Figure 4. Coupling is associated with a longer timescale of population codes for choice in PPC.

a, Example population, instantaneous decoder posterior calculated on a single trial in PPC. b, Example demonstrating the calculation of posterior correlation. At times t and t+lag, the correlation coefficient between the posteriors at this interval was calculated for all trials (dots). c–d, Mean posterior correlation measured between all pairs of time points in the trial for stimulus and choice decoding. e, Time extent for which the mean autocorrelation functions in c–d is above 0.3 (dashed gray lines in c,d). *** indicates p < 0.001, z-test. f, τ₂ of double exponential fits to posterior correlation functions. Colored bars indicate unshuffled data; gray bars indicate data shuffled to disrupt coupling. The difference between colored and gray bars tests the contribution of coupling to the temporal consistency. Asterisks indicate significant differences between real and shuffled data, z-test from confidence intervals of fits. Error bars: 95% confidence intervals. g, Time-shifted coupling index, as in Fig. 3f, for pre- and post-turn periods. Asterisks indicate significant differences between pre-turn and post-turn data, rank sum test. h, τ₂ of double exponential fits to posterior correlation time courses in pre-turn and post-turn data. Brackets show comparisons between the contribution of coupling (unshuffled – shuffled) to the consistency of the choice signal across conditions, z-test. Error bars: 95% confidence intervals. i–j, Same as g–h, for behaviorally correct vs. error trials. * indicates p < 0.05; ** p < 0.01; *** p < 0.001.