Noise in the Verifiability Approach to Lie Detection

Abstract

This study examined the effectiveness of Verifiability Approach (VA) in deception detection. Additionally, we examined potential sources of judgmental noise, including subjective variability in detail counts and inconsistencies in truthfulness assessments. Two experiments were conducted. In the first, untrained participants counted verifiable and unverifiable details in alibi statements without assessing truthfulness, revealing variability in detail counts as a significant noise source. In the second, participants assessed truthfulness using four methods: personal judgment, verifiable detail-assisted judgment, algorithmic classification, and a new consensus-based judgment protocol. Algorithmic classification improved accuracy (61.00%) compared to personal and assisted judgments but resulted in a high false alarm rate (59%). The consensus-based judgment or anti-noise protocol reduced false alarms (30%) and showed promise for enhancing reliability. While the VA enables detection of deception above chance level, its effectiveness is limited by subjective judgment biases. Providing judge training and applying an anti-noise protocol may improve its reliability.