Figure 5.

Simulations of candidate mechanisms. A–C illustrate three possible perceptual pathways shown in Figure 1, each representing a putative location for the modulation. The middle rows (D–I) show predicted perceptual responses for each of these models, for a simulated psychophysical choice for a favored stimulus (D–F, weighted by a factor of 2.0) or an unfavored stimulus (G–I, weighted by a factor of 0.75). In each case, the simulated observer compares the weighted sensory signal with a neutral test stimulus and chooses the “brighter” of the two. For each model, three psychometric curves were constructed from trials using the dimmest, middle, and brightest third of the stimulus luminance distribution, respectively. These three curves illustrate the effect of hypothetical weighting factors on input sensory signals and internal noise: weighting factors that scale sensory signals amplify the shift between the means of the psychometric functions (E vs H, F vs I), but the motor weighting model (D vs G) does not; weighting factors that occur downstream of internal sensory noise decrease the slopes of the psychometric curves (F vs I), but the early visual weighting (E vs H) or motor weighting (D vs G) models do not. The bottom row of J–L shows the predicted relationship between perceptual noise (i.e., SD of the cumulative Gaussian fits) and perceptual gain (i.e., shift between the means of the cumulative Gaussians for the 3 curves) for 20 uniformly distributed weight levels (0.5–2.0) for each model. For the motor model, perceptual gain and noise vary stochastically around a single fixed value (J). For the two visual weighting models, perceptual gain converges to w and perceptual noise varies around a fixed value (K); for the late visual weighting model, perceptual gain and noise covary (L).