Measuring Efficient Lie Detection: Classification Accuracy in Relation to Truth-Lie Effects

Researchers regularly rely on within-test “truth-lie” effects (baseline truth responses vs. responses that are to be evaluated as truthful or deceptive) to compare and evaluate the efficiency of different deception detection methods, designs, and paradigms. However, they provide no clarification regarding how this effect relates to diagnostic efficiency (such as the rates of correctly detected liars and truthtellers), which is the single crucial information from a practical perspective. In the present paper, a series of simulated demonstrations proves and highlights that the relation between truth-lie effect and diagnostic efficiency can be to a great degree haphazard, and generally cannot be relied on. Additional demonstrations clarify how between-condition effects (liar vs. truthteller predictors), areas under the curves, and correct detection rates relate to each other – in particular, it is shown that each of these three measures may vary entirely independently from the rest. The main conclusion is that whenever a diagnostic method is studied, transparent diagnostic measures should be reported for the key outcomes. If diagnostic measures cannot be calculated or inferred, this limitation should be stated explicitly.