DISENTANGLING MODELS OF EVIDENCE INTEGRATION

Decision making has been studied using a variety of experimental paradigms. Here, we focus on dynamically changing noisy perceptual stimuli that require the accumulation of evidence over time. Such decisions have often and successfully been accounted for by models implementing a psychophysically inspired sequential sampling framework. This framework represents a mechanistic approach to Signal Detection Theory where observers accumulate multiple samples of perceptual evidence to a predefined decision criterion. However, a multitude of such models exist which, despite their profound structural differences, all fit existing empirical data well. We propose an approach for comparing models which is based on isolating a specific model attribute to produce qualitative, rather than quantitative, predictions via computational simulations. Simulations demonstrate that, some models (mainly but not exclusively independent ones) speed up (due to statistical facilitation) while others slow down. Our results provide strong support for the presence of high level competition and against independent models of decision making.