Abstract
One important saccadic decision is how long to fixate a given location. Two views of saccadic timing have been proposed. In Evidence models, the eye remains fixated until necessary visual information is acquired. In Timer models, the duration of a fixation is governed by an internal clock. To distinguish these models, subjects viewed sequentially accumulating dots (1 dot/53 ms) sampled from a Gaussian distribution and decided whether the dots came from a mean to the left or right of a reference line. Level of difficulty was determined by the distance of the mean to the reference line. Blocks of trials were either easy, difficult, or a mix. RTs in the mixed condition fell between RTs for the blocked conditions (context effects). The value of the accumulated evidence on each trial (computed based on the locations of the presented dots) increased with RT. A timer model failed to predict the RT distributions in the mixed condition, and the observed relationship between RT and evidence. An evidence model (with noisy criteria and an optimal decision rule), captured the context effect, but did not predict the observed relationship between RT and accumulated evidence. A hybrid model used a weighted average of two independent probabilistic stopping signals, time and evidence. Both signals increased stopping probability sigmoidally. The weight parameter reflects the level of influence of evidence and time. Weights varied across subjects for easy blocks; however, all subjects gave higher weight to evidence in difficult blocks. The model captured the RT distribution, effects of context and relationship between evidence and RT. The success of the hybrid model shows the flexibility of strategies of saccadic timing, allowing emphasis on either evidence or elapsed time in attempts to produce the most efficient saccadic scanning patterns.
Meeting abstract presented at VSS 2014