Fig. 4 |. Value decoding in OFC and ACC.

a, We trained an LDA decoder to predict the value (1–4) on forced trials from neuronal firing rates. We applied the decoder weights to compute the posterior probability for each value in individual free trials in overlapping windows of 20 ms stepped by 5 ms. States are sustained periods (≥35 ms) of confidence decoding of a specific value. In both OFC and ACC, we observed multiple flips between chosen and unchosen value states of each trial. b, We observed more chosen (red) than unchosen (blue) than unavailable (gray) value states in both OFC and ACC, as determined from a one-way analysis of variance with post hoc t-tests, *P < 0.01 and **P < 0.0001. Error bars denote the s.e. of the mean number of states observed in a single trial (n = 3,865 trials for both ACC and OFC in subject C; n = 3,247 trials for both ACC and OFC in subject G). c, We observed longer chosen than unchosen than unavailable value states. Conventions are the same as b. Error bars denote the s.e. of the mean duration of each observed state. In subject C’s OFC, we observed 6,107 chosen states, 3,045 unchosen states and 2,259 unavailable states. In subject G’s OFC, we observed 6,420 chosen states, 3,877 unchosen states and 3,251 unavailable states. In subject C’s ACC, we observed 6,265 chosen states, 3,037 unchosen states and 2,450 unavailable states. In subject G’s ACC, we observed 5,762 chosen states, 3,651 unchosen states and 3,228 unavailable states. d, Choice response times were faster when the chosen value (red) was decoded more strongly and slower when the unchosen value (blue) was decoded more strongly. We built a linear regression for response time with chosen and unchosen decoding strength as predictors. Bold lines indicate significance at P < 0.01 value decoding. e, The posterior probabilities associated with value states were significantly positively correlated between OFC and ACC. We calculated the Pearson correlation coefficient for each session separately, and we then plotted the mean of these correlations. We assessed significance at each time point by performing a one-sample t-test against zero on the Fisher-transformed correlation coefficients. The t-tests were two-sided. To correct for multiple comparisons, we adjusted our significance level so that correlations had to be significantly different from zero (assessed at P < 0.01 and indicated by the thick horizontal lines) for more than 20 ms. This criterion ensured that there were no significant correlations in the 500 ms before picture onset. Interregional value correlations were significant and positive in both subjects shortly after the choice options became available. Error bars denote the s.e. of the mean of the correlation coefficients.