Fig. 4. Choice is distributed, near orthogonal to other components and with a ventral-stream dominance.

a Angles between state axes averaged across animals. Choice axes (pre- and post-movement) were orthogonal to all other axes (smallest angle 84 ± 7°). Attention and wheel had the smallest angular separation (44 ± 3°), followed by wheel and saccades (56 ± 4°). b Hierarchical clustering from the angle distances in a. Attention and wheel movements were most similar. Choice pre- and post-movement onset clustered together, whereas stimulus and saccades had unique profiles. c Angle between choice and movement state axes averaged across animals aligned to movement onset (line for average across animals and shaded area for its 95% CI, n = 7 animals); shaded gray band is the expected angular distance range for statistically independent axes; observed angles never significantly deviated from the statistically independent condition. d Spatial-Distribution index (SDI) for each state axis. Choice had the largest SDI (30 ± 4%); dots are different animals; middle lines and shaded areas are means and their 95% CI (n = 7 animals). e We computed five d’ values, each derived by restricting locaNMF components to one of the five area groups (insets), thus defining a 5-D space for d’ components. The five broad area groups consisted of the dorsal stream (PM and AM), ventral stream (L), posterior parietal (A, AL, and RL), somatosensory (SSt and SSb) and retrosplenial (RS) regions. Each plot shows a 2-D projection of the five broad area groups, where each dot corresponds to the d’ values for a given animal and the large dot with errorbars to the average across animals and s.e. (n = 7 animals).