Meta-analyses on the detection of deception with linguistic and verbal content cues

Abstract

This dissertation reports two meta-analyses on verbal cues to deception. Whereas the first synthesis focuses on the validity of linguistic cues to deception, the second article focuses on the inter-rater reliability of verbal content cues. In general, the validity deals with the question if and to what extent a certain indicator of deception distinguishes truthful from deceptive statements. On the other side, the inter-rater reliability describes the amount of agreement that can be reached from several evaluators when rating specific verbal content cues.More specifically, the first meta-analysis investigates the validity of linguistic cues to deception that are assessed with computer programs. From 44 studies meeting the inclusion criteria, operational definitions for 79 linguistic cues were identified and allocated to six broader research questions. As predicted, meta-analyses showed that relative to truth-tellers, liars experienced greater cognitive load, expressed more negative emotions, and distanced themselves more from events. On the other side, liars expressed fewer sensory-perceptual words, and referred less often to cognitive processes. However, compared to liars, truth-tellers slightly used more terms related to uncertainty. Most main effects were moderated by several important independent variables such as event type, personal involvement, emotional valence, intensity of interaction, motivation, production mode, type of computer program and publication status. Although the average effect size was small, theoretical predictions were partially supported indicating that (a) liars and truth-tellers seem to use different words in a specific context and (b) computer programs can be designed to count some of these linguistic differences. However, at this point, computer programs are far from being applied in real life deception detection contexts. These findings not only further our knowledge about the usefulness of linguistic cues to detect deception with computers in applied settings but also elucidate the relationship between language and deception.The second meta-analysis examines the inter-rater reliability of a different kind of verbal content criteria, the so-called Criteria-based Content Analysis (CBCA). CBCA consists of 19 credibility criteria and constitutes an important component of Statement Validity Assessment (SVA). SVA is a forensic assessment procedure used in many countries to evaluate whether statements (e.g., of sexual abuse) are based on experienced or fabricated events. Furthermore, these criteria have frequently been adapted for research on the detection of deception as a credibility assessment tool . A total of 82 hypothesis tests from 52 English and 22 German studies were included and revealed high inter-rater reliabilities for most CBCA criteria as measured with several reliability indices. Due to large heterogeneity, moderator analyses and meta-regression were conducted on Pearson s r. Significant findings occurred for research paradigm, intensity of rater training, type of rating scale used, and the frequency of occurrence of CBCA criteria (base rates) for some criteria. Implications for future research and forensic practice are discussed.In summary, these meta-analyses suggest that human language is probably the most promising source to differentiate liars from truth-tellers. Moreover, these results show that several linguistic and verbal content cues fulfilled psychometric quality standards like validity and inter-rater reliability to some extent and under specific conditions. Taken several limitations into account, implications for research and practice are discussed.