48) and test block (F[8, 120] = 3.831, p < 0.001, η2 = 0.20), as in the AO group. We also found an interaction of session × gamble pair (F[3, 45] = 12.15, p < 0.0001, η2 = 0.45) which was, as in Experiment

1, driven by observers lower accuracy for the 40/20 pwin pair compared to actors (t[15] = 5.89, p < 0.0001) (see Fig. S4). The between-subject effect of group, i.e. Experiment 1 versus Experiment Selleckchem Talazoparib 3, interacted only with the main effects of session (F[1, 30] = 4.39, p < 0.05, η2 = 0.13) and of gamble pair (F[3, 90] = 3.36, p < 0.05, η2 = 0.10). Therefore, the session × gamble pair interaction in choice accuracy, seen in Experiment 1, was replicated but now within the loss domain, with this effect being driven solely by observers’ impaired accuracy for the lowest value 40/20

win pair (now 60/80 loss pair). In the explicit estimates, there was a significant main effect of session (F[1, 15] = 12.86, p < 0.005, η2 = 0.46) and of gamble (F[3, 45] = 75.85, p < 0.0001, η2 = 0.84), along with a gamble × session interaction (F[3, 45] = 8.87, p < 0.0005, η2 = 0.37). Therefore, participants’ explicit estimates of ploss for each stimulus also replicated the results of Experiment 1, supporting an over-valuing of the lowest value options (i.e. participants underestimated ploss for the 80% loss option) rather than an over-estimation of small probabilities (participants showed high estimation accuracy for options with the lower ploss) (see Fig. S5). However, in the context of this argument, it is not click here obvious why the 40% win option was not also overvalued. One possibility is that the 20% win option may be qualitatively, as well as quantitatively, of lower value since it is the only option never paired with an option of an even lower value. This might explain why we find over-valuation only for the 20% win option, but we accept that this conjecture needs to be tested directly. In Experiment 3, we also found a slight

undervaluation of 80% loss (t[15] = −2.48, p < 0.05). Observer accuracy when choosing between the 80/20 win pair also showed a trend to be lower than for actors (t[15] = 1.83, p < 0.1). The magnitude of this effect was much smaller than in the 20/40 condition and this asymmetrical effect cannot be explained solely by an error in probability assessment. However, this finding hints that both a large over-valuation for low-value options and also a smaller mis-estimation Flavopiridol (Alvocidib) of low probabilities may be at play in Experiment 3. Experiments 1 and 3 both show an over-valuation for low-value options during observational learning, an effect evident across implicit (i.e. choice preference) and explicit indices of subjective value. This difference was evident despite the observational and operant learning tasks being matched for visual information, and for monetary incentives to learn. In contrast, Experiment 2 shows that learning is generally improved between two active learning sessions despite the time delay and the novel stimuli being learned.