Due Process in Dual Process: Model‐Recovery Simulations of Decision‐Bound Strategy Analysis in Category Learning

Abstract

Behavioral evidence for the COVIS dual-process model of category learning has been widely reported in over a hundred publications (Ashby & Valentin, ). It is generally accepted that the validity of such evidence depends on the accurate identification of individual participants' categorization strategies, a task that usually falls to Decision Bound analysis (Maddox & Ashby, ). Here, we examine the accuracy of this analysis in a series of model-recovery simulations. In Simulation 1, over a third of simulated participants using an Explicit (conjunctive) strategy were misidentified as using a Procedural strategy. In Simulation 2, nearly all simulated participants using a Procedural strategy were misidentified as using an Explicit strategy. In Simulation 3, we re-examined a recently reported COVIS-supporting dissociation (Smith et al., ) and found that these misidentification errors permit an alternative, single-process, explanation of the results. Implications for due process in the future evaluation of dual-process theories, including recommendations for future practice, are discussed.

A dimensional summation account of polymorphous category learning

“…Ashby et al, 1998;Kemler Nelson, 1984;Smith & Shapiro, 1989;Ward, 1983), it is fully consistent with a substantial body of more recent work, across several different procedures. For example, it is consistent with results from the match-to-standards procedure Milton & Wills, 2004;Milton et al, 2009;Wills et al, 2013), the triad procedure (Wills et al, 2015), the criterial-attribute procedure (Wills et al, 2015), and information-integration category learning procedure (Carpenter et al, 2016;Edmunds et al, 2015Edmunds et al, , 2018Edmunds et al, , 2019Newell et al, 2013). It finds support from not only human behavioral data, but also from comparative work with rats and pigeons (Lea et al, 2018(Lea et al, , 2006Wills et al, 2009) and from functional imaging data in humans (Carpenter et al, 2016;Milton et al, 2009).…”

Absence of cross-modality analogical transfer in perceptual categorization

“…Ashby, Queller, & Berretty, 1999;Milton, Longmore, & Wills, 2008;Wills, Inkster, & Milton, 2015). Further, using a unidimensional rule in II tasks results in performance around 75%, well above chance (Edmunds, Milton, & Wills, 2018). To test this possibility we added control conditions (one for each category task), where participants had to complete the final transfer test without prior categorization training.…”

“…Note that in previous experiments using these types of category structures, researchers have used a model-based decision-bound analysis informed by signal detection theory (Ashby & Gott, 1988;. However, recent work has shown that this analysis can be unreliable and consequently hard to interpret (Donkin, Newell, Kalish, Dunn, & Nosofsky, 2015;Edmunds et al, 2018). Therefore, we have not included this analysis in the main text, although it is available in the Appendix for those interested.…”

Dissociable learning processes, associative theory, and testimonial reviews: A comment on Smith and Church (2018)

“…One might see such a claim as supported by the fact that the strategies people verbalize in this task are poorly related to the strategies indicated by standard decision-bound computational modeling of their categorization responses (Edmunds et al, 2015(Edmunds et al, , 2019. However, this decision-bound modeling is flawed (Edmunds et al, 2018). The flaw is sufficiently fundamental that the results reported in Smith et al (2014b)-a key paper in Smith and Church's reviewcan be modeled under the assumption that all participants in the information-integration task use easily verbalizable multidimensional rules (Edmunds et al, 2018), see also Donkin et al (2015) for a related conclusion.…”

“…However, this decision-bound modeling is flawed (Edmunds et al, 2018). The flaw is sufficiently fundamental that the results reported in Smith et al (2014b)-a key paper in Smith and Church's reviewcan be modeled under the assumption that all participants in the information-integration task use easily verbalizable multidimensional rules (Edmunds et al, 2018), see also Donkin et al (2015) for a related conclusion.…”