Deception detection accuracy is a predictable linear function of message veracity base-rate: A formal test of Park and Levine's probability model

This study provided the first empirical test of point predictions made by the Park-Levine probability model of deception detection accuracy. Participants viewed a series of interviews containing truthful answers, unsanctioned, high-stakes lies, or some combination of both. One randomly selected set of participants (n = 50) made judgments where the probability that each message was honest was P(H) = .50. Accuracy judgments in this condition were used to generate point predictions generated from the model and tested against the results from a second set of data (n = 413). Participants were randomly assigned to one of eight base-rate conditions where the probability that a message was honest systematically varied from 0.00 to 1.00. Consistent with the veracity effect, participants in P(H) = .50 condition were significantly more likely to judge messages as truths than as lies, and consequently truths (67%) were identified with greater accuracy than lies (34%). As predicted by the model, overall accuracy was a linear function of message veracity base-rate, the base-rate induction explained 24% of the variance in accuracy scores, and, on average, raw accuracy scores for specific conditions were predicted to within approximately +/- 2.6%. The findings show that specific deception detection accuracy scores can be precisely predicted with the Park-Levine model.