ROSENFELD, J. P. (1997) Scaled P300 Scalp Profiles in Detection of Deception September 2002, Report No. DoDPI02-R-0005. Department of Defense Polygraph Institute, Fort Jackson, SC 29207-5000.
Three studies were performed. The first two dealt with countermeasures to brain wave-based detection of deception in concealed information test protocols. There are two kinds of such protocols extant. One, the "6-probe" protocol utilizes multiple different crime details whose brain responses are averaged together. This protocol was easily defeated in the first study, as the detection rates dropped from 82% detection in the simple guilty group to 18% in the guilty group using a countermeasure. Although the average reaction time distinguished these two groups, there was enough overlap in their reaction time distributions such that in any individual case, one could not use reaction time to infer deception. The second protocol, the " 1-probe" protocol uses one crime detail as a probe in each of as many runs as one wishes. One group was run in three successive weeks as 1) a guilty group, 2) a countermeasure group, and 3) finally without the explicit use of the countermeasure. In the first week, 92% of the subjects were detected. The countermeasure dropped this rate to 50%. In the final third week, without explicit use of the countermeasure, only 58% were detected. There was no overlap in the reaction time distributions of the first two weeks, suggesting that the explicit countermeasure use could be detected with reaction time. In the third week, the reaction time distributions looked like those of the first week, so that test beaters would not be detected with reaction time. Other matters examined were 1) a comparison of individual brain wave analysis methods; 2) a comparison of naive versus sophisticated subjects, and 3) a comparison in terms of workload between the 1-probe and the 6-probe protocols. The third study looked at brain maps in detection of deception. In it, twenty-three subjects had P300 event-related brain potentials (ERPs) from 30 scalp sites recorded in response to relevant, control, irrelevant, and target questions as in screening. They also gave behavioral responses. A Principal Component analysis (like a factor analysis) was performed on the sites so as to reveal clusters (principal components) of correlated sites. Four clusters were found to account for 74% of the total variance in all averaged ERPs. Sites within clusters were then averaged to yield virtual sites on which analyses were performed. We compared scalp distributions to relevant versus control stimuli and found that 73% of the subjects guilty of either one (n=7) or two(n=8) items were correctly identified using criteria which did not produce any false positives in seven innocent subjects. We also saw that simple amplitude at the first principal component was greater for guilty responses than for control responses. Principal component clusters were more sensitive in this regard than individual sites. It is recommended that the distribution approach be further investigated since it is probably less vulnerable to a countermeasure than simple amplitude.
Key Words: Psychophysiological detection of deception, P300, screening tests, event-related potentials, brain maps.