BLACKWELL, N. J. An evaluation of the effectiveness of the Polygraph Automated Scoring System (PASS) in detecting deception in a mock crime analog study. March 1994, Report No. DoDPI94- R-0003. Department of Defense Polygraph Institute, Fort McClellan, AL 36205.
The purpose of this research was to evaluate the effectiveness of a prototype scoring algorithm, the Polygraph Automated Scoring System (PASS), Version 2.0, when analyzing mock crime data. A database of psychophysiological detection of deception (PDD) measurements was collected using the Axciton Polygraph System. The PDD test format used was the Zone Comparison Test (ZCT). In addition, two types of control question tests (CQTs) were used during the study: (a) an experimental version of the directed lie control (DLC), and (b) the conventional probable lie control (PLC) currently in use throughout the PDD community. The data set is composed of 120 local civilian citizens who underwent a PDD examination, half of which committed a mock theft of money. The results showed that PASS accuracy on controlled laboratory data with known ground truth was far below that attained by APL during algorithm development when using "live" PDD examinations. In fact, rates of accuracy for both PASS and the PDD examiners were lower than anticipated. PASS was clearly more accurate in identifying individuals who had been programmed "innocent," while the PDD examiners were more accurate in detecting individuals who had been programmed "guilty." Use of the DLC versus the PLC made no apparent difference in overall accuracy for either PASS or the PDD examiners, though there were statistically significant differences when analyzed from the standpoint of programmed condition. Analysis also showed a PDD examiner performance decrement over time which contributed to the lower than expected examiner accuracy rates.
Key-words: Polygraph Automated Scoring System (PASS), Axciton, computerized scoring algorithms, probable lie control (PLC) test format, directed lie control (DLC) test format, control question test (CQT), Zone Comparison Test (ZCT), polygraph, forensic psychophysiology, psychophysiological detection of deception (PDD)
This report describes a study designed to evaluate the relative accuracy of examiner and computer algorithm scoring of psychophysiological detection of deception (PDD) examinations. It is important, if not essential, that this type of cross validation study be completed to assess the validity of computer algorithms designed to evaluate PDD examinations. Results of this study demonstrate that there are no significant differences between overall examiner and computer algorithm accuracies, suggesting that computer algorithm scoring can be as accurate as human scoring. It should be noted that the reported comparison was made using data collected following a laboratory mock-crime while the algorithm was designed using the results of actual criminal examinations. This may have reduced overall accuracy rates if there are intrinsic differences between data collected following actual and mock crimes.