A Machine Learning Investigation of Numeric and Verbal Metamemory Judgments across Basic and Applied Memory Studies 39 views

Across three experiments spanning both basic and applied memory paradigms, this dissertation examined how people evaluate and communicate the accuracy of their memories. A central question was whether traditional numeric judgments capture all of the diagnostically useful information in people’s reports or whether verbal responses provide additional information about memory accuracy. More specifically, these experiments addressed five broad questions: (1) do verbal metamemory judgments provide diagnostic value beyond traditional judgments-of-learning (JOLs) or numeric confidence ratings?; (2) are different components of verbal reports – namely verbal certainty statements and justifications – equally predictive of memory performance?; (3) does the predictive value of verbal metamemory judgments generalize from encoding-based judgments to retrieval-based eyewitness confidence judgments?; (4) how do differences in face recognition ability and lineup type (i.e., same- versus cross-race identifications) influence the predictive value of eyewitness confidence?; and (5) are well-established estimator variables in eyewitness identification, including numeric confidence, decision time, and face recognition ability, robust to the cross-race effect? Overall, the findings from these three experiments suggest that: (1) verbal metamemory judgments contain diagnostically useful information that is not captured by numeric ratings; (2) this predictive value is driven primarily by justifications rather than verbal certainty statements for JOLs in a cued recall memory task but is influenced by the interaction between justifications and verbal certainty statements in an eyewitness memory task; (3) verbal metamemory judgments afford this additional predictive value at both encoding and retrieval; (4) stronger face recognizers are better able to use numeric confidence to distinguish between correct and incorrect same- and cross-race identifications; and (5) even when identification accuracy declines for cross-race lineup decisions, estimator variables – especially confidence – retain their predictive value. Together, these studies advance our understanding of how people evaluate and communicate the reliability of their own memories and provide a framework for integrating verbal and numeric metamemory judgments across basic and applied domains of memory research.

PHD (Doctor of Philosophy)

machine learning; judgments of learning; confidence; eyewitness memory; cross-race effect

English

All rights reserved by the author (no additional license for public reuse)

2026-06-04