Figure 3.

The results of the simulation in the counterfactual learning context. (a–d) The results of the true model with asymmetric learning rates assuming confirmation bias (${\alpha }_c^{+}=0.5,{\alpha }_c^{-}=0.2,{\alpha }_u^{+}=0.2,{\alpha }_u^{-}=0.5$). (e–h) The results of the true model with asymmetric learning rates assuming opposite confirmation bias (${\alpha }_c^{+}=0.2,{\alpha }_c^{-}=0.5,{\alpha }_u^{+}=0.5,{\alpha }_u^{-}=0.2$). (i–l) The panel shows the results of the true model with symmetric learning rates (${\alpha }_c^{+}={\alpha }_c^{-}={\alpha }_u^{+}={\alpha }_u^{-}=0.5$) and choice perseverance (φ = 1.5). (a,e,i) The first column indicates the learning rates (${\alpha }_c^{+}$, ${\alpha }_c^{-},{\alpha }_u^{+}$, ${\alpha }_u^{-}$) in the Asymmetry model. (b,f,j) The second column indicates the learning rates in the Hybrid (gradual) model. (c,g,k) The third column indicates the degree of confirmation bias $\left(\frac{{\mathrm{\alpha }}_{\text{c}}^{+}+{\mathrm{\alpha }}_{\text{u}}^{-}}{2}-\frac{{\mathrm{\alpha }}_{\text{c}}^{-}+{\mathrm{\alpha }}_{\text{u}}^{+}}{2}\right)$. (d,h,l) The final column shows the perseverance parameter (φ) in the Perseverance (gradual) and Hybrid (gradual) models. ***p < 0.001, **p < 0.01 and *p < 0.05. The error bars represent the standard error of the mean. The diamonds denote the ground-truth value of the parameters used in the data generation.