33
0
–
–
–
Face shape
12
100
0
–
10
85
Eye colour
5
75
11
78
4
100
Nose
3
37
2
–
–
–
Facial hair
10
3
0
–
2
100
Mouth
2
39
1
–
2
20
Accent
31
32
-
–
47
99
Note
‘ – ’ = unavailable or incalculable data.
75.6%. Results of others ’ studies are similar or slightly higher (Fashing et al .,
2004 ). More precisely, witnesses ’ descriptions of sex and ethnicity are almost
always consistently accurate, whereas descriptors of facial and other character-
istics are poorer or inconsistent across studies (see Table 14.2 ).
Thus, witnesses have a general impression of their assailants, but cannot
recall discrete features (nose, mouth, etc.). In other words, eyewitnesses ’ verbal
descriptions of criminals are usually too few in number and too vague to
identify a specifi c suspect. Furthermore, the majority of information available
for those in search of a perpetrator (e.g. age, build, height, hair, etc.) may be
incorrect or can be easily altered (e.g. hair colour or length). This results in
an inappropriate selection of likely suspects from a mugshot database because
the algorithm may select suspects who match the vague description of the
criminal, but not their actual appearance.
Results of laboratory studies dealing with various aspects of human memory
offer another way to investigate the nature of person descriptions. The majority
of descriptions provided contain approximately 10 features, with an overall
accuracy rate greater than 70% (see Table 14.3 ). In a study of verbal recall,
244
Handbook of Psychology of Investigative Interviewing
Table 14.3: Completeness and accuracy of descriptions from laboratory studies
Studies
Completeness
Accuracy (%)
Wells, Rydell & Seelau (1993)
5.1
–
Lindsay, Martin & Webber (1994)
7.4
–
Finger & Pezdek (1999) a
11.6
90
Finger & Pezdek (1999) b
12.1
90
Finger & Pezdek (1999) c
15.2
85
Finger & Pezdek (1999) d
12.9
88
Geiselman et al . (2000)
5.3
87
Tunnicliff & Clark (2000)
6.3
–
Meissner, Brigham & Kelley (2001)
9.0
90
Meissner (2002) e
6.5
83
Meissner (2002) f
5.7
83
Meissner (2002) f
6.0
83
Meissner (2002) g
4.9
77
Brown & Lloyd - Jones (2003)
15.9
71
Pozzulo & Warren (2003)
9.9
87
Notes
‘ – ’ = unavailable data.
a. experiment 1, 10 min delayed recall. b. experiment 2, 1 hr delayed recall. c. experiment 3,
5 min delayed recall. d. experiment 3, 24 min delayed recall. e. experiment 1, 15 min delayed
recall. f. experiment 2, 15 min delayed recall. g. experiment 2, 1 week delayed recall.
Lindsay, Martin & Webber (1994) observed that height was reported by 86%
of their participants, followed by build (51%), gender (46%), age (45%), eth-
nicity (43%) and weight (22%). For facial features, upper features are reported
more than lower, and in the following relative frequency order: hair, eyes,
nose, face shape, eyebrows, chin, lips and mouth (Ellis, Shepherd & Davies,
1980 ; Laughery, Duval & Wogalter, 1986 ). It is worth noting that Davies,
van der Willik & Morrison (2000) observed a similar order among participants
who constructed a composite of a face: participants began their assigned task
by selecting hair, and then applied a downward selection strategy. In addition,
all of these results are compatible with face processing, in particular the preva-
lence of upper facial features (notably hair) in face perception and recognition
(Tanaka & Farah, 1993 ). In fact, although descriptions reported by experi-
ment participants are signifi cantly more complete than those provided by real
witnesses (Lindsay et al ., 1994 ), they remain insuffi cient in terms of the infor-
mative value which might help police offi cers to detect the criminal. Despite
this, they contain more information about ethnicity, height, body build, hair
length, colour and style, eyes, facial hair, complexion and shape.
How c an p erson d escriptions b e i mproved?
Several methods can be used to collect person descriptions. For example, wit-
nesses can write down their description (free recall instruction: ‘ Tell me all
A Method to Enhance Person Description
245
you can about the physical appearance of the criminal ’ ) or be administered a
facial feature adjective list (FFAL; Ellis, 1986 ). The FFAL contains items where
the participant rates individual facial features and shapes, with the option of
responding ‘ Don ’ t know ’ or ‘ Not applicable ’ . Simply by the nature of the task,
the checklists, however, force the witness to consider a particular feature, even
if they have no memory for it, by determining if it corresponds to a visual
memory of the face. This is why a feature checklist produces more incorrect
features than a free recall task (Wogalter, 1991; 1996 ). This method, although
easy to use and appearing to be a priori common sense, is not compatible with
the endeavours of law professionals because of the poor quality information
obtained from witnesses. Although a feature checklist produces more exhaus-
tive recall than self - generated descriptions, feature checklists lead to an increase
in the error rate in comparison to spontaneous testimony (Stern, 1902 ; Borst,
1904 ; Whipple, 1909, 1913 ; Cady, 1924 ; Goulding, 1971 ; Dent & Stephenson,
1979 ). In addition, feature checklists may subsequently interfere with partici-
pants ’ ability to identify the target person (Wogalter, 1991, 1996 ; Demarchi,
Py, Parain & Groud - Tan, 2006 ). The counterargument is that a verbatim free
report is usually poor and ineffective in making an arrest (Lipton,
1977 ),
despite a good accuracy rate.
The nature of the instructions given to the witness also modulates
both the quantity and quality of descriptors produced by limiting, expand-
ing or cautioning the participants on the information to be generated.
For example, Meissner, Brigham & Kelley (2001) found that asking some-
one to
‘ be sure to report only those details that they are confi dent of,
and do not attempt to guess at any particular feature ’ ( warning instruction )
signifi cantly increases the overall description quality in comparison to a stan-
dard ( ‘ Describe in as much detail as possible the face you saw ’ ) or a forced
recall ( ‘ It ’ s important to report everything. Try not to leave out any details
about the face even if you think they are not important
’ ). Nevertheless,
even when the warning instruction was used, the quality of descriptions was
very poor.
Despite the crucial importance of a perpetrator ’ s description in the investi-
gative process, few studies have attempted to increase the completeness and
quality of descriptions. Some researchers, for example, used the standard
Cognitive Interview (CI) protocol (Fisher, Geiselman, Raymond, Jurkewich
& Warhaftig, 1987 ; see Ginet & Py, 2001 , for a French version of the CI) to
enhance person recall. The CI incorporates four retrieval mnemonics: report
all , context reinstatement , change of order and change of perspective . The results
of several studies show that, despite these methods signifi cantly increasing the
completeness of descriptions, the information gain is weak (Geiselman, Fisher,
Firstenberg, Hutton, Sullivan, Avetissian & Prosk, 1984 ). Nevertheless, some
results lead us to conclude that particular instructions of the CI are more
effi cient for person description. For example, Boon
& Noon
(1994) and
Clifford & George (1996) have shown that the report all and context reinstate-
ment instructions were the more useful. These results, however, show that the
246
Handbook of Psychology of Investigative Interviewing
CI, even if it is a powerful tool in obtaining exhaustive event recall, had poor
results concerning person recall.
However, removing the inappropriate instructions of the standard CI
appears to increase the completeness of description. Finger & Pezdek (1999)
eliminated the change order and change perspective instructions, and added a
novel one, imagery , from which participants tried to visualize the face and
mentally form an image of it. Results showed that the modifi ed CI was more
effective than a standard interview. Py & Demarchi (2006) obtained similar
results in a fi eld study in collaboration with police offi cers. They kept both
the report all and the context reinstatement instruction, and added a holistic
processing instruction (based Craik & Lockhart, 1972 ; Bower & Karlin, 1974 ).
The modifi ed CI, in comparison to a standard police interview, signifi cantly
enhanced both the completeness and the quality of physical information
reported by participants about a person they had previously met informally
(experiment 1). Compared to a standard police interview, it also permitted
easier detection of the target person from an array of similar individuals
(experiment 2). This research showed that a modifi ed CI can enhance person
recall, providing that new protocols are used to complement a standard
CI for events and facts recall. Consequently, there is an increase in the time
taken to conduct two CIs. There is a pressing need to produce a simpler
method.
The p erson d escription i nterview ( PDI )
One way to improve person descriptions is to take a similar approach to that
used in the CI. A good understanding of episodic memory is required to create
a tool compatible with memory processes. The method must be based on the
following fundamental principles:
1. It is useful not only to ask someone to describe a person, but also to give
instructions to facilitate the recall task.
2. In accordance with ergonomic practice, any tools or instructions must
be adapted to the natural human process or strategy. An instruction
based on an unnatural (or reversed ) one could decrease the global quality
of the response, in terms of increasing the total amount of errors, for
example.
3. If only one strategy is spontaneously used to do a particular task, then a
complementary approach should improve it, such as an inversed strategy
(e.g. the change of order instruction in the CI). This is only relevant if the
complementary approach does not have a disruptive infl uence.
Scientifi c literature contains little information about the cognitive processes
or strategies involved in describing people and methods to improve them.
Therefore, it is necessary to understand how person verbal recall is organized
A Method to Enhance Person Description
247
and structured before creating an interview method. Two preliminary studies
were conducted to determine the strategies spontaneously used to describe a
person. In study 1, participants met a man (target) informally. After a 10 -
minute delay during which they completed a fi ller task, they had to describe
the target following a standard instruction ( ‘ Tell me all you can about the
person you met ’ ). Descriptive analysis showed that 84% of participants spon-
taneously reported general features fi rst (age, height, weight, attitudes, etc.),
and then carried on with more local traits (hair, eyes, etc.). We described this
spontaneous strategy as g eneral to specifi c . Half the participants continued to
use a downward top - to - bottom descriptive strategy, where upper facial elements
(e.g. hair) were reported fi rst, followed by lower traits (eyes, nose, mouth,
etc.). Finally, only 16% of the participants used no apparent strategy at all. In
study 2, which concerned only face description, participants were presented
with a single photograph of the same target man. All participants were given
the same instruction as in study 1 and subsequently described the target.
Descriptive analysis confi rmed that the majority of participants (87%) used the
top - to - bottom strategy to describe a face. In accordance with laboratory and
real - life case studies, it was also observed that lower facial features were never
recalled.
These observations lead to the conclusion that two instructions can be
created (fi rst fundamental principle presented above). The fi rst one, related to
the general - to - specifi c strategy, can be created which concerns all the physical
and trait judgements (second fundamental principle), and the second one,
related to the
top - to - bottom strategy, which concerns only face description
(third fundamental principle). This means that it is necessary to use two suc-
cessive free recalls, each following an instruction. The fi rst instruction aims to
enhance the general - to - specifi c strategy, which seems to be the only spontane-
ously used method to describe a person, because, as in the experiments above,
none of the participants started their spontaneous free recall by reporting
details fi rst. We ’ ve termed this strategy the general - to - specifi c instruction (or
GSI) and is articulated as follows:
Try to describe the person you saw. Be as complete as you can. Try not to provide
only detail, and please begin your description by reporting general and global
features of the person you saw, like silhouette, height, build, ethnicity, or per-
sonality, occupational impressions, etc., and carry on with facial or clothing
details.
Our pilot studies suggest that this instruction must be given fi rst. The
second instruction was specifi cally designed to increase the number of facial
descriptors, especially lower facial features. Neither a free recall instruction nor
one based on a natural strategy would be suffi cient to obtain lower facial
descriptors. In order to solve this problem, a reverse - order descriptive strategy
was used. We termed it the down - to - up instruction (or DUI). It is formulated
as follows:
248
Handbook of Psychology of Investigative Interviewing
Try to describe the face of the person you saw. Begin with the lower part of his
face – such as his chin – and then go up to the top.
Because it is concerned with only face description, this instruction must be
considered as complementing the fi rst one and must be used after the initial
description has been completed.
The effi ciency of these instructions was tested in comparison with a standard
instruction (Demarchi, 2003 ). Participants in this condition were presented
with the following instructions:
Please describe the person you saw. Try to be as complete as possible.
In the fi rst experiment, the GSI used at the beginning of the interview
obtained a 70% increase in correct information ( M = 12.35) compared to the
standard instruction ( M = 7.25), without a signifi cant increase in the number
of errors. Results from the second experiment showed that the DUI, used after
an initial free recall, led participants to reporting 242% more correct facial
information ( M = 5.30) compared to the standard instruction ( M = 1.55), but
also an 127% increase in errors (respectively M = 1.10 and M = 2.50 for stan-
dard and GSI instructions). In conclusion, the GSI led to more complete and
accurate descriptions than a standard instruction, and the DUI provides more
facial descriptors because participants were prompted to report lower facial
features, which would not otherwise have been recalled. In fact, an optimal
interviewing method for acquiring a person (or criminal) description consists
of two consecutive free recalls each following a particular instruction: fi rst, a
GSI, and second, a DUI. We have termed this the person description interview
(or PDI).
The laboratory test of the methods that comprise the PDI reported by
Demarchi (2003) is very encouraging. However, it is essential that the com-
Handbook of Psychology of Investigative Interviewing: Current Developments and Future Directions Page 47
