An invariant for timing of saccades during visual search

The variable latency of a saccade to the onset of a single target reveals our brain's hypothesis testing about the target's presence. Search in complex scenes involves multiple objects that compete to become fixated. The initiation of a saccade in this case involves two hypotheses: (1) a potential target is present outside the fovea and (2) the currently fixated object is not the target. Previous models suggest that these hypotheses are evaluated independently, each involving a decision signal that races towards threshold. We show here that the skewed latency distributions during search comply with strong competition between these decision signals rather than independence. Moreover, the thresholds for the two competing processes are not independent either but conform to an invariant that suggests that saccades in complex scenes are made when the odds for the target's presence outside the fovea versus within the fovea are about four.