Title
Bayesian adaptive optimal design for discriminating models of cognition
Speaker
Jay Myung, The Ohio State University
Abstract
Experimentation is fundamental to the advancement of science, whether one is interested in studying the neuronal basis of a sensory process in cognitive neuroscience or assessing the efficacy of a new drug in clinical trials. Adaptive design optimization, in which the information learned from each experiment is used to inform subsequent experiments, is a particularly attractive methodology because it can potentially reduce the time required for data collection while simultaneously increasing the informativeness of the knowledge learned in the experiment. More concretely, the problem to be solved in adaptive sequential design optimization for model discrimination is to identify an experimental design under which one can infer the underlying model, among a set of candidate models of interest, in the fewest possible steps. This problem is challenging because of the many, sometimes arbitrary, choices that must be made when designing an experiment. Nevertheless, it is generally possible to find a design that is optimal in a defined sense. In this paper, addressing the design optimization problem in discrimination of formal models of cognition, we apply a simulated-based Bayesian method that was recently introduced in statistics (Muller, Sanso & DeIorio, 2004). We use a utility function based on mutual information, and give three intuitive interpretations of the utility function in terms of Bayesian posterior estimates. Finally, we demonstrate the potential of adaptive design optimization for improving experimentation in psychology by implementing the method in experiments with simulated as well as human participants.
