Hierarchical Nonlinear Models in Cognition and Perception
Jeffrey N. Rouder · IMBS Colloquium · 2006
Simple but nonlinear models are relatively common in modern perception and cognition. Popular examples include the theory of signal detection, the process dissociation model, and the three-parameter power and exponential models of skill acquisition. In typical applications, unintended variability is introduced by the selection of participants and/or items. I show how this real-world constraint results in asymptotically biased estimation which, in turn, compromises inference. My colleagues and I have been advocating hierarchical Bayesian models as a principled means of analyzing nonlinear models in real-world contexts. In this talk, I demonstrate the problems of conventional analyses and the potentially dramatic benefits of hierarchical models in three domains: dissociation of memory processes, subliminal/near-liminal priming, and lexical access