shown 295 live line - ups and found suspect, fi ller and no identifi cation rates of
41%, 21% and 39% respectively. More recently Wright & Skagerberg (2007)
reported suspect, fi ller and no identifi cation rates of 58%, 21% and 21% respec-
tively in a study of 134 witnesses shown video line - ups.
Although there are differences across these studies (some reported and some
not known – e.g., it is possible that some datasets include identifi cations made
in non - stranger cases), the general pattern that emerges is that nearly one in
three positive identifi cations made by witnesses in these cases is clearly wrong as
the overall weighted averages across the studies are suspect, fi ller and no iden-
tifi cation rates of 45%, 21% and 40% respectively (numbers do not add to 100
due to missing data in the studies reported by Pike et al ., 2002 ). Note that it
is likely that witnesses in most of these studies were confronted with nine -
person arrays, the British standard. Given that 21% of witnesses in these studies
were ‘ guessing ’ one of the innocent fi llers (an average of 2.6% per fi ller), we
Recent Developments in Identifi cation Science and Practice
275
would have to expect – assuming these arrays were all perfectly fair – that
another 2.6% of the witnesses guessed the suspect by chance. Unfortunately,
as we detail below, other evidence indicates arrays are not entirely fair; thus,
a higher (perhaps much higher) percentage of witnesses may be guessing the
suspect.
A body of research on actual witnesses is beginning to emerge in North
America, though the studies to date all suffer from signifi cant limitations.
Tollestrup, Turtle & Yuille (1994) examined Royal Canadian Mounted Police
records on 119 robbery and 66 fraud cases (42 and 45 fi les respectively were
discarded because they lacked eyewitness descriptions or identifi cation
attempts) reported positive identifi cations of suspects in 31.7% of the Vancouver
photo arrays (most with eight photographs) they studied (vs. 45% reported in
the UK studies). Overall, 46% of robbery victims, 33% of robbery witnesses
and 25% of fraud victims identifi ed suspects. Unfortunately, the police records
were incomplete and therefore did not permit the researchers to differentiate
non - identifi cations from fi ller identifi cations:
Identifi cation outcomes were entered in the fi les in a variety of ways such as
‘ negative results, ’ ‘ unable to identify police suspect, ’ ‘ not in the line - up, ’ ‘ pointed
out suspect and one other as looking like perpetrator, ’ ‘ positive ID, ’ and ‘ weak
ID. ’ … for the most part, we could not distinguish reliably between outcomes
in which an eyewitness rejected the photo spread and those in which he or she
failed to select the police suspect. (p. 153)
Studies in the USA are more recent. Behrman & Davey (2001) examined
Sacramento, California - area police records for 271 cases involving 349 crimes
(the vast majority were armed robberies). They examined a total of 258 fi eld
show - ups, 289 photographic line - ups and 58 live line - ups, and reported that
76% of witnesses who viewed show - ups made an identifi cation and 48% of
witnesses who observed a photographic line - up identifi ed the suspect. However,
because the police fi les were incomplete, Behrman & Davey (like Tollestrup
et al ., 1994 ) could not determine how many identifi cations of known innocent
fi llers were made by witnesses viewing photo - arrays. In live line - ups, where the
records did include reports of mistaken fi ller identifi cations, Behrman & Davey
reported that 50% of witnesses identifi ed the suspect (an unknown percentage
of those identifi cations were errors); the rate of erroneous identifi cations of
fi llers was 24% and 26% of the witnesses did not make a choice. As in the UK
archival data, about one in three positive identifi cations was clearly wrong and
the average rate of guessing was 5% per fi ller, which might mean (per the
suspect bias multiplier noted earlier) that 15% of suspect identifi cations were
also guesses. Behrman & Richards (2005) also reported a study based on some
of the data included in the Behrman & Davey (2001) project, but it appears
that the 2005 data also do not offer a suitable basis for estimating overall
witness error rates (selections of fi llers). The 2005 study draws on 424 pho-
tographic line - ups (for which Behrman & Davey were unable to determine
276
Handbook of Psychology of Investigative Interviewing
fi ller identifi cation rates) and 37 live line - ups. The focus of the analyses is 238
suspect identifi cations vs. 68 fi ller identifi cations which the authors were able
to identify.
Klobuchar, Steblay & Caligiuri (2006) reported the results of an unusual
fi eld study conducted in Minnesota in 2004, in which identifi cation informa-
tion was collected in 280 line - ups from 117 cases involving 206 eyewitnesses.
The unusual aspects of the study were that the police used blind procedures
(i.e., the line - up administrator did not know the identity of the suspect) and
sequential line - ups when presenting arrays to witnesses. In the sequential pro-
cedure employed in Minnesota, photos were presented one at a time, with a
decision made about each photo (witnesses were permitted to view the entire
sequential display as often as desired).
In 178 line
- ups the witnesses were looking for perpetrators who were
strangers to them. In these line - ups 35% of witnesses picked the suspect and
11% picked a fi ller. Thus, about one in four positive identifi cations was clearly
wrong, and the fi ller guessing rate was about 2% per fi ller. Assuming the
suspect is identifi ed three times more often than the fi llers due to bias in the
line - up, perhaps 6% of the 35% of suspect identifi cations were guesses. Overall,
53% made no choice. As noted above, sequential arrays have been shown to
reduce choosing by witnesses, perhaps because they reduce witnesses ’ oppor-
tunity to compare the members of an array and possibly pick the face that
offers the best match to the memory of the witness. In the Minnesota study
about 46% of witnesses saw the sequential array more than once. For 33 wit-
nesses who viewed the array only once, the rates of suspect and fi ller identifi ca-
tion were 42% and 3% respectively compared to 32% and 13% for 31 witnesses
who viewed the array twice and 43% and 29% for 14 witnesses who viewed it
three times. Repeated views were thus associated with a signifi cantly increased
likelihood of a mistaken identifi cation.
The results of a second large fi eld study in Illinois were reported in 2006
(Mecklenburg, 2006 ). This study compared witness selections in non - blind
simultaneous line - ups (i.e., the presenting offi cer knew the identity of the
suspect) and blind sequential line - ups (the presenting offi cer did not know the
identity of the suspect) in several jurisdictions in the Chicago area. Table 15.3
reports the results in cases involving 548 stranger - identifi cation attempts.
Table 15.3: Effects of simultaneous vs. sequential
presentation on identifi cation rates
&nbs
p; Simultaneous
Sequential
(Number)
(319)
(229)
Suspect
59.9%
45.0%
Filler ID
2.8%
9.2%
No ID
37.6%
47.2%
Recent Developments in Identifi cation Science and Practice
277
The Illinois data were complemented by a report (p. 43 of the Illinois
report) of suspect and fi ller identifi cation rates in 2,677 simultaneous non -
blind live line
- ups conducted in Queens, New York during 2001
– 5. The
overall suspect identifi cation rate was 54% and the reported fi ller identifi cation
rate was 3%. Although the Illinois results have been claimed by some as evi-
dence for superior performance in non - blind simultaneous arrays, the study
has been criticized by leading psychologists (non
- eyewitness researchers)
Schacter, Dawes, Jacoby, Kahneman, Lempert, Roediger & Rosenthal (2008)
for confounding blind and sequential vs. non - blind and simultaneous proce-
dures. Schacter et al . note: ‘ Our reading of the materials forces us to conclude
that the confound has devastating consequences for assessing the real - world
implications of this particular study ’ (p. 4).
It will be apparent that the non - blind simultaneous fi ller identifi cation rates
from Illinois (3%) and Queens (3%) are markedly discrepant from those
reported in the various UK studies (which averaged 21%) and the Behrman &
Davey live line - up rate of 24%. Possible explanations for the discrepancy are
buried in the Illinois report: the low number of fi ller identifi cations emerged
from a non - blind simultaneous procedure in which it appears that the police
could ignore identifi cations that did not meet their ‘ probable cause ’ standards
(pp. iii – iv), a practice that seems akin to the Queens practice in which ‘ an
identifi cation [was recorded] only if it was based upon a high level of confi -
dence ’ , so that all tentative identifi cations were recorded as ‘ no identifi cation ’
(p. 43). These practices raise two questions:
1. Are substandard suspect IDs ignored just as often as substandard fi ller IDs?
Some might think that results in Illinois and Queens may be far more
skewed by selective non - reporting of fi ller IDs in non - blind presentations
than through any effort of the police to steer witnesses to suspects.
2. Does the failure to report even substandard IDs deprive defendants of
information that may be useful in their defence? Jurors and judges may
wish to know about such errors.
There are also indications that a substantial proportion of identifi cations in
Queens are actually second/confi rmatory identifi cations of suspects following
photographic or other forms of identifi cations of suspects; this practice would
also conceal foil or other forms of misidentifi cations prior to the confi rmatory
line - ups. Analyses by Wells
& Lindsay
(1980) and Clark
& Wells
(2008)
underscore that in many situations fi ller identifi cations and non - identifi cations
can be highly diagnostic with respect to the quality of witness memory and
identifi cation accuracy in single and multiple witness situations.
Unfortunately, it seems that legal decision - makers such as jurors may not
be properly attuned to the diagnostic value of non - identifi cations. Even if the
non - identifying witness is offered in court, McAllister & Bregman (1986) raise
doubts about the ability of jurors to give effective weight to non - identifi cation
evidence. These researchers presented mock jurors with a series of one and
278
Handbook of Psychology of Investigative Interviewing
two witness combinations of identifi cation evidence. Their results, using ratings
on a 1 (innocent) to 9 (guilty) scale, showed a reliable increase in guilt ratings
for a single identifying witness (M = 6.64) relative to a non - identifying witness
who neither identifi ed nor rejected the defendant as the perpetrator (M = 5.14),
but only a very slight and non - signifi cant decrease in guilt ratings for a single
witness who said the perpetrator was not in the line - up (5.06) relative to the
non - identifying control. The results show that mock jurors believed the single
identifi cation to be diagnostic of guilt, but mistakenly believed the rejection
of the line - up to have essentially no probative value.
As suggested above, fi ller identifi cations are important partly because
they raise questions about the witnesses
’ memory, but also because fi ller
identifi cations can serve as a basis for estimating the rates of innocent
suspect identifi cations. If witnesses are ‘ guessing ’ fi llers at a rate of 25%, as in
Behrman
& Davey
’ s
(2001) archival study or Steblay
& Dysart
’ s
(2008a)
meta - analysis, the average fi ller draws about 5% of witnesses selections in per-
fectly fair arrays, and 5% would have to be guessing the suspect. If the suspect
is the perpetrator 80% of the time, then four of those fi ve guesses would iden-
tify a guilty person (though based on somewhat questionable ‘ evidence ’ ) and
one in fi ve suspect guesses would identify an innocent person – the overall rate
of mistaken identifi cations of innocent suspects would be 1 in 100 in this
scenario.
Unfortunately, as we have noted, there is a growing body of research indi-
cating arrays are biased against suspects. It is not implausible that 15% of
witnesses (rather than 5% in this instance) are guessing the suspect; and, again,
if 20% of suspects are innocent, then 3% of witnesses will identify an innocent
suspect. And lest the point be lost, another 25% of witnesses would be guess-
ing a fi ller, thus raising signifi cant doubts about the quality of their memory
and impugning the reliability of any subsequent identifi cations they might
make.
The bottom line is that if we are interested in getting some sense of the
number of innocent suspects chosen by witnesses with defective memories in
fi eld studies, we need to know the fi ller identifi cation rates of witnesses and
the degree of line - up bias. With this information a reasonable estimate of the
rates of innocent suspect identifi cations can be made, which would be valuable
to defendants to raise questions about the reliability of identifi cations by wit-
nesses who have previously identifi ed a fi ller.
The s tate of e yewitness i dentifi cation
s cience and p ractice
Although psychological research on the sources of eyewitness identifi cation
error started more than a century ago (Whipple, 1909 , in the fi rst of a series
of reports on research mostly conducted in Europe, provides a succinct English
Recent Developments in Identifi cation Science and Practice
279
language summary of the earliest research), and despite the fact that we have
long had systematic evidence that eyewitness error is a primary source of erro-
neou
s convictions, it is only in the past two decades that the rapidly growing
body of scientifi c research and practice has begun to converge. Several factors
have provided an impetus to this convergence. Recent DNA exculpations and
exploration of the causes of those exculpations have signifi cantly raised public
and criminal justice practitioner awareness of the important role that eyewit-
ness errors play in producing erroneous convictions. Archival research, particu-
larly from the UK, supplemented by research conducted in controlled settings,
has provided further compelling documentation of the high rates of errors
among actual witnesses who choose/guess foils. Decomposition of those
errors indicates that the guessing rate is signifi cantly higher than foil choices
alone would indicate. In the past few decades scientifi c research on the sources
of errors and the role that identifi cation procedures and practices play in pro-
ducing errors has matured to the point where it was possible, a mere decade
ago, to advance a conservative set of practice recommendations (Wells et al .,
1998 ), based on a mix of scientifi c and logical analyses. Some of these recom-
mendations were quickly embraced by policy - makers in the form of the NIJ
guidelines on identifi cation procedures.
Recommended practices – suspects should not stand out from foils in arrays,
witnesses should be instructed that the person who committed the crime may
or may not be in the line - up, confi dence judgements should be collected
and recorded at the time of an identifi cation – found their way into the NIJ
guidelines. Other recommendations were considered, but were not incorpo-
rated. The guidelines did not include the recommendation that identifi cations
should be conducted double - blind. Nor did they include a recommendation
for sequential line - ups, a practice that received favourable consideration by
Wells et al .
In recent years there have been halting steps to evaluate the impact of these
practices in fi eld settings using experimental and non - experimental methods.
Although the early studies have drawn signifi cant criticism with respect to
their designs and data - recording practices, further research – some employing
stronger research designs – is under way. It is likely that in the next few years
we will possess strong evidence about the ways in which the procedures rec-
Handbook of Psychology of Investigative Interviewing: Current Developments and Future Directions Page 53
