Orne, M. T., & Thackray, R. I. Group GSR technique in the detection of deception. Perceptual and Motor Skills, 1967, 25, 809-816.
MARTIN T. ORNE 2 AND RICHARD I. THACKRAY 2,3
Institute of the Pennsylvania Hospital and University of Pennsylvania
Summary.-- A technique to average GSR across Ss is proposed to facilitate detection of deception. Differential responsivity to "lie" vs "neutral" stimuli is augmented by averaging out idiosyncratic responses to neutral stimuli. Its feasibility was demonstrated by recording individual and group GSRs of Ss questioned in small groups. The probability of detecting shared information was greatly increased. Limitations and possible applications of the technique are discussed.
The detection of deception is possible only to the extent that physiological responses to significant critical stimuli (items of information concerning which S is attempting to deceive) clearly differ from responses to irrelevant stimuli (control or neutral items). The interrogator, who must discriminate between these two kinds of events, is essentially faced with a problem of signal detection. As in other such problems, it is essential to isolate the signal (physiological response to critical items) from the noise (physiological response to non-critical items). Detection will be improved by any operation which increases the signal-to-noise ratio, whether it enhances the signal, reduces the noise, or both.
Previous work has shown that the signal can be enhanced (i.e., responses to critical items can be magnified) by manipulating such variables as S's motivation, role perception, the requirement to give a "no" response to the critical items, or the method of stimulus presentation (Gustafson & Orne, 1963, 1964, 1965a, 1965b). It has also been possible to reduce the effective noise level by averaging over a relatively large number of critical and non-critical items. Lykken (1960) achieved a substantial increase in the probability of detection by this method.
Another way to decrease noise level offers itself if a number of Ss share the same critical items of information. One may then average across Ss as well as across items. In any particular experimental situation, the probability of detection achieved in this way should be greater than what would be attainable by considering Ss individually. Any single S is sure to make some idiosyncratic responses to non-critical items that happen to have special meaning for him; such responses are "noise" which make detection more difficult. In averaging across Ss, this noise tends to cancel itself out, while the signal relating to shared information becomes more clearly discernible.
1 This work was supported in part by the United States Army Research and Development Command, Department of the Army, under Research Contract No. DA-49-193-MD-2647.
2 The authors wish to express their appreciation to Mary Jo Bryan for her assistance in the running of Ss and data reduction, as well as to Howard A. Keiser who served as E1 in both experiments. The authors also wish to express their appreciation to Frederick J. Evans, Ulric Neisser, Donald N. O'Connell, and Peter W. Sheehan for their critical comments in the preparation of this manuscript.
3 Now at Civil Aeromedical Institute, Oklahoma.
809
810 M. T. ORNE & R. I. THACKRAY
As part of a program of research dealing with the detection of deception, two separate studies were carried out which utilized a group GSR technique and explored its general feasibility. The substantive findings of these studies are reported elsewhere (Thackray & Orne, 1967). The experimental designs are therefore only briefly outlined below. The relationship between individual and group responses which appears in our results is probably independent of the specific experimental situations in which the detection of deception was studied.
We have been unable to find any use of group GSR techniques in the detection of deception. However, the use of such procedures in evaluating group attitudes has recently come to our attention (cf. Hansel, 1951).
METHOD
Subjects
Male college undergraduates were solicited for participation through bulletin boards and advertisements. All Ss were between the ages of 18 and 23, and none had participated in similar research previously. In both studies Ss were run in groups of six or seven. Twenty such groups were employed in the first experiment and 12 in the second. Half of the groups in each experiment were assigned at random to the "guilty" and half to the "innocent" conditions. Altogether, 66 guilty and 68 innocent Ss were tested in the first experiment; 38 guilty and 35 innocent Ss participated in the second.
Procedure
Upon reporting to the lab, a given group was taken to a room and an experimenter (E l) played taped instructions appropriate for the condition to which the whole group had been assigned. The recording played to members of "guilty" groups explained that each of them was to assume the role of an enemy espionage agent (courier) who knows certain code words that are required for deciphering a message carried by another agent. They were told that they would be interrogated later by another person (E 2), who would attempt to determine whether or not they actually were couriers. Ss in the guilty groups were then required to learn the code words, associate to them, reproduce them in the original, reverse and alphabetical order, and perform arithmetic and other irrelevant tasks prior to again reproducing the code words. Ss were told that failure to reproduce the words correctly would disqualify them from the experiment. The procedure took approximately 1 hr. and was designed to ensure overlearning as well as increase Ss' ego involvement in the task. (The words varied with groups and were randomly selected from common categories such as animals, flowers or trees.) Ss in "innocent" groups were not told the words; they were informed that they were under suspicion of being enemy agents and would be interrogated in the same manner as their counterparts in the real courier groups. They spent an hour completing a questionnaire about their attitudes toward various aspects of lie detection.
811 GROUP GSR IN DETECTING DECEPTION
At the completion of the hour, Ss were informed that they would now be interrogated. Innocent Ss were reassured that they had no reason to fear the procedure since they had nothing to hide. Guilty Ss were told that the interrogator would do his best to obtain a confession, that it would be difficult to deceive him, but that it was possible to "beat the polygraph;" highly intelligent, mature individuals would be able to control their emotions well enough to succeed .4 The group was then turned over to E2 (the interrogator) who had no prior knowledge as to whether it was composed of innocent or guilty Ss. E2, however, did know the six code words.
An Offner Type R eight-channel Dynograph was used to record GSR responses. Two Grass EEG electrodes were taped to the index and middle fingers of each S's hand, with the leads going to individual bridge circuits. The output of each bridge was AC-coupled through a 1-sec. time constant to one channel of the Dynograph. A constant gain setting was used for all Ss. To obtain the group GSR, each preamplifier output was coupled through separate 1 megohm resistors to the input of the driver amplifier on the eighth channel. Thus, the voltage appearing across the input of this channel was the voltage produced by the sum of the currents flowing through each of the 1 megohm resistors in the individual preamplifier outputs. Ss wore earphones and were blindfolded to minimize group interactions.
In Exp. I, certain procedures intervened between the instruction-hour and the actual interrogation. They were designed to give Ss varying kinds of feedback about the effectiveness of the polygraph as a lie-detector. Each S was asked to select a card from a deck; unknown to S, E2 was able to determine which card would be selected. Ss were then told that, as a preliminary to the interrogation about the code words, they would be asked about their card and that the polygraph would automatically deliver a shock each time they lied. The 13 values and 4 suits were then mentioned, each S being required to answer "no" each time. This procedure automatically forced all Ss to lie twice. Two Ss in each group were shocked when they responded "no" to the card they had actually drawn, just as the interrogator had promised. However, two others were shocked on cards which they had not drawn, and the rest received no shocks at all. At the end of this procedure, each S was given an opportunity to "confess" whether he was a courier; none did so. The main interrogation then began. A tape-recording was played which included five critical words (the sixth was used later, in a procedure irrelevant to the present paper), six neutral words, and a few other items, at intervals of 8 to 15 sec. S was required to say "yes" or "no," but was told that any reply of "yes" would be an admission of guilt. No further shocks were administered.
In Exp. II, the intervening procedure with card selection and shock was
812 M. T. ORNE & R. I. THACKRAY
omitted entirely. Each critical word was embedded in a separate block of six neutral words. After all 42 words had been presented, the whole series was repeated in reverse order. In addition, a different type of interrogation, in which Ss knew what each word would be before it was presented, was also explored. The results obtained with this latter procedure, which sometimes preceded and sometimes followed the regular one, are not included in the data presented here. 5
RESULTS
The basic response data were the changes in skin resistance (GSRs) produced by the verbal stimuli. Magnitude of GSR to each stimulus was defined as the maximum pen deflection (in millimeters) which occurred within 3 sec. after the onset of the word. To assess the efficiency of detection, it was necessary to compare the GSRs given to code words with those produced by neutral items. In the first study each critical word was ranked relative to six neutral words. A rank of 1 would indicate that the largest response occurred to the critical word whereas 7 would mean that it elicited the smallest response. The rank of each of the 6 critical words was averaged to obtain an average rank score for each S. Fig. 1 shows the average rank of each individual S. All Ss were run in groups
813 GROUP GSR IN DETECTING DECEPTION
of 6 or 7 (which included the three subgroups of Ss differentially shocked during the pretest) and their group GSRs, which automatically averaged across individuals, were recorded simultaneously (Fig. 1 shows these data). These group data were analyzed in the same way: each critical response ranked in reference to neutral words and the ranks for each critical word averaged to yield an average rank score for each group of Ss. Fig. 2 shows these data.
In Exp. II Ss were also run in groups of 6, but no shock was used, all Ss being treated alike. Further, each code word was embedded in a block of four neutral words. Scoring procedures were identical except that each code word had a possible rank of 1 to 5. The ranks for each of the 6 critical words were again averaged to yield each S's score. Fig. 3 shows the distribution for the individual scores and Fig. 4 shows the average rank group GSR for each group of 6 Ss.
In both experiments the differences between guilty and innocent Ss are well beyond the .001 level (Mann-Whitney U tests). However, because of the way the studies were designed, the responses of the innocent group of necessity approximate a normal distribution and there is, therefore, a considerable overlap between the distribution of the guilty Ss and the innocent ones. As a result, the ability of the procedure to discriminate guilty individuals from innocent individuals is relatively poor. For the purpose of discriminating, results from the group
814 M. T. ORNE & R. I. THACKRAY
815 GROUP GSR IN DETECTING DECEPTION
data are far more satisfactory. In Exp. I (Fig. 3), the innocent and guilty distributions are entirely separate except for one innocent group, and in Exp. II there is no overlap at all. 6
DISCUSSION
A comparison of the innocent and guilty distributions of the group GSRs with the corresponding individual distributions clearly demonstrates the advantage of averaging across individuals. Such a procedure may have practical value in certain real-life situations, as when it is known that a group of individuals share information they are unwilling to divulge. Under these conditions, significant questions should yield the largest responses. The typical countermeasure of emitting false positives to irrelevant items would be difficult to employ when a number of individuals are interrogated simultaneously, unless they knew the nature of the interrogation and were given the opportunity to agree beforehand. False positive responses by individuals would randomize across Ss, whereas the systematic variance contributed by the items of common knowledge would be directly additive and thus easily discernible.
It should be pointed out that very special conditions existed in the studies reported. All Ss were either guilty or innocent and, if they were guilty, they all shared the same information. In a real-life situation it is more likely that any given group will contain individuals innocent of any special information as well as those privy to it. Future work will need to establish the relationship of group size and the number of individuals sharing the guilty information to the probability of detection. One would predict that the larger the group, the greater the percentage of innocent individuals it could contain while continuing to provide a high probability of detection. While considerable work will be required to explore the parameters and the potential utility of the group GSR to detect shared guilty information, our data suggest that the technique itself is feasible and should have potentially significant applications.
REFERENCES
GUSTAFSON, L. A., & ORNE, M. T. Effects of heightened motivation on the detection of deception. Journal of Applied Psychology, 1963, 47, 408-411.
GUSTAFSON, L. A., & ORNE, M. T. The effects of task and method of stimulus presentation on the detection of deception. Journal of Applied Psychology, 1964, 48, 383-387.
GUSTAFSON, L. A., & ORNE, M. T. Effects of perceived role and role success on the detection of deception. Journal of Applied Psychology, 1965, 49, 412-417. (a)
GUSTAFSON, L. A., & ORNE, M. T. The effects of verbal responses on the laboratory detection of deception. Psychophysiology, 1965, 2, 10-13. (b)
HANSEL, C. E. M. Measurement of "group responses." Quarterly Bulletin of the British Psychological Society, 1951, 2, 6-7.
816 M. T. ORNE & R. I. THACKRAY
LYKKEN, D. T. The validity of the guilty knowledge technique: the effects of faking. Journal of Applied Psychology, 1960, 44, 258-262.
THACKRAY, R. I., & ORNE, M. T. Methodological studies in detection of deception.United States Army Medical Research and Development Command Report, filed with Defense Documentation Center No. AD-645102, January, 1967.
Accepted November 6, 1967.