Advances in eye-tracking technology have promoted its widespread use to understand and improve target searches in psychology, industrial engineering, human factors, medical diagnostics, and marketing. Eye movements are the realization of a complex, unobserved spatiotemporal attention process with many sources of variation. Eye-tracking data often have been aggregated and/or summarized descriptively, because few adequate statistical models are available for their analysis. This article proposes a model that may serve to uncover the latent attention processes of people searching for targets in complex scenes. It recognizes the spatial nature of eye movements and represents two latent attention states, a localization state and an identification state, between which people may switch over time according to a Markov process. A saliency map, based on low-level perceptual features and the scene's organization, guide target searches in the localization state. In the identification state, people verify whether a selected candidate object is the target. The model is applied to analyze commercial eye-tracking data from more than 100 people engaged in a target search task on a computer-simulated retail shelf display. Rapid switching between attention states over time is revealed. Estimates of the feature and saliency maps are provided and found to be related to search performance. The results facilitate the evaluation of the effectiveness of alternative visual search strategies.

Gibbs sampling, Hidden Markov model, Hierarchical model, Slice sampling, Spatial analysis,
American Statistical Association. Journal
Erasmus Research Institute of Management

van der Lans, R.J.A, Pieters, R, & Wedel, M. (2008). Eye-movement analysis of search effectiveness. American Statistical Association. Journal, 103(482), 452–461. doi:10.1198/016214507000000437