Verbal response time and duration indices of deception in humans interviewed by a computer-generated agent

To examine the relationship between verbal response time, response duration, and deception during an interview with a computer-generated agent, we developed a model using logistic regression conducted on a training group (n=90) and cross-validated the model on an additional 127 participants who either did or did not engage in a simulated crime at a mock security checkpoint. Verbal responses during the interviews required simple “yes” or “no” utterances, which examinees were instructed to produce “promptly” but not in a speeded manner. The results showed that, overall, 75 of 127 (59.1%) participants in the cross-validation group were correctly classified (p<.05). This result was due to the ability to correctly classify nondeceptive participants (specificity), and we interpreted this finding as being due to the presence of subpopulations of deceptive participants. Group analyses also revealed that response durations to mock crime-relevant questions were shorter when participants believed that a human was controlling the interview (p<.05), and that relative changes in response durations among different types of questions during the interview were related to deceptive status. The response duration shortening observed in the present study was attributed to variations in social dominance, motivational systems, or some combination of these factors.