Encyclopedia of Psychology and Law

so as to make a decision. In the WITNESS model, the

decision to make an identification is based on the best

match and the next best match. Specifically, the evi-

dence in favor of identifying the best match is given

by a weighted combination of the value of the best

match and the difference between the best match and

the next best match. This is illustrated below as

EVID=wb(BEST) +wb−n(BEST −NEXT),

where BEST is the value of the best-matching lineup

member and BEST −NEXT is the difference between

the best-matching and the next best-matching lineup

member. Thus, a person may be identified if he or she

is a very good match to memory or if he or she is a

much better match than anyone else in the lineup.

These two ways of making an identification are similar

to the distinction between absolute versus relative judg-

ments made by Gary Wells in 1984. The degree to

which the identification is based on absolute versus rel-

ative judgments depends on the values of the weights

wband wb−n(where wband wb−nmust sum to 1.0). A high

value of wbis consistent with an absolute judgment,

whereas a high value of wb−nis consistent with a rela-

tive judgment. According to the model, the witness

makes an identification if EVIDis above a criterion c

and makes no identification if EVIDis below c.

Applications of the WITNESS Model

In fitting the WITNESS model (or any model) to data,

the model’s parameters are free to vary. Of course, the

parameters should vary in ways that are sensible. A

few examples of how the model’s parameters vary

with experimental manipulations are illustrated below.

OObbsseerrvvaattiioonn aanndd MMeemmoorryy

The changes in the conditions of observation and

memory are produced in the model by variation in the

storage parameter s.The value of swill be higher to the

extent that the witness has a better opportunity to view

the perpetrator, and the delay between the crime and the

presentation of the lineup is shorter. Thus, values of s

reflect both the failure to store information in memory at

the time of the observation and the failure to retain that

information over time. When there is less accurate infor-

mation about the perpetrator in memory, the match val-

ues become noisy and converge, making it less likely

that the witness will correctly identify the perpetrator

and more likely that the witness will identify an inno-

cent person. Any factor that affects the storage or loss of

information in memory would be modeled in terms of

lower values of s,including shorter exposure durations,

longer retention intervals, high levels of stress and fear,

and distraction due to weapon focus effects.

LLiinneeuupp CCoommppoossiittiioonn

Police officers usually select foils (innocent per-

sons in a lineup) based on their similarity to the sus-

pect in the lineup. In the model, the critical parameter

is S(F,S)—that is, the similarity between the foils and

the suspect. Specifically,S(F,S) is the proportion of

features shared by both the suspect and the foil. The

suspect should be more likely to be identified when

S(F,S) is low and less likely to be identified when

S(F,S) is high. In the extreme, the model predicts that

if S(F,S) is very low, the suspect may be very likely

to be identified, whether he or she is guilty or not.

LLiinneeuupp IInnssttrruuccttiioonnss

Because the suspect may be innocent, it is impor-

tant to instruct the witness that the true perpetrator

may or may not be in the lineup and that the witness

is not obligated to make an identification. These

instructions are called unbiased because they are

unbiased with respect to the guilt or innocence of the

suspect. A biasedinstruction implies or explicitly

states that the perpetrator isin the lineup and that it is

for the witness to identify him or her (disallowing a

“None of the above” response). It is not surprising that

biased instructions result in a higher identification

rate, irrespective of whether the suspect is guilty or

innocent. The problem, of course, is that as the identi-

fication rate increases (without any change in the

accuracy of memory), the likelihood that an innocent

person will be misidentified also increases.

One way to model the effect of instructions is to lower

the value of c(the decision criterion) for biased instruc-

tions and raise it for unbiased instructions. However, it is

not the only way to model instruction effects. One could

assume that biased instructions result in changes in the

weights. For example, the biased instructions could have

their effect by inducing witnesses to make identifications

even if the best match is not that high, provided that he

or she is a better match than anyone else in the lineup. In

the model, this shift would be instantiated as a decrease

in wband an increase in wb−n.

WITNESS Model——— 865

W-Cutler (Encyc)-45463.qxd 11/18/2007 12:44 PM Page 865