Eyes don't lie: Eye tracking reveals whether an eyewitness saw the crime.

Past research has suggested that eye movements can be used to uncover perpetrators of a crime to some extent (around 65 % accuracy). We extended this work to examine whether similar or better results could be obtained for eyewitnesses by employing a data-driven eye-tracking approach. We expect that participants who saw the crime before: (1) look more at where the crime happened, (2) differ in the frame-by-frame viewing location, and (3) differ in the frame-by-frame variability in viewing location, compared to non-exposed participants when viewing the now-empty crime scene. Machine learning was used to classify the eye movements of exposed participants (who had seen the knife crime the day before, n = 34) and non-exposed participants (who had not seen the crime before, n = 25) while both groups viewed a video of the now-empty crime scene. Eye-tracking showed that participants who saw the crime previously were more consistent in their viewing patterns and looked more at the perpetrator regions when viewing the same, but empty, crime scene. Fixated regions predicted group membership with moderate accuracy (AUC = 0.758), but the consistency in viewing patterns led to very good classification of observers into exposed and non-exposed participants (AUC = 0.898), although some group differences remained while and after viewing the crime. These results suggest that eye movement patterns can be primed by previous observations, persisting after two days. While currently theoretical, these results may be developed as an implicit measure for detecting previous crime scene exposure through visual attention patterns.