So What? On the Meaning of Parameter Estimates From Reinforcement-Learning Models


    loading  Checking for direct PDF access through Ovid

Abstract

In their comment on the article by Steingroever, Wetzels, and Wagenmakers (2014), Konstantinidis, Speekenbrink, Stout, Ahn, and Shanks (2014) convincingly argue why a wide range of sophisticated model comparison methods is required to select a good model for the Iowa gambling task (IGT). While we agree with Konstantinidis et al. on this count, the focus of Steingroever et al. was not on model comparison. Here we clarify our initial goal, which is to illustrate why assessment of absolute model performance is necessary to avoid premature conclusions about the psychological processes that drive performance on the IGT. In addition, we elaborate on the advantages and drawbacks of both the post hoc absolute fit method and the simulation method. Finally, we highlight the distinction between statistical aspects of model adequacy and psychological relevance of parameter estimates.

    loading  Loading Related Articles