Access to this full-text is provided by Frontiers.
Content available from Frontiers in Psychology
This content is subject to copyright.
Frontiers in Psychology | www.frontiersin.org 1 August 2019 | Volume 10 | Article 1962
ORIGINAL RESEARCH
published: 28 August 2019
doi: 10.3389/fpsyg.2019.01962
Edited by:
Renate Volbert,
Psychologische Hochschule Berlin,
Germany
Reviewed by:
Anna Sagana,
Maastricht University, Netherlands
Josh Paul Davis,
University of Greenwich,
United Kingdom
*Correspondence:
Graham E. Pike
graham.pike@open.ac.uk
Specialty section:
This article was submitted to
Forensic and Legal Psychology,
a section of the journal
Frontiers in Psychology
Received: 21 May 2019
Accepted: 09 August 2019
Published: 28 August 2019
Citation:
Pike GE, Brace NA, Turner J, Ness H
and Vredeveldt A (2019) Advances in
Facial Composite Technology,
Utilizing Holistic Construction, Do Not
Lead to an Increase in Eyewitness
Misidentications Compared to Older
Feature-Based Systems.
Front. Psychol. 10:1962.
doi: 10.3389/fpsyg.2019.01962
Advances in Facial Composite
Technology, Utilizing Holistic
Construction, Do Not Lead to an
Increase in Eyewitness
Misidentications Compared to
Older Feature-Based Systems
GrahamE.Pike1
*, NicolaA.Brace1, JimTurner1, HayleyNess1 and AnneliesVredeveldt2
1 School of Psychology and Counselling, The Open University, Milton Keynes, United Kingdom, 2 Department of Criminal Law
and Criminology, Faculty of Law, VU University, Amsterdam, Netherlands
An eyewitness can contribute to a police investigation both by creating a composite image
of the face of the perpetrator and by attempting to identify them during an identication
procedure. This raises the potential issue that creating a composite of a perpetrator might
then interfere with the subsequent identication of that perpetrator. Previous research
exploring this issue has tended to use older feature-based composite systems, but the
introduction of new holistic composite systems is an important development as they were
designed to bea better match for human cognition and are likely to interact with memory
in a different way. This issue was explored in the current experiment. Participants were
randomly assigned to a feature-based composite construction condition (using E-FIT), a
holistic-based composite construction condition (using EFIT-V) or a control condition. An
ecologically valid delay between seeing a staged crime, creating the composite, and
completing the identication task was employed to better match conditions in real
investigations. The results showed that neither type of composite construction had an
effect on participants’ accuracy on a subsequent identication task. This suggests that
facial composite systems, including holistic systems, may not negatively impact subsequent
eyewitness identication evidence.
Keywords: facial composite, eyewitness identication, eyewitness memory, post-event information, unconscious
transference
INTRODUCTION
e role that an eyewitness plays in a police investigation is likely to dier depending on
whether or not the police are able to identify a suspect at a relatively early stage. If there is
a suspect, then the eyewitness will usually beasked to participate in some form of identication
procedure such as a lineup (commonly used in the US) or a video identication parade
(commonly used in the UK). If the investigation has not yet generated a suspect, then the
Pike et al. Facial Composites and Eyewitness Identication
Frontiers in Psychology | www.frontiersin.org 2 August 2019 | Volume 10 | Article 1962
eyewitness might beasked to provide further assistance, including
looking through albums of mug-shots (Dysart et al., 2001) or
by creating a facial composite image (Brace et al., 2006).
e rst generation of facial composite systems consisted
of individual facial features that were printed on acetate and
could be combined to form a face. Two such systems were
Identikit, which used drawings of facial features, and Photot,
which used actual photographs. Toward the end of the 1980s,
due to the increasing low cost and portability of personal
computers, computerized systems such as E-FIT were introduced,
which stored large image databases of facial features that could
be combined using an interface that also allowed basic image
manipulation. E-FIT construction begins by the operator
interviewing the witness to gain a description of the perpetrator
and then entering this into the system by selecting from lists
of feature descriptors. An initial image, which is greyscale and
fairly low resolution by contemporary standards, comprising
the best match for the description provided is then displayed
and shown to the witness, who attempts to improve it. is
is done by the system displaying dierent features, always seen
within the face, until the witness is happy that the particular
feature is the best match possible. e system can also move
and resize features; change skin, hair and eye tones; and
add paraphernalia.
Research into the accuracy of the facial images produced
by these feature-based systems found the images produced to
bea generally poor likeness of the perpetrator (Laughery etal.,
1977; Ellis et al., 1978; Laughery and Fowler, 1980). However,
Brace et al. (2006) noted that this poor quality was not
necessarily due to any limitations of the systems, as a skilled
operator was able to produce a good likeness of a face he/
she was familiar with, but rather had to do with the memory
(particularly recall) and communication skills of the witness
being insucient for the task and not compatible with a
feature-based approach. Making a comparable point, Wells etal.
(2005) suggest that the task of composite building is an
“unnatural act,” as it requires a witness to recognize and
assemble a face from its component parts, a task that human
cognition struggles with.
e move to a computer-based approach, such as E-FIT,
meant that the witness always saw the individual features as
being part of a whole face, rather than having to search through
albums of individual facial features. However, construction still
proceeded by selecting and manipulating individual facial
features and so essentially used the same piecemeal approach
as the earlier systems, an approach that had been shown to
be problematic for witnesses (Pike etal., 2005). In an attempt
to overcome this fundamental mismatch with human cognition,
a new generation of composite construction systems was
developed that not only involved showing a whole face to the
witness, but utilized a holistic construction process as well
(Tredoux etal., 2006; Frowd etal., 2007; Solomon etal., 2012).
One such “holistic” composite system is EFIT-V (branded
as EFIT6 at the time of writing), which was the system used
in the experiment reported in this paper. EFIT-V is based
on a principal components analysis (PCA) approach (for a
detailed technical description of how the EFIT-V system
operates, see Gibson et al., 2003; Solomon et al., 2012). e
basis of the system is that a database of faces is analyzed
using PCA to determine a set of eigenfaces that describe the
variation across the database. ese eigenfaces can then
be combined using dierent weightings in order to create
any face within the “face space” described by the original
faces in the database. If the database is suitably representative
of the entire population of faces, then the system can be used
to create any face. Using PCA means that the faces are stored
in a holistic manner, i.e., a description of the entire face is
used, rather than decomposing it into individual features,
and that the construction process does not require the witness
to work with individual features. EFIT-V uses an interface
in which a grid of nine faces are presented to the witness,
who selects the face they think is most similar to that of
the perpetrator (although more complex selection tools are
available, should the witness wish to do more than select
the best match). e system then produces another grid of
nine faces that resemble the face selected by the witness. e
variation between the faces in the grid is gradually reduced:
the rst grid contains considerable variation, but by the 10th
grid, the faces look far more similar to each other. In this
way, the system uses information provided by the witness,
through his/her selection of which face is the best match,
to generate faces that should besuccessively more and more
like that of the perpetrator.
is form of representation, presentation, and manipulation
is entirely holistic, which is a good match for the way in
which humans process faces (Laurence and Hole, 2012). Further,
it is more based on recognition than recall, as the witness is
not required to verbally describe the face at the start of the
construction process, as was the case with older feature-based
systems such as E-FIT. Indeed, systems using a holistic approach,
such as E-FITV and EvoFIT (another holistic facial composite
system utilizing a similar method to EFIT-V, but presenting
grids of 18 greyscale facial images), were designed explicitly
to be a good match for human cognition and to avoid a
situation in which a witness is required to process a facial
image in a way that would be “unnatural” (Gibson et al.,
2003). Research has found that the holistic, PCA approach to
composite construction does appear to be more compatible
with human cognition (Brace et al., 2008) and, critically, to
produce facial images that are a better likeness to the perpetrator
(Frowd et al., 2010, 2012).
It is important to remember that composite images are
used to help generate a possible suspect as part of a police
investigation rather than a method of directly identifying the
perpetrator. If the only evidence available is from an eyewitness,
then generating suspects using the facial composites (and/or
verbal descriptions) they provide may be the only way of
progressing the investigation. If a suspect is generated, the
witness may subsequently be asked to attempt to identify the
perpetrator from a lineup or video parade in which the suspect
appears. is means that a witness might produce a composite
and be shown a lineup. It is therefore important to assess
composite systems not just in terms of the accuracy of the
facial image produced, but also in terms of whether they might
Pike et al. Facial Composites and Eyewitness Identication
Frontiers in Psychology | www.frontiersin.org 3 August 2019 | Volume 10 | Article 1962
interfere with the witness’ memory of the face to the extent
that later identication from a lineup is aected.
Previous research has been conducted to explore what eect
creating a composite, mostly using feature-based systems (e.g.,
Photot or E-FIT), might have on subsequent eyewitness
identication performance, and the results have not been
consistent. Some studies have found that composite production
interferes negatively with later identication accuracy (Well s
et al., 2005; Topp-Manriquez et al., 2016), some that it does
not have any signicant eect (Yu and Geiselman, 1993; Davis
et al., 2016; Pike et al., 2019), and some that it actually has
a positive eect (Meissner and Brigham, 2001; Davis et al.,
2014). Tredoux etal. (2016) conducted a meta-analysis of the
research exploring the eect composite production might have
on subsequent performance at an eyewitness identication
procedure and concluded that creating a composite does not
appear to have a statistically signicant eect. However, many
(52 out of 72) of the eect sizes included in the Tredoux
etal. meta-analysis arose from experiments using older, feature-
based composite systems or sketch artists.
Given the variability of results, it is worth considering in
more detail the ways in which facial composites might interfere
with memory. In explaining the impairment reported in target-
present identication tasks, Wells et al. (2005) state that they
“…strongly suspect that the impairment [to performance on
the identication task] results from the process of having to
break the face down into individual features to perform the
composite building task” (p.151). ere is, of course, considerable
support for the notion that faces are processed and remembered
holistically, and not decomposed into their individual features
(e.g., Baddeley, 1979; Tanaka and Farah, 1993; Wilford and
Wells, 2010). is suggests that if witnesses were allowed to
construct the composite in a more holistic manner, the
impairment would have been reduced. In contrast, however,
the notion that featural construction impairs memory for the
target face is contradicted by ndings that composites constructed
in a featural manner can also improve subsequent face recognition
(E-FIT; Davis et al., 2014; Identi-Kit; Mauldin and Laughery,
1981; free-hand drawing; McClure and Shaw, 2002).
An alternative explanation as to why composites might
negatively impact later eyewitness identication accuracy relates
to an observation by Mauldin and Laughery (1981), who suggest
that more accurate or realistic representations of features “…
may be more similar to the memory representation of the
target face…” and therefore “…[a] signicant amount of
interference may result” (p.356). ree studies have examined
the role of similarity in more detail by varying the lineup
images’ degree of similarity to the target face and the created
composite (Comish, 1987; Kempen and Tredoux, 2012; Topp-
Manriquez etal., 2016). In these studies, the target was always
a composite face rather than a natural face to permit realistic
morphing. All three studies found that participants’ memory
for the composite they had created interfered with their memory
for the target, although it is worth noting that the composites
and targets were all composite faces, rather than real faces,
so may have had a high degree of similarity to each other.
However, a composite can also be similar to the target when
the target is a real face and the composite itself also looks
more like a real face than does a typical composite, as is the
case when a modern, holistic system is used (e.g., Tredoux
et al., 2006; Frowd et al., 2007; Gawrylowicz et al., 2012).
us, on the one hand, wemight expect less impairment when
witnesses can construct the composite in a holistic manner,
because it is a more natural process (Wells et al., 2005); on
the other hand, we might expect more interference when the
composite is created in a holistic manner, because it looks
more realistic and is therefore more likely to interfere with
the memory of the original face (Mauldin and Laughery, 1981).
As part of the present paper, we provide a direct test of these
competing hypotheses.
One study that has sought to answer this question was
conducted by Davis et al. (2014), whose rst experiment
compared the inuence of composite construction using E-FIT
or EFIT-V on subsequent identication performance. ey
found that composite construction using either system resulted
in improved identication performance compared to a
no-construction control condition, but there was no signicant
dierence between E-FIT and EFIT-V. In their second experiment,
they again found a benecial eect of EFIT-V composite
construction (E-FIT was not tested), but in a follow-up study
(Davis et al., 2016), there was no signicant eect of EFIT-V
construction. Although the work by Davis etal. was an important
rst step in assessing the inuence of holistic composite
construction, only one of their studies directly compared EFIT-V
to E-FIT (i.e., a PCA-based to a feature-based composite
process). Further, participants in all of their studies experienced
only short delays between encoding and composite construction
(0–30 min) and between composite construction and
identication (5 min to 32 h). e current experiment was
designed to compare the eects of composite building using
a PCA-based system (EFIT-V) or a feature-based system (E-FIT)
to a no-construction control condition, using delays more
typical of real criminal investigations (Frowd et al., 2005),
namely 2 days (mean 52 h) between encoding and composite
construction and an average of 20 days between composite
construction and identication.
MATERIALS AND METHODS
Participants and Design
A total of 245 participants completed all three phases of the
experiment. e data for a further 10 participants were excluded
from analyses either because they did not attend all three
stages or, in the case of two participants, because checks
conducted at the end of the experiment indicated a chance
that they may have previously seen the target. e participants
(70% females, mean age 37.94, SD 12.76) were all working at
or visiting the main campus of the Open University in a variety
of capacities.
Participants were randomly assigned to one of six conditions
of a 3 (Condition: E-FIT, EFIT-V, control) × 2 (Lineup: target-
present, target-absent) factorial design. Participants were treated
according to the ethical guidelines of e British Psychological
Pike et al. Facial Composites and Eyewitness Identication
Frontiers in Psychology | www.frontiersin.org 4 August 2019 | Volume 10 | Article 1962
Society and ethical clearance to conduct the experiment was
granted by the human research ethics panel of e Open University.
is sample size was enough to detect a medium eect (of
φ = 0.31 for the target-present condition and φ = 0.32 for the
target-absent condition) with power = 0.80 at α = 0.05. is
meant that the sample was sucient to detect an eect considerably
smaller than φ = 0.74 that was reported by Wells et al. (2005).
Materials
Video sequences of staged crime scenarios were created for
four targets, all of whom were Caucasian, aged approximately
30, and appeared without glasses or other paraphernalia. Two
of the targets were male, with short, dark brown hair, and
two female, with medium length, light brown hair. Each target
was shown walking down a corridor while attempting to open
the doors to a number of oces. When one door was found
to be open, the target was seen entering the oce, searching
around, nding a mobile phone in an unattended jacket, placing
it in his/her pocket and leaving the room. is sequence was
edited to ensure that each target was viewed close-up for a
minimum of 10 s as well as from a distance and from all
angles. e edited sequences were 1 min 31 s, 1 min 50 s,
1 min 38 s, and 1 min 54 s in length for targets male 1,
male 2, female 1, and female 2, respectively.
Composites were created by an operator using E-FIT (a
feature-based composite system) or EFIT-V (a holistic, PCA-based
composite system). e operator was a researcher trained in
both composite systems. In the E-FIT condition, the participant’s
verbal description of the target was used to generate an initial
likeness. e participant then guided the operator in searching
through and selecting alternative features and in moving and
resizing these features. is process followed that used by police
operators and witnesses in criminal investigations. In the EFIT-V
condition, the participant’s verbal description was used only
to select the appropriate database for sex and ethnicity. e
EFIT-V system then displayed its rst “generation” of nine
randomly created face images from the eigenfaces within that
sex and ethnicity set. e participant selected the “best match”
from the face images displayed and the system then produced
the next “generation” based on their selection. is continued
until the participants were satised that they had produced
as good a likeness as they could.
Although facial composite systems are used by law enforcement
around the world, many of the systems were developed in the
UK, where the standard is to use video identication procedures,
not lineups. We decided to employ lineups in the current
experiment so as to use a method more comparable to previous
research in this area, most notably Wells etal. (2005). A target-
absent and a target-present photo lineup were created for each
target. e target-absent lineups consisted of nine foil images,
while the target-present lineups consisted of eight foil images
and an image of the target. e target’s image was a photograph
taken separately (i.e., not a still from the crime video) and the
lighting, background, clothing, and hair cues were dierent to
those seen in the video. e foils for all lineups were selected
by matching potential images to verbal descriptions of the targets
provided by three participants who were unfamiliar with them.
As part of the procedure for calibrating the lineups, three
independent judges then checked each image to make sure that
it was an approximate visual match for the target, for example
to avoid foils containing any particularly distinctive features
not mentioned in the verbal description. e foil images were
sourced from the Pics image database maintained by Stirling
University1. All of the images for the lineups were standardized
so as to be pictorially similar (e.g., matched for size and
resolution). e eective size of the lineup was measured using
the technique advocated by Tredoux (1998). Sixty mock witnesses
(who had not seen the targets) made a forced-choice selection
from each lineup based on the modal description (derived from
the descriptions provided by the three unfamiliar people). Analysis
showed that all four lineups included at least seven plausible
choices, with eective sizes for each lineup of 7.50, 8.57, 7.93,
and 7.03. Analysis of lineup bias (see Malpass et al., 2007)
revealed no statistically signicant bias against the four targets,
with exact probabilities of 0.09, 0.13, 0.16, and 0.13. Although
non-signicant, the probability of bias for target 1 (0.093) was
a little high and less than 0.1, but as well as being non-signicant,
two foils were selected more frequently by the mock witnesses,
so the lineup was deemed to beunbiased (Malpass etal., 2007).
Procedure
Participants completed the experiment individually. ere were
three phases to the experiment. In the rst phase, participants
were informed that they were about to act as witnesses and
would be shown a short video of a staged “incident.” ey
were told that they would see the video once only and that
aer seeing the video they would beasked to return approximately
2 days later to “give evidence.”
Approximately 2 days later (mean 51.6 h, range 42.42–73.0),
participants returned to the lab for the second phase. ey were
rst asked how well they could remember the perpetrator on
a scale ranging from 0 (no memory of the target) to 100 (perfect
memory of the target). ey were then asked to provide a
verbal description of the person depicted in the video, writing
down as much as they could from the video sequence including
clothing, body, and facial details. Aer providing their description,
the 81 participants in the control condition le while the remaining
participants constructed a facial composite of the target (84
participants using E-FIT and 80 participants using EFIT-V). On
average, the time taken to construct the E-FIT composites (mean
30.33 min, SD 8.47) was considerably longer than the time
taken for the EFIT-V images (mean 17.62 min, SD 7.05). is
was partly because E-FIT construction requires an additional,
initial stage in which the verbal description is entered and partly
because the grid-based selection used in EFIT-V was designed
to bea more ecient interface, avoiding having to look through
large numbers of individual features as is the case with E-FIT.
Aer constructing the composite, participants were asked to
rate (out of 10) how similar they thought their composite was
to the target (based on their memory of the target’s face).
1
http://pics.stir.ac.uk
Pike et al. Facial Composites and Eyewitness Identication
Frontiers in Psychology | www.frontiersin.org 5 August 2019 | Volume 10 | Article 1962
Between 11 and 35 days aer the interview (M = 20.26,
SD = 4.85), participants returned to complete the third phase
of the experiment. e large variation in this delay resulted from
the practicalities in nding a time when the participant was able
to reattend. In this nal session, participants were again asked
to rate how well they could remember the target, aer which
124 participants were shown a simultaneous, target-present lineup
and 121 a target-absent lineup. All were informed that the target
may or may not be present in the lineup. e participant rst
indicated whether or not the target was present. If they thought
the target was present, they indicated which lineup member they
thought was the target. In either case, they were asked for a
condence rating from 0 “completely guessing” to 100 “completely
certain.” Finally, participants were thanked and fully debriefed.
e facial composites that were produced by the participants
were shown to three people who were familiar with the four
targets employed in the study, who were asked to rate (out of
10) the similarity of each composite to the target. e raters
judged similarity by comparing each composite image to their
memory of the target’s face, rather than being provided with
an image of the target. Memorial judgments were employed as
previous research on composite construction has suggested that
using a particular image can lead to matching specic pictorial
elements between composite and photograph rather than face
recognition per se (Brace et al., 2006) and because in a police
investigation the hope is that someone familiar with the suspect
will recognize the composite image, rather than compare the
composite to a picture. e most forensically relevant method
of evaluating composites would be to use spontaneous naming,
as this would reect how composites are used in a real investigation.
While possessing ecological validity, spontaneous naming tends
to lead to oor eects and is dicult to operationalize because
priming eects mean a participant can only see one composite.
However, it is important to note that ratings are a proxy for
real-world naming and are not as ecologically valid.
RESULTS
As stated previously, of the 255 participants tested, data from
245 were included in subsequent analyses. To determine whether
the delays between viewing the crime and constructing the
composite/verbal description, or between constructing the
composite and viewing the lineup, diered between conditions,
3 (Condition) × 2 (Lineup: target-present, target-absent) ANOVAs
were conducted using the lengths of the two delays as the
dependent variables. e results showed no statistically signicant
main eects or interactions, with
h
p
2 values no larger than
0.016. In addition, logistic regression analysis revealed
non-signicant associations using lineup accuracy (whether a
correct or incorrect decision was made) as the DV and the
delays between crime and composite construction/description,
B = −0.001, p= 0.99, and composite construction/description
and lineup, B = −0.261, p = 0.34, as predictor variables.
Summary data of lineup outcomes are presented in Tab l e 1 ,
using the signal detection outcomes of hits, false alarms, misses,
and correct rejections.
e data in Tab l e 1 show that most participants (in all
conditions) made a correct identication from the target-present
lineup. Analysis of these data was based on the association
between condition and the outcome of the lineup: where
outcome was a “hit,” “miss,” or “false alarm” for target-present
lineups, or a “correct rejection” or “false alarm” for target-
absent lineups. A 3 (Condition) × 3 (Outcome) chi-square
test conducted on the target-present lineup data revealed a
statistically non-signicant result with a small eect size,
χ2(4) =1.3, p =0.86, ϕc =0.07. Tabl e 1 also shows that most
participants (in all conditions) correctly rejected the target-
absent lineup. A 3 (Condition) × 2 (Outcome) chi-square test
on the data from the target-absent lineups revealed a statistically
non-signicant result with a small eect size, χ2(2) = 0.47,
p = 0.79, ϕc = 0.06.
Although descriptive statistics (see Tab l e 1) showed
performance in the two composite conditions to bemore accurate
than the control condition (i.e., there were more hits for the
target-present lineups and more correct rejections for the target-
absent lineups), analyses of lineup outcome revealed
non-signicant dierences, with small eect sizes between the
E-FIT and control conditions for target-present lineups,
χ2(2) = 1.7, p = 0.92, ϕc = 0.05, and target-absent lineups,
χ2(1) = 0.10, p = 0.75, ϕc = 0.04, and for the EFIT-V and
control conditions for target-present lineups, χ2(2)=1.3, p=0.53,
ϕc = 0.13, and target-absent lineups, χ2(1) = 0.47, p = 0.49,
ϕc = 0.08. e dierence in lineup outcome between the E-FIT
and EFIT-V conditions was also non-signicant with small eect
sizes for both target-present, χ2(2) = 0.55, p = 0.76, ϕc =0.08,
and target-absent lineups, χ2(1) = 0.14, p = 0.82, ϕc = 0.04.
To further explore the dierence, specically the lack of a
dierence, between the conditions in terms of lineup outcome,
equivalence testing for categorical variables was performed
employing the procedure recommended by Shiskina et al.
(2018). is analysis, which is based on the Cramer’s V measure
of eect size and follows the Beribisky et al., (2018, as cited
in Shiskina et al., 2018) approach of setting the equivalence
TABLE 1 | Lineup outcomes by condition.
E-FIT EFIT-V Control
Target-present N=43 N=40 N=41
Hit 28
0.65 (0.50, 0.78)
29
0.73 (0.57, 0.84)
25
0.61 (0.46, 0.74)
False alarm 10
0.23 (0.13, 0.38)
7
0.18 (0.09, 0.32)
11
0.27 (0.16, 0.42)
Miss 5
0.12 (0.05, 0.24)
4
0.10 (0.04, 0.23)
5
0.12 (0.05, 0.26)
Target-absent N=41 N=40 N=40
Correct reject 25
0.61 (0.46, 0.74)
26
0.65 (0.50, 0.78)
23
0.58 (0.42, 0.71)
False alarm 16
0.39 (0.26, 0.54)
14
0.35 (0.22, 0.50)
17
0.43 (0.29, 0.58)
First row in each cell is the count, and second row is the proportion with 95% CI in
parentheses.
Pike et al. Facial Composites and Eyewitness Identication
Frontiers in Psychology | www.frontiersin.org 6 August 2019 | Volume 10 | Article 1962
bound at δ=0.3. As there is no direct method for computing
the lower and upper condence intervals in this instance, an
iterative approach was taken to determine values for ΔL and
ΔU (Smithson, 2003), using a function from the “DescTools”
package for the R programming environment (Signorell etal.,
2017). is procedure produced values for ΔL and ΔU for
the condition by lineup outcome analysis in the target-present
condition of (0.0, 0.13) and in the target-absent condition of
(0.0, 0.20). In both cases, the upper bound falls below the
equivalence bound of δ = 0.3, meaning that it is possible to
conclude that the relationship between condition and lineup
outcome is negligible for both target-present and target-
absent lineups.
e quality of the composites produced by the participants
in the E-FIT and EFIT-V conditions, as determined by the
ratings of the independent judges, was compared to see if the
images from one system were rated as being more like the
targets than the other. Analysis revealed that the likeness ratings
for the composites in the EFIT-V condition (M=5.01, SD=1.64)
were signicantly higher than those in the E-FIT condition
(M= 4.11, SD=1.36) with a medium eect size, t(118)=3.29,
p = 0.001, d = 0.60. e relationship between the likeness
rating given to the composite constructed by a participant
and that participant’s subsequent performance at the lineup is
of interest because it might be that composite construction
could impact memory dierentially depending on how well
the participant used the system. For example, the memory of
participants who produced a poorly rated composite may
be negatively aected, while the memory of those creating a
highly rated composite might be improved. To test this
relationship, logistic regression analysis was performed using
whether the participant made a correct or incorrect decision
at the lineup as the DV, and the mean likeness rating of the
independent judges as the predictor variable, and revealed a
non-signicant association, B = 0.07, p = 0.54.
On average, the likeness ratings provided by the witnesses
(M = 6.97, SD = 1.41) were higher than the mean ratings
provided by the judges (M=4.55, SD= 1.57), which a by-item
analysis revealed as signicant, t(119)=13.9, p<0.001, d=1.61.
ere was also a statistically signicant correlation between
the two measures, r(120) = 0.19, p = 0.037, albeit with a
relatively small eect size. Analysis of the witnesses’ ratings
revealed the same pattern of results as for the judges’ ratings,
though with a smaller eect size, with participants who created
an EFIT-V providing higher ratings on average (M = 7.28,
SD=1.43) than did participants who created an E-FIT (M=6.65,
SD = 1.34), t(118) = 2.51, p = 0.014, d = 0.45.
e condence data (see Tab l e 2) were analyzed using a
3 (Condition: E-FIT, EFIT-V, control) × 2 (Lineup: target-
present, target-absent) × 2 (Decision: correct, incorrect) ANOVA.
is revealed a non-signicant main eect of condition, F(2,
233) = 1.51, p = 0.22,
h
p
2 = 0.01; a non-signicant main
eect of lineup type; F(1, 233) = 0.06, p = 0.81,
h
p
2 = 0.00;
and a signicant main eect of decision accuracy (with higher
condence in correct than incorrect decisions), F(1, 233)=49.98,
p<0.001,
h
p
2=0.18. All of the two- and three-way interactions
were non-signicant.
For participants who made a selection (i.e., who either chose
the target or a foil) from the lineup, there were relatively
strong and statistically signicant positive condence-accuracy
correlations in the EFIT-V condition, rpb(50)=0.57, p <0.001,
and in the control condition rpb(53) = 0.44, p< 0.001, and a
weaker, but signicant, correlation in the E-FIT condition,
rpb(54) = 0.37, p = 0.007. e correlation coecient in the
control condition did not dier signicantly from the correlation
coecients in either the EFIT-V condition, Fisher’s Z = 0.86,
p = 0.39, or the E-FIT condition, Fisher’s Z = 0.42, p = 0.67,
and the dierence between the EFIT-V and E-FIT conditions
was also non-signicant, Fisher’s Z = 1.28, p = 0.20. For
participants who did not make a selection from the lineup,
there was a relatively strong and statistically signicant positive
condence-accuracy correlations in the control condition,
rpb(28) = 0.57, p = 0.002, and non-signicant correlations in
the E-FIT-V, rpb(30) =0.28, p=0.13 and the E-FIT condition,
rpb(30) = 0.24, p = 0.21. However, the correlation coecient
in the control condition did not dier signicantly from the
correlation coecients in either the EFIT-V condition, Fisher’s
Z = 1.30, p = 0.19, or the E-FIT condition, Fisher’s Z = 1.45,
p = 0.15, and the dierence between the EFIT-V and E-FIT
conditions was also non-signicant, Fisher’s Z =0.16, p=0.87.
DISCUSSION
e results of the current experiment showed no detrimental
or benecial eect of constructing a facial composite compared
to giving a verbal description only, regardless of whether a
feature-based system (E-FIT) or a PCA-based system (EFIT-V)
was used to construct the composite. Overall, participants
tended to perform accurately on the identication task, with
between 57.5 and 72.5% making a correct decision. Although
these gures are a little higher than those reported in some
eyewitness identication research, particularly aer the long
delay involved, the false alarm rates (which varied from 17.5
to 42.5%) are a good match for those in previous research
TABLE 2 | Mean condence (0–100%) in identication decision.
E-FIT EFIT-V Control Total
Target-present N=43 N=40 N=41
Hit 79.12 (16.95)
[72.83, 85.38]
78.93 (16.07)
[73.08, 84.78]
79.2 (17.36)
[72.39, 86.01]
79.07 (16.56)
[75.49, 82.66]
False alarm 71.0 (14.49)
[62.02, 79.98]
40.0 (20.0)
[25.18, 54.82]
60.91 (19.73)
[49.25, 72.57]
59.29 (21.24)
[51.42, 67.15]
Miss 58.0 (24.9)
[36.17, 79.83]
57.5 (23.63)
[34.34, 80.66]
40.0 (33.91)
[10.28, 69.72]
51.43 (27.42)
[37.07, 65.80]
Target-absent N=41 N=40 N=40
Correct reject 71.2 (20.27)
[63.25, 79.15]
73.45 (18.16)
[66.29, 80.25]
76.09 (17.45)
[68.96, 83.22]
73.45 (18.54)
[69.22, 77.67]
False alarm 62.81 (18.35)
[53.82, 71.80]
58.93 (21.32)
[47.76, 70.10]
58.93 (24.86)
[46.71, 70.35]
60.11 (21.38)
[53.99, 66.22]
Total 71.43 (19.26)
[67.31, 75.55]
69.11 (21.53)
[64.39, 73.83]
69.07 (23.15)
[64.03, 74.11]
69.89 (21.29)
[67.23, 72.56]
First row in each cell is the mean % with SD in parentheses, second is 95% CI in brackets.
Pike et al. Facial Composites and Eyewitness Identication
Frontiers in Psychology | www.frontiersin.org 7 August 2019 | Volume 10 | Article 1962
and estimates for misidentications in real investigations (Pike
and Clark, 2018). is suggests that there were no oor or
ceiling eects and that identication rates were at a level that
would have reected either a detrimental eect of composite
production, resulting in a reduction in the identication rate
compared to the control condition, or a benecial eect, whereby
identication rates would have increased compared to the
control condition.
e results, therefore, revealed that composite construction
using either a featural or holistic composite system did not
have an adverse or benecial eect on subsequent identication
accuracy, using methods and materials that were a “fair” test
of face memory (non-biased constructions and instructions
were employed) and a good match for the procedures and
delays involved in a real investigation.
ese results dier from those reported by Davis et al.
(2014), who found that creating either an E-FIT or EFIT-V
image led to more accurate performance at the subsequent
identication task. e likely explanation of this dierence is
that Davis etal. used relatively short delays between the initial
exposure to the target face and composite construction (from
0 to 30min) and then between construction and the identication
task (5 min to 32 h), while the current experiment employed
more ecologically valid delays of 2days and 11–35 days. is
explanation is supported by Mauldin and Laughery (1981),
who reported that the initial benecial eect of creating a
composite on subsequent identication diminished as the delay
between construction and identication increased.
e current results are similar to those of Davis etal. (2016),
who also found that composite construction using a holistic
system did not have an eect on later identication. As noted
by Tredoux etal. (2016), individual studies of possible composite
interference eects do show some variability, with only one
outlier article showing a strong, negative relationship between
composite construction and identication (Wells et al., 2005),
although it should benoted that other research (Topp-Manriquez
etal., 2016) using the (piecemeal, feature-based) FACES composite
system employed by Wells et al. has also reported a large
negative eect. Moreover, although some studies show a marginal,
if statistically signicant, eect in either direction, the general
picture that emerges is that creating a composite does not
alter a witness’ memory of the target face suciently to impact
later identication.
e results reported here found that the EFIT-V system
tended to produce more accurate composites, as determined
through ratings provided by independent judges. As was stated
previously, ratings are only a proxy for the spontaneous naming
that would be needed in a real investigation, so it does not
necessarily follow that the EFIT-V images would have been
more recognized than the E-FIT images. e images produced
by EFIT-V are much more photo-realistic than those of E-FIT,
which are not only lower resolution and greyscale, but do not
integrate the individual facial features seamlessly. is dierence
in the quality of the image produced is partly intrinsic to the
two systems of course, and not a factor that can be easily
separated out from the underlying representations and
construction techniques used to construct the facial image
itself. It could be that the high quality of the EFIT-V images
was a factor that contributed to their higher ratings, though
equally it is possible that their more photo-realistic nature
could have raised expectations regarding their accuracy,
potentially leading to stricter judgments if the representation
of the face was not so perfect. In the current experiment, the
ratings for the images produced by the two systems were fairly
similar, albeit statistically signicantly dierent, and there was
considerable overlap in the range of ratings for the E-FITs
(1.33–7.33) and EFIT-Vs (1.67–8.33), although it is possible
that the increase in the use of the higher ratings (e.g., >7)
for the EFIT-V images was driven in part by their higher
image quality.
Analyses of the quality of the composites produced (here
achieved with independent likeness ratings), can also shed light
on an interesting theoretical distinction between the piecemeal
and holistic construction processes and their dierential eects
on memory. Holistic systems (such as EFIT-V and EvoFIT)
were developed to overcome the aws in their predecessor
piecemeal systems (such as E-FIT) and, in so doing, create
images that were a better match to the face of the suspect.
However, if the quality of the composites produced by both
types of system was essentially the same, then this would
suggest that either the experimental conditions were insucient
to allow the eects of holistic construction to manifest, or
alternatively that holistic construction does not lead to more
accurate composites. In either case, an experiment where the
likeness of piecemeal and holistic composites was not suciently
dierent would also mean that the potential eects (whether
these be positive or negative) of holistic construction on the
memory of the participants for the face of the target would
not be revealed. In other words, if holistic construction was
insuciently inuential to impact composite quality, it could
also be insuciently inuential to impact eyewitness memory
for the face of the target. In the experiment reported here,
the holistic composites were judged to be more like the targets
than the piecemeal composites, meaning that the impact of
using a holistic construction method was apparent. However,
although the descriptive statistics did show a tendency for
participants who had constructed a holistic composite to bemore
accurate at the lineup, suggesting perhaps that their memory
of the target’s face had been enhanced, this dierence was not
statistically signicant from the performance of either the
participants who created a piecemeal composite or the control
participants. is suggests that although the holistic construction
method did lead to improved composites, any enhancement
of the participants’ memories was too short-lived to impact
subsequent identication of the target in the lineup. is
conclusion is consistent with that of Mauldin and Laughery
(1981), and the results of Davis et al. (2014) where a shorter
delay between construction and lineup was used, and
improvements to face memory were found.
Analyses of the likeness ratings can also contribute to choosing
between two potential hypotheses regarding why holistic and
piecemeal systems might aect eyewitness memory dierentially.
If holistic systems produce better images, then regardless of
the technique used in construction it could be that memory
Pike et al. Facial Composites and Eyewitness Identication
Frontiers in Psychology | www.frontiersin.org 8 August 2019 | Volume 10 | Article 1962
would be more aected simply because the witness would
be viewing an image more like that of the perpetrator, an
explanation very similar to that suggested by Mauldin and
Laughery (1981) in relation to piecemeal composites. is could
either have an enhancing eect, resulting through rehearsal of
the target face, or a negative eect because the similarity of
the image is more likely to change the original memory trace.
Alternatively, it could be that the holistic construction process
itself had a greater eect on memory for the target than
piecemeal construction, regardless of the quality of the resulting
composite image. e results reported here, that holistic
composites were judged on average to be more like the targets
but lineup performance was the same across holistic, piecemeal,
and control conditions, and particularly that likeness was not
a predictor of lineup accuracy, provide evidence that the former
hypothesis is unlikely to oer the best explanation. Instead,
the fact that holistic composites were rated as being more like
the targets than piecemeal composites provides evidence that
holistic construction is more eective than piecemeal, suggesting
further research on its impact on face memory is needed.
e participants in the EFIT-V condition also tended to
rate the accuracy of their own composite more highly than
did participants in the E-FIT condition. is result suggests
that the holistic construction approach adopted in the EFIT-V
system resulted in the participants thinking they had produced
better quality composites than the feature-based approach used
in E-FIT. Interestingly, even though there was an increase in
the quality of the EFIT-V images, there was no subsequent
impact on identication performance compared to either the
E-FIT or control conditions. is suggests that although holistic
construction may lead to subjective judgments that a better
composite has been produced, this does not interfere with, or
indeed enhance, long-term memory for the target face.
Although the current experiment employed methods that
sought a degree of ecological validity in some respects, not
every aspect of the methodology was an eective match for
a police investigation. One dierence was that while a witness
seeing a crime is likely to experience a degree, even a very
high degree, of stress, participants in an experiment will
experience a relatively low, or indeed no, degree of stress. e
ethical requirements of conducting research mean stressing a
participant is problematic, so this dierence between experimental
and real-life conditions is a standard issue for eyewitness
research (Lane and Houston, 2019). Deenbacher et al. (2004)
conducted a meta-analysis of the eects of stress on eyewitness
recall (36 independent tests) and identication (27 independent
tests), which provided considerable support for the hypothesis
that both recall and identication are adversely aected by
high levels of stress. Morgan et al. (2004) drew a similar
conclusion in one of the few studies to have introduced a
high level of stress to experimental participants, by employing
soldiers who were asked to identify their interrogators aer
having taken part in either a high- or low-stress interrogation.
Some studies have found contradictory ndings; for example,
Sauerland etal. (2016) found that stress had no robust, negative
impact on identication performance in a study that measured
salivary cortisol levels to control for the eectiveness of the
stressor. In addition, Brace et al. (2009), who used a eld test
in which participant-witnesses attended identication procedures
at a police station, found that dierences in the stress self-
reported by participants were not reected in their scores on
a standardized checklist.
As Brace etal. (2009) note, as well as the stress experienced
at the encoding stage while witnessing the crime, the stress
induced by the various aspects of police investigation at the
retrieval stage is also an important factor. In the case of
composite construction, police operators have reported that
the apparent stress levels of the witness can change during
construction, with some witnesses becoming notably anxious
as the face begins to resemble that of the perpetrator (Clark
et al., 2000), suggesting that the eects of stress on composite
construction may be complex and signicant, and certainly
worth further exploration.
Although nding that composite systems could lead to an
improvement in eyewitness identication evidence would have
been a positive eect for law enforcement, in the circumstances
of a real criminal investigation “no eect” is in many ways a
positive result. “No eect” in this case suggests that the same
witness can be asked both to create a composite of the perpetrator
and attempt to identify that perpetrator in an identication
procedure, without the former interfering with the latter. Pike
et al. (2019) drew a similar conclusion but pointed out that
the delay between composite creation and seeing a lineup is
oen much shorter in experimental work than in a real case,
so that care needs to be taken in translating the results. Even
though a relatively long delay was employed in the current
study, in a real case, the delay may be longer still; so again,
care is needed in translating the results to practice. It is also
important to remember that the research reported here, like
previous studies, involved constructing a composite fairly soon
aer seeing the “crime” and sometime before seeing a lineup,
so the results do not apply to an investigation with a long
delay between crime and composite construction and a relatively
short interval between construction and lineup.
In addition to variations in delay and stress levels, when
considering translation of the results reported here into a policing
context, it is important to remember that the current study
involves just a single experiment; so although the results showed
that composite construction did not interfere with subsequent
eyewitness identication accuracy, employing delays that match
those likely in a real criminal investigation, these data alone
are not grounds for making claims about policing practice.
DATA AVAILABILITY
e datasets generated for this study are available on request
to the corresponding author.
ETHICS STATEMENT
e studies involving human participants were reviewed and
approved by Human Research Ethics Committee, e Open
Pike et al. Facial Composites and Eyewitness Identication
Frontiers in Psychology | www.frontiersin.org 9 August 2019 | Volume 10 | Article 1962
University. e patients/participants provided their written
informed consent to participate in this study.
AUTHOR CONTRIBUTIONS
GP, NB, and JT contributed to the conception and design of
the study. GP and AV performed the statistical analysis. GP,
JT, and AV wrote the rst dra of the manuscript. GP, NB,
JT, HN, and AV wrote sections of the manuscript. All authors
contributed to manuscript revision, read and approved the
submitted version.
FUNDING
e research reported here was supported by EPSRC grant
GR/S06745/01.
ACKNOWLEDGMENTS
e authors would like to thank Dr. Anna Bobak (University
of Stirling) for her expert advice, and Sally Kynan (e Open
University) and Clare Jonas (University of Sussex) for their
assistance with data collection.
REFERENCES
Baddeley, A. D. (1979). “Applied cognitive and cognitive applied psychology:
e case of face recognition” in Perspectives on memory research. ed. L. G.
Nilsson (Hove, UK: Psychology Press), 367–388.
Brace, N. A., Pike, G. E., Allen, P., and Kemp, R. I. (2006). Identifying composites
of famous faces: investigating memory, language and system issues. Psychol.
Crime Law 12, 351–366. doi: 10.1080/10683160500151159
Brace, N. A., Pike, G. E., Kemp, R. I., and Turner, J. (2009). Eyewitness
identication procedures and stress: a comparison of live and video identication
procedures. Int. J. Police Sci. Manag. 11, 183–192. doi: 10.1350/ijps.2009.11.2.122
Brace, N. A., Pike, G. E., Kemp, R. I., Turner, J., and Bennett, P. (2006). Does
the presentation of multiple facial-composites improve suspect identication?
Appl. Cogn. Psychol. 20, 213–266. doi: 10.1002/acp.1181
Brace, N. A., Pike, G. E., and Turner, J. (2008). Holistic facial composite systems:
are they compatible with witness recall? Int. J. Cogn. Technol. 13, 42–49.
Clark, C., Pike, G., Brace, N., and Kemp, R. (2000). How important is interviewing
in composite construction? Paper presented at the 10th Annual Conference
of the European Association of Psychology and Law, Cyprus.
Comish, S. E. (1987). Recognition of facial stimuli following an intervening
task involving the Identi-kit. J. Appl. Psychol. 72, 488–491. doi:
10.1037/0021-9010.72.3.488
Davis, J. P., Gibson, S., and Solomon, C. (2014). e positive inuence of
creating a holistic facial composite on video line-up identication. Appl.
Cogn. Psychol. 28, 634–639. doi: 10.1002/acp.3045
Davis, J. P., orniley, S., Gibson, S., and Solomon, C. (2016). Holistic facial
composite construction and subsequent lineup identication accuracy:
comparing adults and children. J. Psychol. 150, 102–118. doi: 10.1080/
00223980.2015.1009867
Deenbacher, K. A., Bornstein, B. H., Penrod, S. D., and McGorty, E. K.
(2004). A meta-analytic review of the eects of high stress on eyewitness
memory. Law Hum. Behav. 28, 687–706. doi: 10.1007/s10979-004-0565-x
Dysart, J. E., Lindsay, R. C. L., Hammond, R., and Dupuis, P. (2001). Mug shot
exposure prior to lineup identication: interference, transference, and commitment
eects. J. Appl. Psychol. 86, 1280–1284. doi: 10.1037/0021-9010.86.6.1280
Ellis, H. D., Davies, G. M., and Shepherd, J. W. (1978). A critical examination
of the PhotoFIT system for recalling faces. Ergonomics 21, 297–307.
Frowd, C. D., Bruce, V., Ness, H., Bowie, L., Paterson, J., omson-Bogner, C.,
et al. (2007). Parallel approaches to composite production: interfaces that
behave contrary to expectation. Ergonomics 50, 562–585. doi: 10.1080/
00140130601154855
Frowd, C. D., Carson, D., Ness, H., McQuiston-Surrett, D., Richardson, J.,
Baldwin, H., et al. (2005). Contemporary composite techniques: the impact
of a forensically-relevant target delay. Leg. Criminol. Psychol. 10, 63–81. doi:
10.1348/135532504X15358
Frowd, C. D., Pitchford, M., Bruce, V., Jackson, S., Hepton, G., Greenall, M.,
et al. (2010). e psychology of face construction: giving evolution a helping
hand. Appl. Cogn. Psychol. 25, 195–203. doi: 10.1002/acp.1662
Frowd, C. D., Skelton, F., Atherton, C., Pitchford, M., Hepton, G., Holden, L.,
et al. (2012). Recovering faces from memory: the distracting inuence of
external facial features. J. Exp. Psychol. Appl. 18, 224–238. doi: 10.1037/
a0027393
Gawrylowicz, J., Gabbert, F., Carson, D., Lindsay, W. R., and Hancock, P. J. B.
(2012). Holistic versus featural facial composite systems for people with mild
intellectual disabilities. Appl. Cogn. Psychol. 26, 716–720. doi: 10.1002/acp.2850
Gibson, S. J., Solomon, C. J., and Bejarano, A. P. (2003, September). “Synthesis
of photographic quality facial composites using evolutionary algorithms” in
BMVC. 1–10.
Kempen, K., and Tredoux, C. G. (2012). ‘Seeing is believing’: the eect of
viewing and constructing a composite on identication performance. S. Afr.
J. Psychol. 42, 434–444. Available at: http://search.ebscohost.com/login.aspx?
direct=true&db=pbh&AN=79962083&site=ehost-live
Lane, S. M., and Houston, K. A. (2019). “Eyewitness memory” in Psychological
science and the law. eds. N. Brewer and A. B. Douglas (New York: e
Guildford Press), 104–129.
Laughery, K. R., Duval, G. C., and Fowler, R. H. (1977). An analysis of procedures
for generating facial images. Mug le project report number UHMUG-2.
(Texas: University of Houston).
Laughery, K. R., and Fowler, R. H. (1980). Sketch artist and identikit procedures
for recalling faces. J. Appl. Psychol. 65, 307–316. doi: 10.1037/0021-9010.65.3.307
Laurence, S., and Hole, G. (2012). Identity specic adaptation with composite
faces. Vis. Cogn. 20, 109–120. doi: 10.1080/13506285.2012.655805
Malpass, R. S., Tredoux, C. G., and McQuiston-Surrett, D. (2007). “Lineup
construction and lineup fairness” in e handbook of eyewitness psychology:
Volume II. eds. R. Lindsay, D. Ross, J. D. Read and M. P. Toglia (Mahwah,
NJ: Lawrence Erlbaum & Associates), 169–192.
Mauldin, M. A., and Laughery, K. R. (1981). Composite production eects on
subsequent facial recognition. J. Appl. Psychol. 66, 351–357. doi: 10.1037/
0021-9010.66.3.351
McClure, K. A., and Shaw, J. S. III. (2002). Participants’ free-hand drawings
of a target face can inuence recognition accuracy and the condence–
accuracy correlation. Appl. Cogn. Psychol. 16, 387–405. doi: 10.1002/acp.802
Meissner, C. A., and Brigham, J. C. (2001). A meta-analysis of the verbal
overshadowing eect in face identication. Appl. Cogn. Psychol. 15, 603–616.
doi: 10.1002/acp.728
Morgan, C. A. III, Hazlett, G., Doran, A., Garrett, S., Hoyt, G., omas, P.,
et al. (2004). Accuracy of eyewitness memory for persons encountered during
exposure to highly intense stress. Int. J. Law Psychiatry 27, 265–279. doi:
10.1016/j.ijlp.2004.03.004
Pike, G. E., Brace, N. A., Turner, J., and Kynan, S. (2005). Making faces with
computers: witness cognition and technology. Pragmatics & Cognition 13,
459–480. doi: 10.1075/pc.13.3.04pik. ISSN 0929-0907
Pike, G. E., Brace, N. A., Turner, J., and Vredeveldt, A. (2019). e eect of facial
composite construction on eyewitness identication accuracy in an ecologically
valid paradigm. Crim. Justice Behav. 46, 319–336. doi: 10.1177/0093854818811376
Pike, G. E., and Clark, C. (2018). “Identication evidence” in e psychology
of criminal investigation: From theory to practice. Issues in forensic psychology.
eds. A. Griths and R. Milne (London: Routledge), 133–153.
Sauerland, M., Raymaekers, L. H., Otgaar, H., Memon, A., Waltjen, T. T.,
Nivo, M., et al. (2016). Stress, stress-induced cortisol responses, and eyewitness
identication performance. Behav. Sci. Law 34, 580–594. doi: 10.1002/bsl.2249
Shiskina, T., Farmus, L., and Cribbie, R. A. (2018). Testing for a lack of
relationship among categorical variables. Quant. Methods Psychol. 14, 167–179.
doi: 10.20982/tqmp.14.3.p167
Pike et al. Facial Composites and Eyewitness Identication
Frontiers in Psychology | www.frontiersin.org 10 August 2019 | Volume 10 | Article 1962
Signorell, A., Aho, K., Alfons, A., Anderegg, N., Aragon, T., Arppe, A., et al.
(2017). DescTools: Tools for descriptive statistics. R package version 0.99.23.
Smithson, M. (2003). Condence intervals. ousand Oaks, CA: Sage.
Solomon, C. J., Gibson, S. J., and Maylin, M. (2012). “EFIT-V: evolutionary
algorithms and computer composites” in Craniofacial identication. eds.
C. Wilkinson and C. Rynn (Cambridge: Cambridge University Press),
24–41.
Tanaka, J. W., and Farah, M. J. (1993). Parts and wholes in face recognition.
Q. J. Exp. Psychol. A 46, 225–245. doi: 10.1080/14640749308401045
Topp-Manriquez, L. D., McQuiston, D., and Malpass, R. S. (2016). Facial
composites and the misinformation eect: how composites distort memory.
Leg. Criminol. Psychol. 21, 372–389. doi: 10.1111/lcrp.12054
Tredoux, C. G. (1998). Statistical inference on measures of lineup fairness.
Law Hum. Behav. 22, 217–237. doi: 10.1023/A:1025746220886
Tredoux, C. G., Nunez, D. T., Oxtoby, O., and Prag, B. (2006). An evaluation
of ID: an eigenface based construction system. S. Afr. Comput. J. 37, 90–97.
Tredoux, C. G., Sporer, S. L., Vredeveldt, A., Kempen, K., and Nortje, A.
(2016). Does making a face composite contaminate a witness’ memory? A
meta-analysis of the extant literature. Paper presented at the European
Association of Psychology and Law conference, Toulouse, France.
Wells, G. L., Charman, S. D., and Olson, E. A. (2005). Building face composites
can harm lineup identication performance. J. Exp. Psychol. Appl. 11,
147–156. doi: 10.1037/1076-898x.11.3.147
Wilford, M. M., and Wells, G. L. (2010). Does facial processing prioritize
change detection? Change blindness illustrates costs and benets of holistic
processing. Psychol. Sci. 21, 1611–1615. doi: 10.1177/0956797610385952
Yu, C. J., and Geiselman, R. E. (1993). Eects of constructing identi-kit composites
on photospread identication performance. Crim. Justice Behav. 20, 280–292.
doi: 10.1177/0093854893020003005
Conict of Interest Statement: e authors declare that the research was conducted
in the absence of any commercial or nancial relationships that could beconstrued
as a potential conict of interest.
Copyright © 2019 Pike, Brace, Turner, Ness and Vredeveldt. is is an open-access
article distributed under the terms of the Creative Commons Attribution License (CC
BY). e use, distribution or reproduction in other forums is permitted, provided the
original author(s) and the copyright owner(s) are credited and that the original publication
in this journal is cited, in accordance with accepted academic practice. No use, distribution
or reproduction is permitted which does not comply with these terms.
Available via license: CC BY
Content may be subject to copyright.
