Caricature Effects in Forensic Contexts: Impact of Altered Face Distinctiveness on Accuracy of Eyewitness Identifications

The overarching aim of the present thesis was to investigate the impact of caricature and anticaricature manipulations on accuracy of face recognition in forensic contexts, integrating behavioural experiments with computational modelling. Caricatures, which are constructed by exaggerating distinctive facial features, have previously been suggested to enhance accuracy of recognition. However, prior empirical findings have been inconsistent, often limited to familiar face recognition or based on stylised or low-quality stimulus sets. To address these issues, the present thesis investigated the impact of caricatures and anticaricatures on accuracy of recognition when faces are encoded in their veridical (i.e., unaltered) form, as would be the case when witnessing a crime or when matching photographs to live individuals (e.g., border control). Chapter 3 investigated caricature effects on unfamiliar face recognition in the context of sequential lineups, and examined their impact on the own-gender bias of face recognition. Chapter 4 combined behavioural experiments with computational modelling to extend this work across three forensically relevant paradigms involving unfamiliar face recognition (i.e., face matching, showups, and simultaneous lineups), and further examined whether caricature effects were influenced by delays between encoding and test. Finally, Chapter 5 investigated the impact of caricature and anticaricature on accuracy of familiar face recognition under varying levels of image degradation, simulating the typical quality of CCTV footage. A consistent pattern emerged across all studies: Caricature did not enhance accuracy of recognition relative to veridical faces, while anticaricature consistently impaired performance. These findings are not in line with face-space theory predictions, which suggest that caricatures should enhance recognition by increasing face distinctiveness. Instead, as all stimuli were initially encoded in their veridical form, the present findings more closely align with the principle of encoding specificity, indicating that while caricatures provide strong identity cues, recognition is optimised when the distinctiveness of stimuli at test more closely matches information present at encoding. This thesis therefore highlights the necessity of revising face-space theory to incorporate encoding specificity as a boundary condition, providing a more accurate account of both familiar and unfamiliar face recognition. Such a revision would not only reconcile the mixed empirical findings present in the caricature literature, but additionally provide clearer guidance for most applied contexts, in which faces are typically encoded in their veridical form.