Abstract
Purpose: Saccadic eye movements provide discrete responses at easily measured latencies, facilitating study of the underlying decision process. Linear rise-to-threshold models of saccadic decisions (i.e. race model, LATER) have been very successful at describing latency (1/rate) distributions, but do not account for the accuracy (% correct) of saccadic choices. We have extended the race model and incorporated predictions derived from probability theory to account for the latency and accuracy of saccadic decisions.
Methods: A macaque observer fixated a central fixation cross on a random noise background (mean 41, sd 9 cd/m^2). Two bounding boxes centered at 6 on either side of fixation demarcated possible target locations. After a random interval, two gaussian-blurred disks were added to the background, with a small luminance increment (0 .48 .97 1.45 1.94 cd/m^2) added to one disk. The monkey was rewarded for making a saccade to the brighter disk. We measured saccadic latency distributions and percent correct for each signal strength.
Results: We modeled the decision process by applying a max rule to the rates drawn from two distributions, each representing one saccade goal. The shapes of the correct and incorrect latency distributions as well as their relative proportions were well fit. As signal strength increased, the monkey's proportion correct increased and saccadic latency decreased; rates associated with correct responses increased with signal strength and rates for incorrect responses remained constant.
Conclusions: By including a max rule, our extension of the race model accounts for both the latency and percent correct of saccadic choices. The max rule incorporates the idea that the overt saccadic choice reflects only one of the multiple saccades prepared during visual discrimination. This version of the race model may be useful in determining how saccadic decisions are affected by task parameters such as signal strength and prior probability.