04). Finally,
goodness of ®t of the regression model was signi®cant,
2
(24) 42.43, p<0.01, indicat-
ing the presence of additional residual variance unexplained by the present moderators.
From the above results it is unclear what might be responsible for the differences in
performance across the post-description delay variable. One possibility is that a `release
of verbal overshadowing' may occur such that performance in the description condition
signi®cantly increases across the delay (Finger and Pezdek, 1999). Alternatively, differ-
ences in performance could result from memory decay across the delay for the control
condition. Finally, a combination of the above processes may be responsible for the
interaction (i.e. increased accuracy for the description condition and decreased accuracy
for the control condition across the delay). To examine more closely this effect, we
analysed the pattern of identi®cation accuracy for the control and description conditions
across the three post-description delay conditions (immediate versus short delay versus
long delay). Although variation in relative levels of identi®cation accuracy may occur as a
function of the particular stimuli used, we felt that such an analysis might nevertheless
provide some preliminary estimate of the in¯uence of a post-description delay. Unfortu-
nately, a mixed factorial ANOVA on the mean proportion of correct identi®cations
demonstrated a nonsigni®cant verbal overshadowing post-description delay interaction,
F(2, 26) 2.34, p0.12, MSE 0.013,
2
0.15. Nevertheless, planned comparisons
con®rmed that performance did signi®cantly differ in the control condition when the
identi®cation task followed either immediately or after a short delay (10 minutes) versus
a long delay (30 minutes), ts>2.55, ps<0.05. Conversely, participants who had
previously generated a description of the target stimulus showed no signi®cant perfor-
mance differences across the three delays, ts<0.30, ns. Figure 1 displays the pattern of
means for the interaction.
Publication bias
As there has been some dif®culty in replicating the verbal overshadowing effect within the
research community, it is possible that, despite our efforts to accumulate both published
and unpublished studies, some degree of publication bias exists in the sample. As a
measure of publication bias, we calculated the number of null studies necessary for our
effect to no longer be signi®cant. Also known as the `®le drawer calculation' (see Begg,
1994) it was estimated that 155 studies with an effect size of zero would be necessary for
the mean weighted effect size to no longer be statistically signi®cant. Thus, it appears that
we can maintain a good degree of con®dence in the results of the meta-analysis indicating
a verbal overshadowing effect.
5
5
In addition, a sensitivity analysis was conducted such that we could determine whether one or more studies
exerted undue in¯uence over the pattern of results (see Greenhouse and Iyengar, 1994). No study appeared to
contribute a large degree of in¯uence, as average effect sizes produced by the analysis fell within an acceptable
range.
610 C. A. Meissner and J. C. Brigham
Copyright #2001 John Wiley & Sons, Ltd. Appl. Cognit. Psychol. 15: 603±616 (2001)
Excluded studies
Several other studies were obtained from the literature search that were generally
consistent with the constraints of the verbal overshadowing paradigm, but were excluded
from the meta-analysis due to their use of multiple target faces with each participant
(Chance and Goldstein, 1976; McClure, 1998, unpublished dissertation; Wogalter, 1991,
1996), or their use of an alternative identi®cation procedure (Mauldin and Laughery, 1981;
Thompson, 1979, unpublished dissertation) (see Table 2).
6
The mean weighted effect size
across these studies was Z
r
0.15, indicating a signi®cant improvement in identi®cation
accuracy following generation of a verbal description (verbal facilitation), Z3.28,
Table 2. Sample size (N) and effect size (Z
r
) estimates for excluded studies
Study Exp./condition NFisher's Z
r
Chance and Goldstein (1976) 40 0.40
Mauldin and Laughery (1981) 4 s 26 0.33
15 s 22 0.50
McClure (1998, unpublished dissertation) Exp. #1 134 0.13
Exp. #2; 3-min delay 88 0.12
Exp. #2; immediate 88 0.02
Thompson (1979, unpublished dissertation) 50 0.00
Wogalter (1991) 43 0.12
Wogalter (1996) 48 0.18
Figure 1. Proportion correct identi®cation as a function of post-description delay
6
Experiment 3 of Dodson et al. (1997) and Experiment 1 of Westerman and Larsen (1997) were not included
in this set. Exclusion of the ®rst was due to the use of a mixed design in the description condition (i.e.
participants described only one of two faces but were later tested on identi®cation for both). The latter was
excluded due to the lack of a no-description control condition resulting from the within-subjects design of
the study.
Verbal overshadowing synthesis 611
Copyright #2001 John Wiley & Sons, Ltd. Appl. Cognit. Psychol. 15: 603±616 (2001)
p<0.001, with 95% con®dence intervals of (0.06, 0.23). In sum, the current set of studies
indicated that individuals who generated a description were 1.38 times more likely to
correctly identify the target(s) when compared with individuals in the no-description
control condition.
Facial composite recall task
Several studies (N8) have examined the in¯uence of a facial composite recall task (i.e.
Photo®t or Identi-Kit) on later identi®cation of a target face. It is possible that such a task
could act to overshadow or interfere with an individual's ability to make a subsequent
identi®cation. However, because the task maintains a visual component as well, one might
predict that the processes acting on visual recall and visual recognition would compliment,
rather than contradict, one another (see Schooler et al., 1997). To examine this effect,
studies that followed the basic overshadowing methodology, but which involved a facial
composite reconstruction rather than a verbal description task, were collected (see Table
3). Across studies, the mean weighted effect size was Z
r
0.21, indicating a signi®cant
improvement in recognition accuracy following facial composite reconstruction, Z2.92,
p<0.01, with 95% con®dence intervals of (0.07, 0.35). Overall, this effect accounted for
4.41% of the variability across studies, and indicated that participants who generated
a facial composite were 1.56 times more likely to correctly identify the target when
compared with a no-description control condition.
DISCUSSION
The results of our meta-analysis indicated a small, yet signi®cant, verbal overshadowing
effect. This overall effect accounted for only 1.44% of the variability across studies, a
®nding that may explain other researchers' failure in replicating the general overshadow-
ing effect. Interestingly, a sample of studies which were excluded from the primary
analysis, but which involved a comparison between description and control conditions,
exhibited a signi®cant verbal facilitation effect. Given that the overshadowing effect is
known to attenuate across repeated trials within-subject (see Schooler et al., 1996), this
effect was somewhat expected. However, less is known regarding the overshadowing
effect using alternative identi®cation procedures, such as a mug shot sorting task (Mauldin
Table 3. Sample size (N) and effect size (Z
r
) estimates for studies involving a facial composite
recall composite recall task
Study Exp./condition NFisher' Z
r
Davies et al. (1978) Exp. #2: <48 h 20 0.14
Exp. #2: 3 weeks 20 0.20
Mauldin and Laughery (1981) 4 s: immediate 20 0.79
4 s: 48 h 20 0.17
15 s: immediate 20 0.62
15 s: 48 h 20 0.37
Thompson (1979, unpublished dissertation) 50 0.26
Yu and Geiselman (1993) 47 0.03
612 C. A. Meissner and J. C. Brigham
Copyright #2001 John Wiley & Sons, Ltd. Appl. Cognit. Psychol. 15: 603±616 (2001)
and Laughery, 1981; Thompson, 1979, unpublished dissertation). Further research
assessing the reliability of these results would be bene®cial.
Due to the variance in effect size estimates across studies, a ®xed-effects analysis
(weighted least-squares regression) was run in which four moderators were entered as
predictors. Both post-description delay and type of description instruction were signi®-
cant. Regarding the post-description delay effect, examination of the mean weighted effect
sizes indicated that verbal overshadowing occurred when identi®cation followed the
description task either immediately or shortly thereafter (10 minutes). However, when a
long delay (30 minutes) was inserted prior to the identi®cation phase, no signi®cant
differences were noted between the control and description conditions.
A more in-depth analysis of the proportion correct across studies indicated that the long
delay (30 minutes) in¯uenced only participants in the no-description control condition
who demonstrated a signi®cant degree of forgetting. Conversely, participants who had
previously described the target face showed no change in performance across the delay.
Overall, it appears that the process of retrieving the stimulus face from memory during the
description task allowed participants to maintain or preserve the memory trace across the
extended post-description delay when compared with the control condition. Curiously,
such a ®nding is inconsistent with that of a recent study by Finger and Pezdek (1999) in
which the description condition demonstrated signi®cant improvement in identi®cation
performance across a 24-minute delay. Unfortunately, the authors included only an
immediate control condition; thus, we were unable to determine whether signi®cant
forgetting might also occur for participants not asked to describe the target face. The
general effect of post-description delay was also inconsistent with an experiment
conducted by Schooler and Engstler-Schooler (1990; Experiment 5 : 2-day delay), and a
study by Boelter and Reisberg (1999, presentation at the Society for Applied Research in
Memory and Cognition, Boulder, CO; 2-day delay), in which signi®cant verbal over-
shadowing effects were observed.
The ®xed-effects analysis also indicated a signi®cant in¯uence of description instruc-
tion on the magnitude of observed effects, a result consistent with our previous studies
(Meissner et al., 2001). Speci®cally, studies that employed instructions intended to elicit
an elaborative and detailed description of the target face were more likely to demonstrate
verbal overshadowing when compared with studies that employed only a standard (free
recall) instruction. First, we believe this effect to have certain implications for the methods
that 1aw-enforcement of®cials might use to elicit descriptions from eyewitnesses. Namely,
it appears that of®cers should allow witnesses to establish their own criterion of
responding when providing a description of the suspect, including the freedom to withhold
information of which they may be unsure. Although such a procedure involves a cost of
less information in the description, law enforcement must weigh this against the potential
for later misidenti®cation of the suspect.
Second, with regard to the theoretical mechanism of description instructions, our
empirical results (Meissner et al., 2001) have indicated that this effect is likely due to the
in¯uence of erroneous descriptors that are more often elicited under elaborative recall.
This erroneous coding of the stimulus appears to confuse participants when they later
attempt to distinguish that which was internally generated as correct versus incorrect, and
thus to match their memory for the target to members in the photo lineup. This retrieval-
based account is consistent with Schooler and Engstler-Schooler's (1990) original
recoding interference hypothesis, and is also in agreement with the results of Finger
and Pezdek (1999).
Verbal overshadowing synthesis 613
Copyright #2001 John Wiley & Sons, Ltd. Appl. Cognit. Psychol. 15: 603±616 (2001)
Several alternatives to this account also seem plausible. First, Schooler and colleagues
(1997) have recently proposed that the interference caused by verbalization may result
from the type of memory processes demanded by the paradigm (i.e. con®gural versus
featural processing of faces (see Diamond and Carey, 1986)), a theory they termed transfer
inappropriate retrieval (TIR). In general, the authors suggested that an individual's
description of a target face would contain only those aspects of the stimulus that were
readily verbalizable (i.e. featural information), a process that would reduce access to (or
overshadow) those aspects which were not recalled or those which could not be verbalized
(i.e. con®gural information). As a result of the suppression of crucial visual information,
individuals would later demonstrate dif®culty in matching their (description-based)
memory for the target with that of the identi®cation stimulus presented them.
With regard to the description instruction effect, TIR might predict that the criterion
shift produced by the elaborative instructions would simply lead to more (continuous)
verbalization such that overemphasis is placed upon the featural (verbalizable) informa-
tion. This overemphasis would then further overshadow crucial non-verbalizable (con®g-
ural) details that appear to be important for identi®cation. While Schooler and colleagues
(1997) discuss a number of other ®ndings in support of TIR, including evidence on cross-
racial effects (Fallshore and Schooler, 1995), re-presentation of the target face prior to the
identi®cation task (Schooler et al., 1996), and description of an alternative non-target face
(Dodson et al., 1997), one particular ®nding from the present meta-analysis is also
noteworthy. Namely, participants who engaged in a facial composite recall task (i.e.
Identi-Kit or Photo®t) demonstrated signi®cant improvement in identi®cation perfor-
mance over that of a no-recall control condition. As the processes activated by a facial
composite task are primarily visual, presumably preserving participants' ability to rely on
con®gural information, con®rmation of increased performance appears to provide yet
further evidence as to the viability of competing memory processes.
7
In addition to the TIR approach that Schooler and colleagues (1997) have advocated, a
second theoretical line has involved examining characteristics of the observer (participant)
him/herself. For example, a study by Ryan and Schooler (1998) found that individual
differences in participants' perceptual and verbal abilities appeared to in¯uence the
overshadowing effect in face identi®cation (although see Memon et al., 1999, unpublished
manuscript). In addition, several studies have indicated that increased perceptual skill in a
given domain (e.g. own-race faces, wine tasting, music, etc.) may also mediate over-
shadowing in tasks involving stimuli taken from the speci®ed domain (Fallshore and
Schooler, 1995; Houser et al., 1997, unpublished manuscript; Melcher and Schooler, 1996;
see also Schooler et al., in press). With regard to the description instruction effect,
individual differences may play a role in the recall criterion that participants naturally
select on such tasks. For example, certain individuals may demonstrate an ability to more
effectively monitor the responses that they generate from memory (see Koriat and
Goldsmith, 1996). Future research would be bene®cial in clarifying the potential for an
individual differences account of verbal overshadowing, and for generalizing to over-
shadowing effects beyond the facial identi®cation domain.
7
In only two effect size estimates was this ®nding not supported (Davies et al., 1978). Unlike other studies in
this sample, Davies and colleagues' participants were asked to return to the lab either less than 48 hours or 3
weeks after initially viewing the target stimulus. However, given such a small sample of studies, it is dif®cult
to conclude whether this moderating variable might solely be responsible for the observed decline in
performance.
614 C. A. Meissner and J. C. Brigham
Copyright #2001 John Wiley & Sons, Ltd. Appl. Cognit. Psychol. 15: 603±616 (2001)
Overall, it is apparent from the present analysis that verbal overshadowing is a genuine
and reliable phenomenon. However, various inconsistencies have plagued the experi-
mental literature, and conclusions as to the mechanisms responsible for the phenomenon
appear complex and elusive. Although we have made progress in assessing a handful of
these inconsistencies, it is our hope that future empirical and theoretical work will clarify
the various conditions under which the effect is reliably observed. Once this is accom-
plished, the applied issues associated with the verbal overshadowing effect may be further
addressed, and a more grounded account of the ®ndings can be presented to law-
enforcement of®cials for their use in designing everyday investigative procedures.
ACKNOWLEDGEMENTS
We are to indebted Anders Ericsson, Colleen Kelly, Kathy Pezdek, Jonathan Schooler, and
Rick Wagner for their comments on earlier versions of this manuscript.
REFERENCES
References denoted by an asterisk (*) were included in some portion of the meta-analysis.
Begg CB. 1994. Publication bias. In The Handbook of Research Synthesis, Cooper H, Hedges LV
(eds). Russell Sage Foundation: New York; 399±409.
*Chance J, Goldstein AG. 1976. Recognition of faces and verbal labels. Bulletin of the Psychonomic
Society 7: 384±386.
*Davies G, Ellis H, Shepherd J. 1978. Face identi®cation: the in¯uence of delay upon accuracy of
photo®t construction. Journal of Police Science and Administration 6: 35± 42.
Diamond R, Carey S. 1986. Why faces are and are not special: an effect of expertise. Journal of
Experimental Psychology:General 115: 107±117.
*Dodson CS, Johnson MK, Schooler JW. 1997. The verbal overshadowing effect: why descriptions
impair face recognition. Memory and Cognition 25: 129±139.
*Fallshore M, Schooler, JW. 1995. The verbal vulnerability of perceptual expertise. Journal of
Experimental Psychology:Learning, Memory, and Cognition 21: 1608±1623.
*Finger K, Pezdek K. 1999. The effect of verbal description on face identi®cation accuracy: `release
form verbal overshadowing'. Journal of Applied Psychology 84: 340±348.
Fleiss JL. 1981. Statistical Methods for Rates and Proportions (2nd edn). Wiley: New York.
Greenhouse JB, Iyengar S. 1994. Sensitivity analysis and diagnostics. In The Handbook of Research
Synthesis, Cooper H, Hedges LV (eds). Russel Sage Foundation: New York; 383±398.
Hedges LV. 1994. Fixed effects models. In The Handbook of Research Synthesis, Cooper H, Hedges
LV (eds). Russell Sage Foundation: New York; 285±300.
Hedges LV, Olkin I. 1985. Statistical Methods for Meta-analysis. Academic Press: Orlando, FL.
Johnson BT, Mullen B, Salas E. 1995. Comparison of three major meta-analytic approaches. Journal
of Applied Psychology 80: 94±106.
Koriat A, Goldsmith M. 1996. Monitoring and control processes in the strategic regulation of
memory accuracy. Psychological Review 103: 490± 517.
*Mauldin MA, Laughery KR. 1981. Composite production effects on subsequent facial recognition.
Journal of Applied Psychology 66: 351±357.
McKelvie SJ. 1976. The effects of verbal labeling on recognition memory for schematic faces.
Quarterly Journal of Experimental Psychology 28: 459±474.
*Meissner CA, Brigham JC, Kelly CM. 2001. The in¯uence of retrieval processes in verbal
overshadowing. Memory and Cognition 29: 176±186.
Melcher JM, Schooler JW. 1996. The misremembrance of wines past: verbal and perceptual
expertise differentially mediate verbal overshadowing of taste memory. Journal of Memory and
Language 35: 231±245.
Verbal overshadowing synthesis 615
Copyright #2001 John Wiley & Sons, Ltd. Appl. Cognit. Psychol. 15: 603±616 (2001)
Raudenbush SW. 1994. Random effects models. In The Handbook of Research Synthesis, Cooper H,
Hedges V (eds). Russell sage foundation: New York; 301±321.
Read JD. 1979. Rehearsal and recognition of human faces. American Journal of Psychology 92:
71±85.
Rosenthal R. 1994. Parametric measures of effect size. In The Handbook of Research Synthesis,
Cooper H, Hedges LV (eds). Russell Sage Foundation: New York; 383±398.
Ross DF, Read JD, Toglia MP. 1994. Adult Eyewitness Testimony:Current Trends and Develop-
ments. Cambridge University Press: New York.
Rubin DC, Wenzel AE. 1996. One hundred years of forgetting: a quantitative description of
retention. Psychological Review 103: 734±760.
*Ryan RS, Schooler JW. 1998. Whom do words hurt?: Individual differences in susceptibility to
verbal overshadowing. Applied Cognitive Psychology 12: 105±126.
*Schooler JW, Engstler-Schooler TY. 1990. Verbal overshadowing of visual memories: some things
are better left unsaid. Cognitive Psychology 22: 36±71.
Schooler JW, Fiore SM, Brandimonte MA. 1997. At a loss from words: verbal overshadowing of
perceptual memories. In Handbook of Learning and Motivation (Vol. 37), Medin D (ed.).
Academic Press: Orlando, FL.
Schooler JW, Ryan RS, Fallshore M, Melcher JM. in press. Knowing more than you can tell: The
relationship between language and expertise. In The Psychology of Expertise, Nisbett RE, Caverni
J (eds). Elsevier: Amsterdam.
*Schooler JW, Ryan RS, Reder LM. 1996. The costs and bene®ts of verbalization. In Basic and
Applied Memory: New ®ndings, Herrmann D, Johnson M, McEvoy C, Hertzog C, Hertel P. (eds).
Erlbaum: Hilisdale, NJ; 51±65.
Shapiro PN, Penrod S. 1986. Meta-analysis of facial identi®cation studies. Psychological Bulletin
100: 139±156.
Sporer SL, Malpass RS, Koehnken G. 1996. Psychological Issues in Eyewitness Identi®cation.
Erlbaum: Mahwah, NJ.
Wells GL. 1993. What do we know about eyewitness identi®cation?American Psychologist 48:
553±571.
*Westerman DL, Larsen JD. 1997. Verbal overshadowing effect: evidence for a general shift in
processing. American Journal of Psychology 110: 417±428.
*Wogalter MS. 1991. Effects of post-exposure description and imaging on subsequent face
recognition performance. Proceedings of the Human Factors Society 35: 575± 579.
*Wogalter MS. 1996. Describing faces from memory: accuracy and effects on subsequent recogni-
tion performance. Proceedings of the Human Factors and Ergonomics Society 40: 536± 540.
*Yu CJ, Geiselman RE. 1993. Effects of constructing identi-kit composites on photospread
identi®cation performance. Criminal Justice and Behavior 20: 280±292.
616 C. A. Meissner and J. C. Brigham
Copyright #2001 John Wiley & Sons, Ltd. Appl. Cognit. Psychol. 15: 603±616 (2001)