Extended Data Fig. 5: Alignment of single-units to random subspaces.

Random subspaces were constructed by independently and identically drawing from a normal distribution with zero mean and unit variance. Each random subspace was then biased towards the covariance structure of the actual data (see Methods). a. Null distributions of alignment indices for trial-averaged data. b. Null distributions of alignment indices for single-trial data. Null alignment distributions are skewed towards the movement-potent subspace due to the unbalanced variance between delay and response epochs (a) or between stationary and movement time points (b), reflecting the strong movement tuning of many units.