Abstract
A decision is a commitment to a proposition or action based on a combination of evidence, prior knowledge and the anticipated costs and benefits. Even simple psychophysical tasks draw on these elements when an observer chooses between two alternatives. Here, we address how sensory evidence is combined with prior knowledge of the probability of the alternatives. Human and nonhuman primates discriminated the direction of dynamic random dot motion. We manipulated both stimulus strength and prior probability that the motion was to the left or right. Observers viewed displays consisting of a pair of peripheral choice targets and a central patch of dynamic random dots. A fraction of the dots chosen at random moved coherently to the left or right. The task was to indicate the net direction of motion by making an eye movement to the left or right target. Motion strength was manipulated by varying the motion coherence and bias was manipulated by varying the probability of a particular direction of motion within a block of trials. We found that the effect of prior probability is equivalent to the effect of motion strength: changing the odds of rightward motion from 1:1 to 4:1 induced a change in both choice and response time equivalent to adding about 5% coherence to the right. This result is inconsistent with the traditional view that prior probability affects decisions by altering the decision criterion. In the framework of sequential sampling (e.g., random walk) models, a change in criterion predicts a pattern of choice and response time that was not seen in our experiments. Our findings are consistent with sequential sampling, but instead of affecting the decision criterion, prior probability affects the representation of the sensory information. This sensory effect resembles feature-based attention. Although the result seems to contradict a tenet of Bayesian decision making, we show that it may be a sensible strategy under some conditions.
Supported by HHMI and NIH grants RR00166 and EY11378