A decade of evolving composites: regression- and meta-analysis
| Pages | 319-334 |
| DOI | https://doi.org/10.1108/JFP-08-2014-0025 |
| Date | 09 November 2015 |
| Published date | 09 November 2015 |
| Author | Charlie D. Frowd,William B. Erickson,James M. Lampinen,Faye C. Skelton,Alex H. McIntyre,Peter J.B. Hancock |
| Subject Matter | Health & social care,Criminology & forensic psychology,Forensic practice |
A decade of evolving composites:
regression- and meta-analysis
Charlie D. Frowd, William B. Erickson, James M. Lampinen, Faye C. Skelton, Alex H. McIntyre
and Peter J.B. Hancock
Dr Charlie D. Frowd is based at
the Department of Psychology,
University of Winchester,
Winchester, UK.
William B. Erickson and
Professor James M. Lampinen,
both are based at the
Department of Psychological
Science, University of
Arkansas, Fayetteville,
Arkansas, USA.
Dr Faye C. Skelton and
Dr Alex H. McIntyr, are
Lecturers, both at the
School of Life, Sport and Social
Sciences, Edinburgh Napier
University, Edinburgh, UK.
Professor Peter J.B. Hancock
is based at the Department
of Psychology, University of
Stirling, Stirling, UK.
Abstract
Purpose –The purpose of this paper is to assess the impact of seven variables that emerge from forensic
research on facial-composite construction and naming using contemporary police systems: EvoFIT, Feature
and Sketch.
Design/methodology/approach –The paperinvolves regression-and meta-analyseson composite-naming
data from 23studies that have followed proceduresused by police practitioners for forensicface construction.
The corpusfor analyses contains6,464 individualnaming responsesfrom 1,069 participants in 41 experimental
conditions.
Findings –The analyses reveal that composites constructed from the holistic EvoFIT system were over
four-times more identifiable than composites from “Feature”(E-FIT and PRO-fit) and Sketch systems; Sketch
was somewhat more effective than Feature systems. EvoFIT was more effective when internal features were
created before rather than after selecting hair and the other (blurred) external features. Adding questions
about the global appearance of the face (as part of the holistic-cognitive interview (H-CI)) gives a valuable
improvement in naming over the standard face-recall cognitive interview (CI) for all three system types tested.
The analysis also confirmed that composites were considerably less effective when constructed from a long
(one to two days) compared with a short (0-3.5 hours) retention interval.
Practical implications –Variables were assessed that are of importance to forensic practitioners who
construct composites with witnesses and victims of crime.
Originality/value –Using a large corpus of forensically-relevant data, the main result is that EvoFIT
using the internal-features method of construction is superior; an H-CI administered prior to face
construction is also advantageous (cf. face-recall CI) for EvoFIT as well as for two further contrasting
production systems.
Keywords Meta-analysis, EvoFIT, Sketch, Facial composite, Feature system, Holistic CI,
Regression analysis
Paper type Research paper
A prevalent view, 15 years ago, among forensic practitioners was that procedures used to
construct composites had been largely optimised and the effectiveness of a composite was
determined by the ability of the witness. The procedures used to construct composites in
a forensicsetting are detailed (describedin Fodarella et al., 2015),with the aim of allowing a witness
(who may alsobe a victim) to create the best likenessof an offender. In brief,for traditional “feature”
systems, a practitioner would administer Cognitive-Interviewing (CI) techniques, to obtain a
description of the offender’s face from a witness, and then prepare an “initial”composite: a face
with facial features(eyes, nose, mouth, etc.)to match this description. Next, thepractitioner would
present alternative features from the software system for the witness to select best-matching
items, with selected features adjusted for size and placement. Finally, a paint package could be
used to add lines,wrinkles, etc. Alternatively, a forensic artist would produce a composite sketch.
The artist wouldobtain a description of the offender’s face from a witness (viaa CI) and prepare an
“initial”,faintly drawn sketch. Artistand witness would work together on theconfigural properties of
Received 16 August 2014
Revised 21 February 2015
18 August 2015
Accepted 18 August 2015
DOI 10.1108/JFP-08-2014-0025 VOL. 17 NO. 4 2015, pp. 319-334, © Emerald Group Publishing Limited, ISSN 2050-8794
j
JOURNAL OF FORENSIC PRACTICE
j
PAG E 31 9
the face (spacing of features), and then to increase the overall level of detail. In either case, the
resulting composites would be shown to other people (police officers and members of the public)
to identify.
To quantify the effectiveness of composites, Frowd et al. (2005a) defined a “gold”standard by
which composite systems (or new techniques) should be assessed in the laboratory: composite
construction should follow procedures used in police interviews and composite effectiveness
should be based on people’s ability to spontaneously name these images. Using this procedure,
a decade of research has revealed that fairly good performance emerges when the interval is up
to a few hours in duration from encoding a target face to constructing a composite of it.
Constructors using sketch and modern feature systems prevalent in the USA, UK and Europe
(e.g. E-FIT, PRO-fit, FACES, Identikit 2000) create composites that other people name with a
mean of around 20 percent correct (e.g. Brace et al., 2000; Bruce et al., 2002; Frowd et al.,
2005a). However, when the retention interval is one or two days, a usual minimum in police
investigations, mean correct naming is usually low (M¼5 percent; e.g. Frowd et al., 2005b,
2007d). Thus, procedures used for face construction seemed to be neither effective nor optimal.
Considerable effort has sought solutions which are more closely aligned to face recognition
(a holistic process) than to face recall (describing a face). As we tend to recognise faces as complete
entities rather than by component parts (facial features) (e.g. Davies and Milne, 1982; Tanaka and
Farah, 1993), face construction should be effective if accomplished likewise. This concept has long
been implemented in modern feature systems: individual features are presented for selection in the
context of a complete face (Skelton et al., 2015). For the emerging “holistic”systems, this concept is
taken one step further: constructors repeatedly select whole faces (or whole-face regions) from arrays
of alternatives, with characteristics of selected items being “bred”together, to “evolve”acomposite.
They also contain scales for changing age and other global properties of an evolved face. Overall, the
approach is based on recognition, which is more stable over time than recall (Davies, 1983), and
requires holistic (global) processing of faces rather than explicit recall of features. There are three main
implementations: EvoFIT, which has been assessed extensively using the gold standard (Frowd,
2015); EFIT-V (Gibson et al., 2009), evaluated using the gold standard in one published study
(Valentine et al., 2010) and ID (Tredoux et al., 2006).
A crucial observation that led to a forensically-useful system, EvoFIT, concerns differences
regarding the way in which faces are processed when constructed and named. Face
construction is performed by a witness who is usually unfamiliar with a target (an offender), and so
a witness’s processing of the face is influenced strongly by external features (hair, ears and neck);
in contrast, internal features (the inner region encompassing eyes, brows, mouth, etc.) are
particularly important for recognition of a familiar face (e.g. Bruce et al., 1999; Ellis et al., 1979;
Young et al., 1985) –in this case, for successful naming of a composite (Frowd et al., 2007a,
2011). Frowd et al. (2010) used a Gaussian (“blur”) filter to de-emphasise external features in
EvoFIT arrays. They demonstrated that this technique helped constructors to create composites
with fairly good correct naming (M¼25 percent) after a two-day retention interval, presumably as
this prevented external features from dominating during construction of an unfamiliar face.
Composites with even higher naming (M¼45 percent) were produced when just internal features
were shown, with external features added thereafter (Frowd et al., 2012d); for an example face array,
see Fodarella et al. (2015) in this special issue.
A further important development was made by facilitating holistic processing prior to face
construction:after witnesses have freely recalleda target face using CI techniques (e.g. Wellset al.,
2007), they reflect silently on its character for one minute and then make seven whole-face
judgements –such as its level of perceived honesty or masculinity. These two whole-face
techniques, when used after a face-recall CI, form the holistic-cognitive interview (H-CI).
Constructors then build the face as normal. The H-CI improves composite naming from EvoFIT
(Frowd et al., 2012a), feature systems (Frowdet al., 2008) and artists’sketches(Kuivaniemi-Smith
and Frowd, unpublished, see Discussion).
Naming is improved still further when composites are viewed as a dynamic caricature (e.g. Frowd
et al., 2007c), an image format that exaggerates and de-emphasises distinctive aspects of
the face; and from side-on, to allow the face to appear long-and-thin (e.g. Davis et al., 2015;
PAGE320
j
JOURNAL OF FORENSIC PRACTICE
j
VOL. 17 NO. 4 2015
To continue reading
Request your trial