Perceptual averaging on relevant and irrelevant featural dimensions

Abstract

Here we report four experiments that explore the nature of perceptual averaging. We examine the evidence that participants recover and store a representation of the mean value of a set of perceptual features that are distributed across the optic array. The extant evidence shows that participants are particularly accurate in estimating the relevant mean value, but we ask whether this might be due to processes that reflect assessing featural similarity rather than computing an average. We set out and test detailed predictions that can be used to adjudicate between these averaging and similarity hypotheses. In each experiment, a memory display of randomly positioned bars was briefly presented followed immediately by a probe bar. Participants had to report in a Yes/No task whether the probed feature value was present. In initial experiments, we examine reports of the orientation of white bars and of the color of vertical bars. Then, in companion experiments, we examine reports of the orientation of bars whose color vary, and of the color of bars whose orientation varies. In this way, we test ideas about whether perceptual averaging occurs on a featural dimension that is irrelevant to the task. Currently, it is not known whether perceptual averaging only takes place on a task-relevant dimension or whether it operates more widely.