Magnitude estimation reveals Poisson-like noise underlying perception

Understanding how physical stimuli map onto perceptual experience is the founding goal of psychophysics. Insights into this process have primarily relied on measurements of sensitivity, how finely people tell stimuli apart, but these cannot identify the internal noise distribution underlying perception because they cannot separate it from signal amplification. Here, using visual contrast, we show that the trial-to-trial variability of subjective ratings derived from magnitude estimation, where people rate stimulus intensity with numbers, provides access to this noise. We validated this estimate by demonstrating that the analysis of this variability alongside mean responses reveals an internal representation comprising a sigmoidal response function and Poisson-like noise that quantitatively predicts sensitivity without free parameters. These predictions include the classic phenomena of the pedestal effect and Weber's law, which we can now trace back to the response nonlinearities and signal-dependent noise. These findings establish subjective rating variability as a behavioral measure of internal noise.