Abstract
A meta-analysis was conducted to determine the overall effectiveness of deception detection training and to identify conditions that may moderate training effectiveness. The analysis was based on a total of 16 studies with 30 separate hypothesis tests, representing the behavior of 2847 trainees. Results indicated that the effect of deception detection training on detection accuracy was positive, significant, and of medium magnitude. Moreover, training effectiveness was moderated by the type of training implemented, training content, trainee expertise, and the type of lie told. Discussion centers on implications for training design and implementation.
Acknowledgement
I acknowledge the contributions of Susan E. Brandon, Department of Defense, and Ruth Willis, Naval Research Laboratory.
Notes
1. Note that the correspondence between the cues described in DePaulo et al. (2003) and the cues described in the studies in this database was not exact. For example, the DePaulo et al. analysis contains the cue ‘fidgeting’, as well as the cues, ‘self-fidgeting’ and ‘facial fidgeting.’ For our purposes, if a training study included a cue related to fidgeting (e.g. self-adaptors, grooming gestures), we simply coded it as fidgeting.
2. For most of the moderators examined (e.g. motivation of the deceiver, trainee expertise, length of training), items were directly coded from the written reports. Other moderators, such as the type of training implemented, required some judgment on the basis of the two raters. For these moderators, initial disagreements occurred on a small number or cases (less than approximately 10%) and these disagreements were resolved through discussion to achieve complete agreement.
3. The assumption that each of the 30 hypothesis tests represented an independent observation is false. However, it can be seen that such an assumption does not seem to render an inflated summary of this research domain. Consider the results of a supplementary meta-analysis of wholly independent effects, in which multiple hypothesis tests obtained from a single study were combined into a single test. This solution precludes the examination of the effects of moderators such as type of training or training content, but it does eliminate the problem of non-independence. This produced 16 distinct, independent hypothesis tests, one from each includable study. The results of this supplemental meta-analysis revealed the same pattern as the general effects reported, with an overall positive effect of deception detection training on judgment accuracy, d + = 0.54 and 95% confidence interval ranging from 0.45 to 0.64.