Understanding Verbal Confidence Statements

The aim of this dissertation is to use machine learning classifiers to examine whether verbal statements contain diagnostic value, and if so, why. Chapter 1 explores whether individual differences moderate the relationship between classifier predictions and eyewitness identification accuracy. We find that stronger face recognizers use language that better discriminates between correct and incorrect lineup identifications than weaker face recognizers. Chapter 2 examines whether verbal classifiers outperform human evaluators (e.g., jurors) in determining the likely accuracy of eyewitness statements collected under optimal and/or suboptimal lineup administration procedures. Interestingly, classifiers show very similar (or improved) diagnostic performance compared to human evaluators, but are susceptible to overestimating accuracy when eyewitnesses are given post-identification feedback. Finally, Appendix A details a study which tests whether Dobbins and colleagues’ recollection account (Dobbins & Kantner, 2019; Selmeczy & Dobbins, 2014) explains the source of the diagnostic information used by verbal classifiers (Dobbins, 2022; Dobbins & Kantner, 2019; Selmeczy & Dobbins, 2014). The results of this study are difficult to reconcile with the recollection sensitivity account, but we consider such conclusions preliminary due to deviations from our preregistered analyses.