Table 7.

Group parameter estimates of the full reinforcement learning diffusion decision model

Parameter	M	SD	2.5% percentile	97.5% percentile
$\varphi \left({\mu }_{{\eta }^{+}}\right)$	0.07	0.02	0.03	0.12
${\sigma }_{{\eta }^{+}}$	0.75	0.15	0.50	1.09
$\varphi \left({\mu }_{{\eta }^{-}}\right)$	0.08	0.02	0.05	0.14
${\sigma }_{{\eta }^{-}}$	0.58	0.13	0.37	0.87
$\exp \left({\mu }_{v_{\text{mod}}}\right)$	0.48	0.10	0.32	0.70
${\sigma }_{v_{\text{mod}}}$	0.85	0.14	0.59	1.14
$\exp \left({\mu }_{v_{\max }}\right)$	3.47	0.25	2.98	3.98
${\sigma }_{v_{\max }}$	0.31	0.07	0.20	0.47
${\mu }_{a_{\text{fixed}}}$	1.00	0.20	0.62	1.39
${\sigma }_{a_{\text{fixed}}}$	0.97	0.14	0.73	1.26
${\mu }_{a_{\text{mod}}}$	−0.010	0.006	−0.021	0.001
${\sigma }_{a_{\text{mod}}}$	0.027	0.004	0.020	0.037
${\mu }_{T_{\mathrm{er}}}$	0.76	0.03	0.71	0.81
${\sigma }_{T_{\mathrm{er}}}$	0.13	0.02	0.10	0.17

Note. The full reinforcement model had separate learning rates η⁺ and η⁻ for positive and negative prediction errors, two parameters to describe the non-linear mapping between the difference in values and the drift rate, a scaling parameter v_mod, and an asymptote $v_{\max }$, one fixed threshold parameter a_fixed, one value-modulation parameter a_mod, and finally one non-decision time T_er. Note that ${\mu }_{{\eta }^{+}}$, ${\mu }_{{\eta }^{-}}$, ${\mu }_{v_{\text{mod}}}$, and ${\mu }_{v_{\max }}$ were transformed for interpretability