Fig. 7: LHb value signals predict online multi-attribute decisions.

a, Schematic of testing if LHb value-related activity is choice predictive. The black bar indicates the analysis time window. CC, corpus callosum; TH, thalamus; SC, superior colliculus; IC, inferior colliculus; Cb, cerebellum. b,c, The LHb example neuron from Figs. 1–4 had trial-to-trial variations in value-related activity during offer 1 (b) and especially during offer 2 (c) that predicted trial-to-trial variations in choice. Scatter plots show the correlation of residual neuronal value signals in response to each offer (x axis) versus residual choice of that offer (y axis). Each dot represents one trial. Colors indicate choices of offer 1 (purple) or offer 2 (orange). Lines are linear fits (type 2 regression). The text indicates rank correlation and its P value. Top, histograms of residual neuronal value signals for choices of offer 1 versus offer 2; the text indicates receiver operating characteristic curve (ROC) area and P value (rank-sum test). d, The LHb population average relationship between normalized residual value signals (x axis) and residual choice (y axis) had a more positive slope for offer 2. Two asterisks (**) indicate the 99% bootstrap confidence interval excluded 0. Error bars represent ±s.e. Shown are all n = 301 attribute-responsive LHb neurons with sufficient data from both offers for this comparison. e,f, Histograms of each neuron’s choice predictive indexes for offer 1 (e) and offer 2 (f). The dark areas indicate significant indexes (P < 0.05). The dashed vertical line and text indicate mean and significance of the median (signed-rank test); text in the lower right indicates the fraction of neurons with significant positive indexes and whether it is above chance (one-tailed binomial test). g, Both animals had higher mean choice predictive indexes for offer 2 (orange) than for offer 1 (purple); *P < 0.05; **P < 0.01; ***P < 0.001; signed-rank tests. Data are shown as mean ± s.e.; n = 122 and 179 for animals R and Z, respectively. h, Venn diagrams show, for neurons in each area with strong attribute effects, the overlap of neurons with high value coding indexes (left), significant RPE coding indexes (right) and significant choice predictive indexes (center); other neurons that met none of those criteria are shown at the bottom. Colored areas indicate neurons with combined coding of all three properties. See Supplementary Table 6 for the details of all tests and P values.