Supplemental Material for The Role of Perception in Generalization: Commentary on Zaman, Yu, and Verheyen (2023)\n

Re-analysis: Mixture of Response StrategiesOne take-away message of our commentary is that it is difficult to investigate the role of perception in generalization.A within-subjects design is needed, and there is always the possibility that early tasks will affect later tasks, regardless of whether they consist of perceptual or learning judgements.However, as stated, we believe that it is worthwhile attempting to dissociate the contributions of perception and the cognitive processes of stimulus-outcome learning between generalization and identification, but that this requires more adequate statistical models.Therefore, we applied our recently proposed method (Schlegelmilch, Wills, & Lee, 2023) to meaningfully disentangle the prevalence and the shape characteristics in generalization and identification gradients for quantitative comparison, to test ZYV's hypotheses more directly.In this re-analysis, we took their data at face value (i.e., ignoring the interpretational issues discussed above).These quantitative measures are outlined below.Importantly, for this, we quantified the initial color ratings via individual regressions (boundary and between color discrimination), and further used the training scores as indicators of learning success, to be correlated with the assessed gradient characteristics. Color Ratings & Training PerformanceFirst, as predictors for gradient characteristics in the following analyses, we calculated the average rate of responding + for each CS+ and CS-in training (mean of 12 trials each, roughly reflecting learning speed), yielding two measures for each individual.As a measure of color perception, we applied to each individual a simple beta-regression to their blueness and greenness ratings, similar to the method by the original authors (mean-centered stimulus values as predictor; using the betareg R package Grn, Kosmidis, & Zeileis, 2011).The measure of interest is the slope of the color rating gradients used as indicator of color discrimination later.There was no evidence favoring a correlation between the difference in training scores on CS+ and