Abstract
What is the nature of the internal representation at the decision stage? Most current theories assume that the decision-stage representation consists of a full probability distribution over possible choices (full probability model). Alternatively, it is possible that the decision-stage representation only includes information on the most likely choice (simple model). Here we report on the first attempt to empirically distinguish between these two possibilities. On each trial, we presented 49 circles displayed in four different colors. In the 4-choice condition, subjects (N = 32) indicated the dominant color (applied to 16 circles; non-dominant colors were applied to 11 circles each) among all possible colors. Critically, we also included a 2-choice condition, in which, after the offset of the stimulus, only two response options were presented – the dominant and a randomly selected non-dominant color. Based on the performance on the 4-choice condition (mean accuracy = 69.2%), the full probability model predicted significantly higher performance on the 2-choice task (predicted accuracy = 83.8%) compared to the simple model (predicted accuracy = 79.7%). The actual performance on the 2-choice condition (mean accuracy = 78%) was much closer to the prediction of the simple model. Formal model comparisons favored the simple model in 28 of the 32 subjects (total AIC difference between the two models = 260). Experiment 2 replicated these results with a different stimulus set (we used symbols, such as % and $, instead of colors) and a higher number of choices (6 vs. 4). Experiment 3 further replicated the main findings but with a different task, in which subjects always made two choices in a row. The results of the three experiments demonstrate that the internal decision-stage representation does not include a full probability distribution over possible choices; instead, it consists of information related exclusively to the most likely choice.
Meeting abstract presented at VSS 2018