Fig. 2: Increased exploration in the long horizon.
From: Value-free random exploration is linked to impulsivity
a Behavioural horizon effects: in the long (versus short) horizon participants sampled less from the high-value bandit (two-sided Wilcoxon signed-rank two-tailed test: V = 157079.5, p = 1.536e-81, Wilcoxon effect size r = 0.797) and more from the novel (V = 10355, p = 1.811e-72, r = 0.750) and low-value bandit (V = 34420, p = 2.817e-24, r = 0.425). b Model parameters: in the long (versus short) horizon participants had higher value of \(\epsilon\) (i.e., value-free random exploration; V = 35367, p = 9.831e-34, r = 0.503), \(\eta\) (i.e., novelty exploration; V = 10334, p = 7.411e-75, r = 0.076) and \({\sigma }_{0}\) (their uncertainty about a bandit’s mean before seeing any samples; V = 54537, p = 1.868e-13, r = 0.306). The parameters were fitted to each participant’s first draw, and they were fitted to each horizon separately. ***p < 0.001. For detailed statistics cf. Supplementary Table 2. For details about model parameters cf. Supplementary Table 4. Data are shown as mean ± 95% CI and each dot/line represent one participant. Sample size for statistics: N = 580 human participants. Source data are provided as a Source Data file. Bar values: High-value bandit: Short Horizon (SH): 52.134, Long Horizon (LH): 42.088; Novel bandit: SH: 36.801, LH: 46.073; Low-value bandit: SH: 3.928, LH: 5.026; \(\epsilon\)-greedy parameter: SH: 0.099, LH: 0.134; Novelty bonus \(\eta\): SH: 1.919, LH: 2.884; Prior variance \({\sigma }_{0}\): SH: 1.085, LH: 1.186.
