Perceptual brightness decisions do not use a difference model
33.412, Sunday, May 12, 8:30 am - 12:30 pm, Orchid Ballroom
Dorion Liston1,2, Leland Stone1; 1NASA Ames Research Center, Moffett Field, CA, 2San Jose State University, San Jose, CA
Many current models of perceptual decision-making utilize a single decision variable encoding the difference between two alternate signals to account for behavioral data, often implemented as a log likelihood ratio. Alternatively, such data can be modeled using a "race" mechanism in which two separate decision variables, encoding each of the alternatives, race against one another in parallel. Consider a generic magnitude discrimination between two signals, a and b. In the noise-free case, these two models predict identical sets of choices (i.e., a-b>0 and a>b are canonically equivalent), but qualitatively distinct relationships between perceived stimulus magnitude and the respective signal strengths of the selected and unselected stimuli. In particular, a difference model predicts strong mirror image correlations of equal magnitude and opposite sign between the selected and unselected signal strengths (because of the equal and opposite weighting in a simple difference). On the other hand, the race model predicts correlations of unequal strength (and both typically positive) between perceived stimulus magnitude and the respective signal strengths of the selected and unselected stimuli (because a race produces positive competition between the two signals). Methods. We used a two-stage task in which observers (n=5) made saccadic choices in a 2AFC spatial brightness discrimination task and then reported the perceived brightness of the selected stimulus using method of adjustment. We compared the reported brightness value to the absolute signal strength of each alternative stimulus. Results. We observed positive correlations between perceived brightness and both the selected and unselected stimulus strength for all observers (bootstrap test, p<0.01). However, we simulated the difference mechanism with three commonly-used sources of internal noise (Brownian, Carpenter, and baseline-to-threshold), and were unable to break the equal and opposite correlation predictions. Conclusion. We conclude that no difference (or log likelihood ratio) model can account for the observed patterns of perceptual performance.