State-trace analysis misinterpreted and misapplied: Reply to Stephens, Matzke, and Hayes (2019)

摘要

After using state-trace analysis to reanalyze results from 63 different categorization studies, Stephens, Matzke, and Hayes (2019) concluded that "the evidence for two distinct category learning systems is much more limited and inconsistent" (p. 14) than Ashby and Valentin (2017) had previously claimed. This reply shows that Stephens et al. (2019) misinterpreted and misapplied state-trace analysis. They report no evidence that favors a single learning system over multiple systems. They acknowledge that they would favor a single-system account, regardless of how their re-analyses had turned out. They justify this bias by claiming that single-system theories are more parsimonious than dual-systems theories, but they use a definition of parsimony that is inconsistent with state-trace analysis, and with the entire statistical field of model selection. By any accepted definition of parsimony, the dual systems COVIS model is more parsimonious than the single-system model they favor in the current applications. The correct interpretation of their results is that none of the 63 studies they examined, by itself, definitively identifies the number of parameters that are varying across the conditions of that study. However, this was never an issue of contention, and was stated explicitly in prior publications. (C) 2019 Elsevier Inc. All rights reserved.