Fig. 4: Effect of consistency on evidence encoding and evidence readout.
From: Confirmation bias through selective readout of information encoded in human parietal cortex
a, b Time courses of stimulus sample information I(S;R) in dorsal visual cortex for consistent (C) and inconsistent (I) samples, relative to cue (a) or choice (b). Time-courses are pooled across samples 7 and 8, showing the behavioral consistency effect in Fig. 2. Bar charts, mean across gray-shaded interval with significant I(S;R). Bars or lines, group mean; error bars or shaded areas, SEM. Horizontal bars, significant difference from (−0.1 s to 0 s) baseline interval (colors) or between consistent and inconsistent (black); p-values from two-sided permutation tests (cluster-corrected for time courses); N = 29 participants. Time courses were smoothed with Gaussian kernel (sigma = 37.5 ms). c, d As (a, b) but for estimation information I(R;E). e, g As (a, b) but for intersection information II(S;R;E). f, h Maps of consistency effects on II(S;R;E) displayed without threshold. Abbreviations: DVC, dorsal visual cortex; IPL, inferior parietal cortex. i Correlation between maps of consistency effects on I(S;R) and I(R;E) in Cue and Choice conditions (Pearson’s correlation coefficient, one-sided). Data points, areas (N = 180); p-values from spatial autocorrelation-preserving permutation tests (Methods). j As (i) but for pattern similarity of consistency effect maps on II(S;R;E). k Schematic of mismatch between neural stimulus encoding and readout. Vectors represent neural codes for sample stimuli (gray) or behavioral estimates (colors) in a two-dimensional state space. The alignment of the neural activity patterns governs the contribution of encoded stimulus information to behavioral report. When samples are consistent, the readout is well-matched (green vector). When samples are inconsistent, the readout is mismatched (orange vector). Also the stimulus vector may change between conditions; we show a single stimulus vector for simplicity: What matters is the relative alignment of the codes for stimulus and estimate within condition. See Supplementary Fig. 5 for simulations.
