Random walk models of binary choice: the effect of deadlines in the presence of assymetric payoffs

Random walk models of binary choice were evaluated using line length discrimination tasks of varying difficulty. Subjects' gained monetary payoffs when their responses were correct and lost when their responses were incorrect, or exceeded a prescribed deadline. The payoff structure heavily favoured one response. Mean latency for each stimulus was a linear function of information acquired for that stimulus, with the same residual motor time for favoured and unfavoured stimuli; supporting a random walk model in discrete time. For all levels of difficulty, responses given in error were never slower than the same responses given correctly. The difference between correct and error mean response times were independent of distance to the random walk boundaries; supporting a modified form of the sequential probability ratio test model and contradicting the predictions of relative judgment theory. Rate of gain of stimulus information per unit time was substantially larger for the unfavoured stimulus in the hard conditions and one easy condition, also contradicting the prediction of relative judgment theory. Subjects' placement of response criteria was neither optimal nor completely predicted by a linear learning model, suggesting a mixed decision strategy.