Theory selection and evaluation in case series research.

Using empirical data to develop theories requires not only evaluating how well a theory accounts for data; it requires using the data to select the best theory from among a set of alternatives. Current case series research is examined in light of these two issues. Theory selection requires that theories make contrasting predictions. In the first section of this commentary, I present novel simulation results showing that existing theories of language production do not make contrasting predictions for the overall distribution of responses over a set of responses categories (e.g., correct response, semantic error, etc.; Dell, Schwartz, Martin, Saffran, & Gagnon, 1997). Given such results, in order to be theoretically productive case series research must focus on those aspects of data that serve to contrast theoretical alternatives. The second section considers evaluation of claims regarding individual differences. Such claims are typically underconstrained. Two approaches to addressing this issue are discussed. I argue that case series research should provide independent evidence for hypothesized individual differences. Second, parametric approaches might provide a means of constraining theories of individual differences. The plausibility of this approach is examined through novel analyses of empirical distributions of individual differences in impairments to lexical access (Schwartz, Dell, Martin, Gahl, & Sobel, 2006).