How Analytic Approaches Shape Deception Detection Results: A Comparison of Raw Percentage and Signal Detection Metrics

Human deception detection can be both poor in an absolute sense and, statistically, substantially better than chance. This empirical paradox is documented and explored in the present article by comparing raw percentages and signal detection metrics in a reanalysis of fourteen prior deception detection experiments (total N = 2,349 respondents; 32,776 truth-lie judgments from 5 different countries). We show that different analytic approaches applied to the same data can yield inconsistent or mixed findings. Measures of raw percent-correct accuracy and sensitivity are nearly perfectly correlated yet are open to very different descriptive interpretations. In contrast, raw and signal detection estimates of bias diverge, indicating competing interpretations of human judgment error. We advocate for avoiding simple face-value interpretations of both approaches and embracing multiple analytic approaches simultaneously. A new R package, liaR, is presented to facilitate raw and signal detection calculations in parallel.