Vision for the blind: visual psychophysics and blinded inference for decision models

Area of Science:

• Cognitive Psychology
• Computational Neuroscience
• Psychophysics

Background:

• Evidence accumulation models, such as the diffusion model, are vital for dissecting sensory and decisional contributions to decision-making speed and accuracy.
• Key parameters like drift rates, decision criteria, and nondecision times offer insights into evidence quality, decision bias, and processing duration.

Purpose of the Study:

• To investigate the limitations of standard diffusion models in the random dot motion (RDM) task, specifically the failure of selective influence.
• To propose and validate a continuous-onset diffusion model that better accounts for the RDM task's perceptual characteristics.

Main Methods:

• Reanalyzed data from Dutilh et al. (2019) using both abrupt and continuous-onset drift and diffusion models.
• Compared model fits and the psychometric validity of parameter estimates.

Main Results:

• The continuous-onset model provided a superior fit to the RDM task data compared to the abrupt-onset model.
• Parameter estimates from the continuous-onset model were more meaningful and aligned with RDM task psychophysics.
• The standard diffusion model exhibited a failure of selective influence, impacting nondecision time estimates.

Conclusions:

• The RDM task's unique perceptual properties necessitate models with continuous-onset drift and diffusion rates for accurate parameter estimation.
• Future selective influence studies must consider the visual properties of evidence to ensure psychometric validity.