Letting go of the Grail: Falsifying the theory of ‘true’ eyewitness identifications

Various formal and informal models of eyewitness memory have been proposed. While serving to guide both the construct and analytical frameworks of research within the field, these models have yet to be critically tested through a process of empirical falsification. This study addresses this gap by critically testing four hypotheses: the hypotheses that eyewitness memory possesses both (1) random-scale and (2) monotonic-likelihood representation; the hypothesis that eyewitness memory data is (3) accurately predicted by high-threshold (HT) models; and the hypothesis that a mathematical model of eyewitness identification provides a (4) good representation of the psychological constructs of eyewitness memory and decision making. After investigating the Block-Marschak inequalities test for random-scale and monotonic-likelihood representation and developing a new critical test for the falsification of the high threshold (HT) models, two experiments were conducted online with a total of 5,056 participants recruited from Amazon Mechanical Turk. Experiment 1 collected k-AFC probabilities for lineup sizes k ∈ {2,…,7}. Experiment 2 collected identification and ranking probabilities from a simultaneous 8-item lineup using a 3 (strong, weak, very weak memory) x 2 (low vs high expectation) x 2 (target-present vs target-absent) between-subject experimental design. Eyewitness identification outcomes were shown to have both random-scale and monotonic likelihood representation, thus allowing for development of a mathematical model. The 2HT models of eyewitness memory were falsified and superseded by an alternative surviving model—signal detection theory (SDT). Finally, the predictive ability of the unequal-variance (UV) SDT model of simultaneous lineup identification (assuming a MAX decision rule) was confirmed, as was the independence of the model’s parameters and its generalizability across task structures. It was concluded that the UV-SDT class of models provide an evidence-based account of eyewitness identification behavior, support the measurement of empirical eyewitness identification data, and have facilitated a shift towards the building of stronger scientific evidence.