Abstract
Signal detection and ideal observer models have been proposed to predict decreasing visual search accuracy with increasing set-size with a fixed target known to the observer (Palmer et al., 2000, Vision Research; Eckstein et al., 2000, Perception & Psychophysics; Verghese, 2001, Neuron). Here, we extend ideal observer and suboptimal models to oddity search, where fewer modeling attempts have been made (Santhi & Reeves, 2004, Vision Research). In an oddity search, the target is not known beforehand and observers must identify and find the target as the ‘odd man out’. Methods: Three observers performed a 2IFC oddity orientation discrimination. Each interval contained N (2, 4, 8, or 16) Gabors (peak sf = 1.98 cpd, full width, half height = 0.79°) presented on an imaginary circle (eccentricity=9.74°). On each trial, a target and distractor orientation were randomly sampled from 10 possible orientations ranging from −45° to 45° in 10° steps. One interval contained N distractors, while the other contained N-1 distractors and a target; the observer judged which interval contained the target. The orientation of each element was perturbed with Gaussian noise (sd=12°). There were two conditions, a target unknown (oddity search), and another with a cue giving the target orientation before each trial (target known). Results: For the ideal observer, set-size effects are more pronounced for the target known condition than the target unknown (oddity) condition, but not for a suboptimal model that does not use information about the target identity and only makes decisions based on element differences. Across both conditions, the suboptimal model best predicted performance for two naïve observers, and the ideal observer best predicted performance for the third observer (WS, author). Thus, with this novel extension of ideal observer models to the oddity search, we were able to assess that observers may differ in the amount of information used about target identity.