Bayesian Analyses of Cognitive Architecture

The question of cognitive architecture-how cognitive processes are temporally organized-has arisen in many areas of psychology. This question has proved difficult to answer, with many proposed solutions turning out to be spurious. Systems factorial technology (Townsend & Nozawa, 1995) provided the first rigorous empirical and analytical method of identifying cognitive architecture, using the survivor interaction contrast (SIC) to determine when people are using multiple sources of information in parallel or in series. Although the SIC is based on rigorous nonparametric mathematical modeling of response time distributions, for many years inference about cognitive architecture has relied solely on visual assessment. Houpt and Townsend (2012) recently introduced null hypothesis significance tests, and here we develop both parametric and nonparametric (encompassing prior) Bayesian inference. We show that the Bayesian approaches can have considerable advantages. Translational Abstract Nearly every judgment we make is based on multiple sources of information. In this article, we are interested in extending the statistical tools available for examining the core properties of how multiple sources of information are used together. Two of those core properties, whether information sources are processed in series or in parallel and how many sources of information are used before making a decision, can be assessed using a contrast of response time distributions. Whereas there are some statistical tests available, they are limited to null-hypothesis testing inferences. In this article we present two Bayesian analyses that can indicate whether information is used in series or parallel and whether a single or multiple sources are used for a judgment. Based on data from a simple perceptual task and simulated data, we demonstrate that these Bayesian approaches are effective and have many advantages over the existing statistical tests.