25, p = 0.036), and the transverse temporal region (t[21] = 2.79, p = 0.011). Accumbens favored win information, showing significant win-tie decoding (55% accuracy, on average, t[21] = 3.77, p < 0.001) but not tie-loss (51% average accuracy, t[21] = 1.17, p = 0.13). Caudal ACC also favored win-tie over tie-loss discriminations (56% versus 52% accuracy), but still showed a significant (p < 0.05, uncorrected) tendency to decode tie-loss (t[21] = 1.74, p = 0.048). Transverse temporal region showed an ability to decode tie versus loss

information (t[21] = 3.21, p = 0.002) but not win versus tie (t[21] = −0.28, p = 0.6). In a similar searchlight analysis, we contrasted the ability of each voxel to decode wins-ties and ties-losses. We found eight small clusters that differed significantly in their IWR-1 mouse ability to perform these two classifications (Table S6; figures not shown because these small clusters did not show up well when projected to the surface). Regions that did better on win-tie than tie-loss (p < 0.001, k = 10) were in the right basal ganglia (medial globus pallidus), Luminespib clinical trial the left ACC, and left middle frontal gyrus. Regions performing better

on ties-losses were the left amygdala and regions in the right IPL, left medial temporal, left fusiform and left middle temporal gyrus. In total, clusters showing these differences only encompassed 136 voxels, far fewer than those with significant three-way win-tie-loss classification (equal to only 0.4% of the number of voxels able to decode win-tie-loss). Of 34,520 above-chance voxels in three-way win-tie-loss classification, only 25 voxels showed a significant difference between win-tie and tie-loss classification

(42 without cluster correction). Therefore, signals related to both reinforcements and punishments were remarkably ubiquitous, and there was very little difference between encoding of the two. The addition of tie outcomes in Experiment 2 afforded the ability to distinguish signals related to reinforcement and punishment from those related to salience. One possible explanation for the ubiquitous reward signals in Experiment 1 is that one of the two outcomes about in the matching pennies game is more attention-demanding or salient (Maunsell, 2004, Bromberg-Martin et al., 2010, Chun et al., 2011 and Litt et al., 2011). By contrast, during rock-paper-scissors, the “tie” outcome should be less salient and arousing than both wins and losses. We evaluated the salience hypothesis by using a pair of classifiers. First, we trained classifiers to discriminate wins from ties (win-tie classifier), then evaluated whether they tended to classify unseen losses as wins or ties. Next, we also trained classifiers to discriminate ties from losses (tie-loss classifier), then evaluated whether they tended to classify unseen win trials as ties or losses.