Fig. 3: Model Comparison and nkRL simulations.
From: Distinct motivations to seek out information in healthy individuals and problem gamblers
BIC comparison of the 5 RL models in HCs (a) and PGs (b). The comparative fit is based on the sum of individual BIC computed by fitting each model to participants’ free choices. In both groups, the novelty-knowledge RL model (nkRL, in green) better explains participants’ behavior compared to a leaky novelty-knowledge RL model (leaky nkRL), a knowledge RL model (kRL), a standard RL model (sRL), and a gamma novelty-knowledge RL model (gnkRL). By using the estimated individual parameters, simulations of nkRL in the first free choice trial reproduced the empirically observed decrease in novelty-seeking in PGs (Unequal Information condition, (c)), comparable choice behavior when choices are equally informative (Equal Information condition, (d)), an increase of preference for partially informative options (general information-seeking, (e)). f nkRL correctly predicts the novelty-familiarity shift in the healthy sample, (g) and its absence in the PG group. Error bars: sem (color figure online).
