Table 2.

Quality of the generative models fitted to behavioral data given as the mean difference (d) in criterion values (AIC and BIC) across participants^a

Model	Criterion	RW#1	RW#2	PH#1
RW#2	dAIC	−6.5
	dBIC	−3.3
PH#1	dAIC	−11.7	−5.1
	dBIC	−8.4	−5.1
PH#2	dAIC	−15.4	−8.9	−3.8
	dBIC	−9.0	−5.7	−0.6

^aRW#1, Rescorla-Wagner model with one learning rate fitted across positive and negative PEs; RW#2, Rescorla-Wagner model with separate learning rates for positive and negative PEs; PH#1, Pearce-Hall model with a fitted initial learning rate and a parameter guiding the trialwise decay in learning rate; PH#2, Adaptive Pearce-Hall model with a fitted initial learning rate and a parameter guiding the trialwise decay in learning rate; AIC, Akaike information criteria; BIC, Bayesian information criteria. Here, PEs are scaled relative to reward variability. Models are fitted across all trials, conditions, and participants.