Ben-Shakhar, G. & Elaad, E. (2003). The validity of
psychophysiological detection of information with the guilty knowledge test: a
meta-analytic review. Journal of Applied Psychology, 88, 131-151.
The field of
law and criminal apprehension is confounded by the deceptions of the criminals
involved. Unfortunately, we as humans
are not very efficient at detecting such deception unaided, even for “experts”
within these fields. Science has
attempted to provide certain psychophysiological tests to aid investigators in
their attempts to circumnavigate these deceptions and apprehend these
criminals.
Three main
psychophysiological tests have emerged for this purpose: the polygraph, the
Control Questions Test (CQT), and the Guilty Knowledge Test (GKT). The standard polygraph is the lie-detector
test that is probably familiar to all of you.
The CQT compares physiological responses measured during answers to
crime relevant and control questions.
The crime relevant questions typically come in the form of “Did you do
it?” questions that relate to the specific crime at hand. The control questions have a more general
basis, but are still about crimes (e.g. “Have you ever stolen anything in your
life?”). Unfortunately many critics are
quick to point out that a lack of standardization and improper control
questions severely limit the CQT.
On the
other hand, the GKT has a stronger backing from the scientific field in
general. The GKT measures the
psychophysiological response of participants during multiple-choice
questions. Each question contains a set
of answers, one being the relevant answer and the rest being control
answers. The relevant answers relate to
the specific crime in question, while the control answers are structured so
that they would not be discernible from the relevant answers by innocent
suspects.
The current
article focuses on reviewing past studies of the GKT through the process of a
meta-analysis. The meta-analysis was
conducted with four main goals in mind. The first was to set a rating of
validity based on a variety of studies.
This validity would span a large set of data, as compared to the
relatively smaller sets contained within each study. The second goal of the meta-analysis was to examine the
differences in results found in the collected studies and what may have had a
hand in causing these differences.
Sampling error may be to blame, or, more interestingly, the differences
could be due to the variations of some of the features within the studies. The third goal was to test for moderating
factors if the variability in the studies exceeded what would be expected from
sampling errors. The fourth goal, as is
usually the case in a meta-analysis, was/is to encourage future research of the
GKT based on results of the current study.
Overall the
results are very promising for the validity of the GKT. The average effect size of the studies is a
very impressive 1.55. The average r
was 0.55, which is impressive as well, although there was a wide variability
between studies.
The
variance found across the studies exceeded that variance expected based on
sampling errors. Tests for homogeneity
also hinted at the possibility of moderating factors as well. Thus the study systematically sought out
potential moderators of three types: motivational based moderators, desire of
participant to deceive the experimenter, and the number of questions on the
GKT.
In the case
of motivational factors, the studies were divided into two categories, high and
low motivation. The high motivation
group yielded a significantly higher effect size than did the low motivation
group. Differences in d were
found in the deceptive response analysis, but the differences were not
significant.
A series of
regressions was conducted on the number and repetition of questions on the
GKT. The regressions revealed that the
number of questions on the GKT predicted the effect size. The number of questions was also found to
account for 12% of the variance found for d.
As another
analysis, the studies were divided into five categories based on the method of
study employed. The mock-crime paradigm
showed the highest effect size at 2.088, significantly higher than the effect
size from the other four categories. A
subset of these mock-crime studies selected as the most relevant to real life
showed an effect size of 3.12 with a mean r of 0.79.
Two main
conclusions were drawn from the results of the study. The first conclusion is that psychophysiological tests can effectively
predict individuals that have guilty knowledge (note: this is different from
actually being guilty) and individuals who do not have this knowledge. The second conclusion is that there is, in
fact, a strong variation between the studies examined, caused by a variety of
moderation factors.
All in all
the results suggest that with proper administration the GKT can be a very
strong indicator of guilt. The effects
size found overall (1.55) is extremely impressive, and when the data set is
whittled down to the studies that are most real world relevant the effect size
increases significantly.
Limitations
with the current study’s application to the real world do exist, however. The information selected for the GKT in
these experiments is specifically taught to the participants of the
studies. In a real criminal
investigation there is no such guarantee that the information chosen for the
GKT is learned by the suspect. In addition
the amount of information available for the GKT varies widely from case to
case, both in salience and quantity. It should be noted, however, that law
enforcement officials in Japan have used the GKT preferentially, suggesting
that case information may not be as limited as anticipated.
Time of
testing is another strong limitation of these findings. In the experiments the participants are
tested almost immediately. In a real
world investigation the GKT may not be administered until large time intervals
have elapsed, again affecting memory.
The leak of
information about the case is a very common real world occurrence. In criminal investigations information may
be leaked from any number of sources. Innocent suspects with knowledge of this
information could be detected by the GKT.
A false-positive detection of this nature is very troubling.
Ben-Shakhar
and Elaad contend that the GKT should become integrated into the law
enforcement field. Additionally they
contend that the GKT has a strong chance to become justified enough to become a
court- admissible measure.