A New Psychophysiological Detection of Deception Examination for Security Screening
Studies 1 - 3
Briefing Report
October, 1994
Sheila Reed
Department of Defense Polygraph Institute
Fort McClellan, AL 36205
Introduction
Although security screening examinations have been utilized by the government for a number of years, there is no research to support the validity of the psychophysiological detection of deception (PDD) format - counterintelligence scope polygraph (CSP) that is currently utilized. The one major study which was conducted to assess the validity of the CSP format, reported less than chance accuracy for correctly identifying the guilty examinees. In addition, there is a great deal of controversy regarding the types of questions utilized with the CSP format. The CSP is a control question format in which the probable lie control (PLC) questions are viewed, by some people, as being intrusive and offensive. Also, it requires careful manipulation of the examinee to properly pretest PLCs so that they are effective. Therefore, a series of studies was designed to develop a new format which would be less offensive to the examinee and which could be shown to have a reasonable measure of validity for both guilty and innocent examinees. Currently, three of these studies have been conducted.
The new security screening format, the test for espionage and sabotage (TES) is the first PDD format to be developed scientifically. The new format, as well as the scoring and decision criteria were based on previous research conducted at the Department of Defense Polygraph Institute (DoDPI). Unlike conventional PDD examinations in which the test (8 to 13 different questions) is repeated three times (three charts), the new format asks only one series of questions. However, the questions are repeated within the series. Therefore, each relevant question is asked three times, as follows: IR1 IR2 1C1 1R1 1R2 1C2 2R1 2R2 2C1 3R1 3R2 2C2. Each relevant is compared to the stronger of the two controls which bracket it. Then the three scores for each relevant is summed, yielding a score for each question. The two question scores are summed to provide a total score. A decision of "NO SIGNIFICANT RESPONDING (NSR)" is rendered if the total score is +4 or greater, with both questions being positive. A decision of "SIGNIFICANT RESPONDING (SR)" is rendered if either question score is -3 or less, or if both question scores are -2.
This new format limits the number of relevant questions (2) that can be asked on an examination. Since the typical CSP examination includes 4 relevant questions, two TES examinations are conducted. Each test addressing two of the 4 relevant questions. Subtest A, for purpose of the research, utilized the relevant questions: R1 - Have you committed an act of sabotage, during this project? and R2 - Have you committed an act of espionage, during this project? Subtest B utilized the relevant questions:
R3 - During this project, have you provided classified material to an unauthorized person and R4 - During this project, have you had unauthorized contact with a foreign national.
The other significant difference between the new TES format and the CSP format is that the TES utilizes a different type of control question - the directed lie control (DLC). The examinee is explicitly directed to lie to the DLCs. Again, previous research conducted at DoDPI suggested that the DLCs yielded greater accuracy than did the PLCs. In addition, the DLC is less intrusive and offensive to the examinee and it does not require much psychological manipulation of the examinee. Therefore, the expertise of the examiner is not as great an issue.
The first study compared the new TES format (TES) to the CSP format utilizing PLCs (PLC) and to the CSP format utilizing DLCs (DLC). The study utilized 306 examinees recruited from the local community. Approximately one third of the examinees were programmed guilty by their participation in one of four mock espionage scenarios. The 18 examiners were DoDPI trained and were certified with at least two years experience. Standard field instruments were utilized to conduct the examinations.
A total of 29 examinees were excluded from the data analyses for the following reasons: Eight guilty examinees gave pretest confessions; six examinees were not given the examination because they were not suitable medically; seven examinations were incomplete; and 8 innocent examinees were not included because they had "real world" explanations for their responses to the relevant questions (e.g. a young college student, the son of a high ranking officer on Ft McClellan, lived in the international house at the local university. His roommate was a Romanian student who continually asked the student questions regarding fort operations and repeatedly asked the student to "get him on the fort". This individual responded to the unauthorized contact and disclosure questions).
For the innocent examinees, the results indicated no significant differences in accuracies among the three groups. However, for the guilty examinees, the TES format was significantly more accurate than either of the other two formats, X2(2) = 6.054, p < .0485, (see Table 1). The accuracies reported do not include inconclusive decisions. The inconclusives are reported separately. There were no significant differences among the groups with respect to inconclusive rates.
Table 1
Original examiners' per cent correct decisions and inconclusive rates for the three formats.
inconclusives % correct decisions initial unresolved Innocent Guilty Inn Gui Inn Gui PLC 95.3% 55.6% 23.1% 10.3% 1.5% 6.9% DLC 95.2% 58.6% 19.7% 29.0% 6.0% 6.5% TES 88.9% 83.3% 21.4% 13.3% 3.5% 0.0% Blind analyses of the data yielded similar results. Table 2 depicts the data. The correlations of the original examiners' question scores with the blind examiners' question scores (correlations were calculated for each question) ranged from .80 to .87 and were significant at p<.0000. Analyses indicated that there were no significant differences among the three conditions for the innocent examinees but the accuracies were significantly different for the guilty examinees, X2(2) = 6.85, p < .0325. The TES format was significantly more accurate with the guilty examinees than either of the other two formats.
Table 2
Blind scorers' per cent correct decisions for the three formats.
% correct decisions Innocent Guilty PLC 93.2% 52.2% DLC 94.4% 42.9% TES 88.5% 81.0% The second study was a validation study of the TES format only. Data from the first study was analyzed and new scoring criteria were developed and utilized during the second study. Otherwise, the parameters of the second study were identical to those of the first study. Examinees (90) were recruited from the community. Approximately one-third were programmed guilty utilizing the scenarios from the first study. The 10 examiners were DoDPI trained, were certified and had utilized the TES format in field situations for a one month period. Standard field instruments were utilized to conduct the examinations.
Eight examinees were excluded from the analyses (1 pretest confession, 2 incomplete examinations and 2 innocent examinees with "real world" explanations for their responses). The accuracy for the innocent examinees was 98.0% and 83.3% for the guilty examinees. The initial inconclusive rate was 17.0% for the innocent and 10.0% for the guilty, with an unresolved inconclusive rate of only 1.9% for the innocent and 0/0% for the guilty.
The third study was designed to 1) answer the question of whether or not examiners' subjective opinions bias their data analyses; 2) collect digitized data to utilize in developing a computer scoring algorithm; 3) expand the parameters of the TES format - e.g. increase the list of acceptable control questions, allow flexibility in the order questions are asked, etc.; and 4) to ascertain whether the phrase "during this project", which was utilized as part of the relevant questions in the first two studies, significantly affected the outcome. The set up of the study was identical to the two previous studies.
Two hundred and eighty one examinees and 30 examiners participated. Eight of the examiners utilized AXCITON computerized systems to conduct their examinations (the AXCITON digitizes the physiological data). Half of the examinations were conducted by an individual examiner who did the pre-test, in-test and test data analysis. The other half of the examinations were conducted by teams of two examiners each. One examiner conducted the pre-test and the other examiner conducted the actual examination. Both examiners independently scored the tests. Immediately after each part of the examination which they conducted, all examiners indicated their subjective opinion of the examinee's innocence or guilt. In addition, half of each group utilized relevant questions which contained the phrase "during this project" and half utilized relevant questions which did not contain the phrase. The relevant questions were reworded slightly to make them more compatible with the questions being utilized in the field. The questions were: R1 - (During this project), have you been involved in espionage; R2 - (During this project), have you committed sabotage; R3 - Have you disclosed classified information to any unauthorized person, (during this project); and R4 - Have you had any unauthorized foreign contacts, (during this project).
A total of 38 examinees were excluded from the data analyses for the following reasons: Four guilty examinees gave pretest confessions; nine examinees were not given the examination because they were not suitable medically; twelve examinations were incomplete; and 13 innocent examinees were not included because they had "real world" explanations for their responses to the relevant questions.
There were no significant differences in accuracies for guilty or innocent, between the group that used the phrase "during this project" and the group that did not utilize the phrase. In addition, there were no significant differences between the groups with respect to inconclusive rates. The accuracies are depicted in Table 3. The outcome accuracies do not include inconclusive. Inconclusives are reported separately.
Table 3
Original examiners' per cent correct decisions and inconclusive rates for the two conditions (not using "during this project" and using "during this project") and all examinees combined.
inconclusives % correct decisions initial unresolved Innocent Guilty Inn Gui Inn Gui Without "during" 83.8% 83.7% 7.9% 19.2% 1.2% 0.0% With "during" 85.7% 70.3% 9.6% 6.9% 0.0% 2.6% Combined 84.8% 77.9% 8.7% 16.3% 65% 1.1% Blind analyses of the data yielded similar results. Table 4 depicts the data. The correlations of the original examiners' question scores with the blind examiners' question scores (correlations were calculated for each question) ranged from .81 to .87 and were significant at p<.0000. Analyses indicated that there were no significant differences in accuracies for guilty or innocent, between the group that used the phrase "during this project" and the group that did not utilize the phrase.
Table 4
Blind scorers' per cent correct decisions for the two conditions (not using "during this project" and using "during this project") and all examinees combined.
% correct decisions Innocent Guilty Without "during" 84.6 75.6% With "during" 85.9% 72.2% Combined 85.3% 74.1% Analyses of the influence of the examiners' bias on the analysis of the test is incomplete.
Although, the accuracy rates of the third study are not as great as the previous two studies (there were several problems which occurred during the third study - including examinee contamination), all three studies indicated scientifically acceptable accuracies for both innocent and guilty examinees. Since during the first study, the DLC group did not perform any differently than the PLC group, it is assumed that the increase in accuracy for the TES format is not specific to the DLC questions. It is hypothesized that the advantage of the TES may be due to the limitation of the number of significant questions that are asked on each examination. With a CSP examination, an examinee must focus on 4 relevant questions and 3 probable lie questions (which they are convinced are as significant as the relevant questions). Thus, their attention is very diffused. With the TES, the examinee focuses on only two relevant questions. The relevant questions on the second TES examination are not mentioned until after the first two questions have been cleared. Therefore, their attention is more focused. It is hypothesized that a single issue test would be even more effective.