Abstract
Prior expectations and contextual information influence perception. These expectations can develop rapidly. Even a single decision to categorize a stimulus can bias a subsequent estimate in the direction of the chosen category. To determine the effect of sensory uncertainty on estimation bias, we examined the mean estimates of orientation categories following a sequence of categorization decisions. On each trial, an ellipse was drawn from one of two partially overlapping categories of orientation (category variance was constant across conditions). Observers reported the category of the ellipse and received feedback. After each block (N = 200), observers estimated the mean orientation of each category and placed confidence intervals around their estimates. To incentivize accurate placement, points were awarded for capturing the true mean and for small confidence intervals. No immediate feedback was provided for category-mean estimates. Task difficulty was manipulated across three sessions by changing the ellipse aspect ratio. Within a session, the category boundary varied across blocks and the distance between category means was set based on a separate threshold-measurement session to yield constant d'. Observers' estimates of category means were biased away from the category boundary. These repulsive effects were greater at cardinal orientations, where sensory thresholds of orientation discrimination are lowest. The difference in bias across orientations was attenuated at the highest level of task difficulty, where sensory thresholds of orientation discrimination are highest. This pattern was predicted by a parameter-free model that ignores stimuli that cross the category boundary (i.e., stimuli that would be categorized incorrectly) and estimates category variance in JNDs. Human judgments are self-consistent. That is, only stimuli consistent with one's categorization decisions are used for estimation. Furthermore, perception of variance is in units of sensory threshold. Our results suggest a fundamental relationship between one's discrimination threshold and estimation bias.
Meeting abstract presented at VSS 2018