Reverse Correlation Reveals Averaging Strategies

Abstract

It is known that individuals can encode summary statistics about groups of elements, such as their mean orientation (e.g., Dakin, 2001). A common account of this averaging process is that people only use a subsample of elements from an array. Here, we used a reverse correlation method to examine the sampling strategy for participants who judged the average (a) head rotation of a group of faces, (b) gaze deviation of a group of faces, and (c) orientation of a group of Gabor patches. Participants (n = 6) categorized (leftwards or rightwards) the average (a) head rotation, (b) gaze deviation, and (c) orientation (clockwise or anticlockwise of vertical) of Gabor patches. For all stimuli, the head rotation (gaze deviation or Gabor orientation) of each face (Gabor) was drawn from a normal distribution with a mean of zero. A minimum of 1,200 trials were collected from each participant. Reverse correlation maps were generated by correlating, on a trial-by-trial basis, the location and value of each face (Gabor) with the participants' responses. Results for head rotation and gaze deviation show a clear sampling area in the center of the array that was significantly larger for judgments of average head rotation than gaze deviation. Variations between participants could not be attributed to differences in their sampling efficiency. Maps from the orientation task reveal smaller apparent sampling areas and lower peak correlations. These data suggest that observers do not average across all elements, but rather display clear biases for centrally localized elements. Individual differences in sampling efficiency (within a category) likely reflect differences in strategies.