Table 2. Model fits. Mean ± SEM.

Model	-LLE	AIC	Pseudo-R2	α	αApproach	αAvoid	β
Working memory
Random choice	240.76±22.860	240.767±45.720	-	-	-	-	-
Canonical	182.068±23.126	186.068±46.252	0.291±0.028	0.152±0.024	-	-	0.286±0.043
Approach/avoidance	174.586±22.641	180.586±45.282	0.320±0.029	-	0.282±0.045	0.077±0.019	0.258±0.027
Habitual
Random choice	79.467±0.581	79.467±1.162	-	-	-	-	-
Canonical	55.801±2.398	59.801±4.796	0.297±0.030	0.223±0.056	-	-	0.229±0.037
Approach/avoidance	51.384±2.696	57.384±5.392	0.354±0.036	-	0.175±0.043	0.221±0.054	0.161±0.032

LLE is the log-likelihood estimate. AIC is the Akaike Information Criterion. α denotes learning rates and β the trade-off between exploration and exploitation. Working memory and Habitual refer to two different learning systems which can be assessed by fitting model parameters to behavioural data during the training and testing phase, respectively [12].