Abstract
The power-rate diffusion model (PRDM; Palmer, Huk, & Shadlen, J. Vis. 5(5): 1, 2005) relates both accuracy and mean response times (RTs) to stimulus strength in perceptual judgment tasks. The model assumes that perceptual evidence accumulates through a noisy diffusion process and that the average rate of accumulation of perceptual evidence scales with the stimulus intensity according to Stevens' power law. The model utilizes the information that is contained in the distribution of RTs allowing one to quantify an observer's sensitivity independent of the observer's standing on the speed-accuracy continuum. Here, we propose the adaptive psiprdm method which optimizes estimation of the four parameters of the PRDM by selecting, on each trial, the stimulus intensity that minimizes the expected entropy in the posterior distribution defined across the model's four parameters (cf. the psi method, Kontsevich & Tyler, Vis. Res., 39, 2729-2737, 1999). After each trial, the posterior distribution is updated based on both the accuracy and RT obtained on the trial. Unlike the original PRDM, in which likelihood functions are based on mean RTs, psiprdm utilizes likelihood functions based on the RTs for individual trials. The performance of the method was tested using simulations in which responses were generated by mimicking the diffusion process. Results indicate that the estimates of all of the model's four parameters are stable and virtually bias-free after approximately thirty trials. Furthermore, estimates of the observer's sensitivity parameter were found to be stable with respect to speed-accuracy trade-off manipulations. A comparison of the psiprdm method with a method that considers only the accuracy (not the RT) of responses indicates that while the former produces a bias-free estimate of the 'half-way accuracy' threshold in the psychometric function within about twenty trials, the latter typically requires hundreds of trials to do the same.
Meeting abstract presented at VSS 2016