Matching identical stimuli with the method of adjustment produces counter-intuitive biases

Poster Presentation: Saturday, May 18, 2024, 8:30 am – 12:30 pm, Pavilion
Session: Decision Making: Perceptual decision making 1

Jiang Mao1 (), Alan Stocker1; 1University of Pennsylvania

The method of adjustment is a widely used experimental procedure for measuring perceptual biases: subjects adjust a probe stimulus until it perceptually matches the test stimulus. Any resulting difference in feature value between probe and test is thought to reflect their relative perceptual difference (i.e. bias). Here, we challenge this view by empirically testing a fundamental prediction of this perceptual matching interpretation: if test and probe are identical then their match should be unbiased because perceptual distortions cancel out between identical stimuli. We run a psychophysical orientation matching experiment using the method of adjustment. Subjects (N=10) were presented with two orientation-filtered noise stimuli (0.5 deg radius Gaussian window, broadband spatial frequency) equidistant (4.5 deg) on either side of a central fixation mark with response-terminated display. Test and probe stimuli either had low or high noise (orientation) randomly assigned, which resulted in two same-noise and two cross-noise conditions. Test orientations were uniformly sampled from the full range (0 to 180 deg) in steps of 10 deg. Counter-intuitively, we found robust and characteristic repulsive biases away from cardinal orientations in both same-noise conditions. These biases cannot be perceptual because test and probe are identical. In addition, in the cross-noise conditions biases were not inverted when switching test and probe, which replicates a previous result. This is also incompatible with the view that the matches are purely perceptual. Finally, we demonstrate that a holistic matching model where perceptual inference and the matching process both simultaneously operate on all levels of the representational hierarchy, can accurately explains subjects’ matching data from our experiment as well as the data from multiple previous studies. Together, our results demonstrate that biases measured with the method of adjustment are not purely perceptual but rather reflect optimal matching behavior according to a holistic objective.