Approach-avoidance decisions

Participants’ decisions within the approach-avoidance paradigm represent a binary outcome (i.e., either a CS−/low reward or a CS+/high reward choice), thus generalized linear mixed models (GLMMs) were particularly suitable for analysis. Accordingly, GLMMs were calculated using R³⁵ as well as the packages lme4 and afex. GLMMs were fit by maximum likelihood (Laplace Approximation) with binomial error distribution and the logit link function, which accounts for the binary nature of the data. Continuous predictors were centered (M = 0) and scaled (SD = 1) prior to analysis and correlations among random terms were disabled. Likelihood ratio tests were applied to obtain p-values for all fixed effects. Follow-up analyses for GLMMs were calculated using the R package emmeans.

First, decisions of all participants were analyzed using a GLMM. Fixed effects comprised the continuous predictors Reward Magnitude of the CS+/high reward option and CS+ Probability, the categorical predictor Group as well as all two-way interactions. In addition, a by participant random intercept as well as by participant random slopes for Reward Magnitude and CS+ Probability were included. A more complex model including the three-way interaction of all predictors yielded the same significant main effects and two-way interactions, however the three-way interaction was non-significant and thus not included in the final model.

Next, to explore whether decisions of the aversive learning group changed across the approach-avoidance paradigm due to fear extinction processes, we ran the same GLMM with the additional continuous predictor Trial Repetition for the aversive learning group only. This predictor counted the number of times a specific trial (i.e., a combination of reward magnitude and CS+ probability) had been presented during the task (1st to 6th repetition). As before, fixed effects included all possible two-way interactions. Further, a by participant random slope for Trial Repetition was introduced. A more complex model including the three-way interaction of all predictors did not converge successfully but yielded the same significant main effects and two-way interactions.