Direct and Indirect Measures of Lie Detection Tell the Same Story

Deception-detection experiments typically find that people are poor lie detectors. Bond and DePaulo (2006) reported that in their meta-analysis, the across-study average level of detection-deception accuracy was 54% (50% reflects baseline chance). Although the nil null hypothesis of chance accuracy could be rejected with confidence (p < .0001), 54% accuracy is unimpressive on its face and is understood as evidence of human fallibility in lie detection.

From time to time, scholars assert that although people perform poorly when directly asked to make truth/lie decisions, they perform better when accuracy is assessed with indirect, less-conscious measures. Although the argument goes back at least to DePaulo, Charlton, Cooper, Lindsay, and Muhlenbruck (1997) and Anderson, DePaulo, Ansfield, Tickle, and Green (1999), ten Brinke, Stimson, and Carney (2014) are the latest to advance the implicit-lie-detection argument in the pages of Psychological Science:

Across two experiments, indirect measures of accuracy in deception detection were superior to traditional, direct measures. These results provide strong evidence for the idea that although humans cannot consciously discriminate liars from truth tellers, they do have a sense, on some less-conscious level, of when someone is lying. (p. 1103)

In ten Brinke et al., the authors reported two experiments in which direct and indirect measures were compared. The results were straightforward. Direct measures produced accuracy rates at or below chance (46.83% and 49.62%). Indirect measures, in contrast, revealed that responses to truths and lies were significantly different (d = 0.32 and 0.27 in Experiments 1 and 2, respectively), a pattern consistent with accurate truth/lie discrimination.

Here is the catch: The 54% accuracy obtained in Bond and DePaulo’s (2006) meta-analysis is also significantly better than chance, and the average effect size for correct truth/lie discrimination with dichotomous direct measures is 0.42. The evidence that ten Brinke et al. reported in support of their claim for the superiority of indirect measures involves estimates of effect size that are below average compared with those obtained over the years using direct measures. The observed superiority of performance on their indirect measures, then, was not a sign of impressive accuracy; rather, it was a sign of atypically inaccurate performance on their direct measures. Simply put, when the control condition is a statistical outlier, even an otherwise unimpressive treatment can look good by comparison.

If effect sizes of 0.27 and 0.32 really provide “strong evidence” for less-conscious lie detection, then the meta-analytic average effect size of 0.42 associated with 54% accuracy provides even stronger evidence that people can consciously discriminate liars from truth tellers. Alternatively, if the meta-analytic average of 54% is dismissed as poor performance, then the evidence provided by ten Brinke et al. for less-conscious detection reflects even poorer accuracy.

We recently tested the indirect-lie-detection hypothesis as well, but instead of collecting new data, we used meta-analysis to approach the issue (Bond, Levine, & Hartwig, in press). Using data from DePaulo et al. (2003), we identified 130 different estimates spanning 24 different indirect measures. With the average effect of each indirect measure as the unit, the median effect size for indirect deception detection was 0.23 (ds were 0.07 for the bottom of the first quartile and 0.35 for the top of the third quartile). We concluded that accuracy is higher with direct measures than with indirect measures.

The estimated effect sizes for indirect deception detection provided by ten Brinke et al. are plausible and well within the range we would expect, given what prior research has found. People are better than chance at distinguishing truths and lies with indirect measures. However, the estimates from ten Brinke et al. do not support the conclusion that accuracy measured indirectly is better than accuracy measured directly.

The methodological problems associated with reliance on the null-hypothesis test are a recurring theme in Psychological Science, and we believe these problems are, in part, what led ten Brinke et al. astray. They cited Bond and DePaulo’s (2006) meta-analysis but noted the percentage correct (54%) without noting the corresponding effect size (d = 0.42). For their own data, they noted significant differences, and they reported effect sizes, but they did not put the effect sizes in the context of the literature. Instead, they were content to discredit the statistical null hypothesis.

To explain why indirect, less-conscious deception detection would be better than direct deception detection, ten Brinke et al. used an evolutionary argument. The argument posits that because deception evolved, deception detection had to evolve as a countermeasure. An alternative explanation is offered by truth-default theory (Levine, 2014), according to which the gains from believing other people (e.g., cooperation, efficient communication) vastly outweigh the costs of occasional deception, and deception prevention in the form of religion and other cultural prohibitions is more efficient than constant active monitoring for deception.

A wave of replicated findings since Bond and DePaulo’s (2006) meta-analysis has demonstrated that deception-detection accuracy can be 70% or greater when deception detection is assessed with direct measures (Blair, Levine, & Shaw, 2010; Hartwig, Granhag, Stromwall, & Kronkvist, 2006; Levine, Blair, & Clare, 2014). Improved accuracy is possible, but the path is not through the unconscious; instead, it is through the conscious processes of effective questioning and careful listening. Outside the lab, most lies are detected either by comparing what is said to some evidence or by confessions of the truth (Park, Levine, McCornack, Morrison, & Ferrerra, 2002).

In conclusion, ten Brinke et al. claimed that people cannot consciously distinguish between truths and lies but nevertheless sense deception. This claim, however, is inconsistent with results from a meta-analysis and is not scientifically defensible. The meta-analysis suggests that accuracy is actually a little higher with direct than with indirect measures, and recent findings suggest other paths to deception-detection accuracy.