Table 1.

Summary of analyses, models, and simulations used in the manuscript

Analysis	Sessions/Units	Data/Parameters	Figures
Principal Components Analysis (PCA) with overlapping RT bins and choice	• 141 sessions • 996 units • 30 ms Gaussian • 801 units (*) • 15 ms Gaussian and 50 ms Boxcar (∘)	Smoothed PSTHs of trial-averaged units combined across sessions and organized by overlapping RT bins and choice; 6D trajectories for associated KiNeT analysis	• Fig. 4 • Fig. S5∘ • Fig. S7 • Fig. S10* • Fig. S11∘ • Fig. S20
Principal Components Analysis (PCA) with coherence and choice	• 141 sessions • 996 units • 30 ms Gaussian	Smoothed PSTHs of trial-averaged units combined across sessions and organized by coherence and choice; 6D trajectories for associated Kinematic analysis of Neural Trajectories (KiNeT)	• Fig. 7
Principal Components Analysis (PCA) with trial outcome and choice	• 141 sessions • 996 units • 30 ms Gaussian	Smoothed PSTHs of trial-averaged units combined across sessions and organized by outcome of trials and choice; 6D trajectories for associated KiNeT	• Fig. 8 • Fig. S18 • Fig. S19 • Fig. S20
Principal Components Analysis (PCA) with nonoverlapping RT bins and choice	• 141 sessions • 996 units • 30 ms Gaussian	Smoothed PSTHs of trial-averaged units combined across sessions and organized by nonoverlapping RT bin and choice; 6D trajectories for associated KiNeT	• Fig. 5
Choice/outcome decoding analysis & RT regression analysis  • Curve Fitting Toolbox  • Statistics & Machine Learning Toolbox	• 51 sessions • 2 to 32 units • 157 to 2349 trials	Binned spike counts (20ms bin size)	• Fig. 6 • Fig. 8 • Fig. S16 • Fig. S17
Single neuron regression analyses	• 141 sessions • 996 units	Smoothed firing rates (50ms causal boxcar)	• Fig. S10
Latent Factor Analysis of Dynamical Systems (LFADS)  • https://github.com/lfads	• 1 session; 23 units (*) • 16 sessions; 11 to 32 units (∘) • 631 to 1403 trials	Binned spike counts (10ms bin size); 8 factors	• Fig. 6* • Fig. S14∘ • Fig. S18*
Tensor Component Analysis (TCA),  • https://www.tensortoolbox.org/  • https://github.com/ahwillia/tensor-demo	• 44 sessions • 2 to 32 units • 370 to 2349 trials	Causally smoothed firing rates (50 ms boxcar), ranks from 1 to 4. Sessions with fewer than 4 units we only ran TCA up to the maximum number of units for that session.	• Fig. S12
Linear Dynamical Systems (LDS),  • https://github.com/gamaleldin/CFR	• 31 sessions • 11 to 32 units • 97 to 960 trials	Causally smoothed firing rates (50 ms boxcar); step of 100 ms. Only left trials were analyzed.	• Fig. S13
Reduced-rank regression  • https://github.com/cmccomb/RedRank	• 41 sessions • 2 to 32 units • 370 to 2349 trials	Causally smoothed firing rates (50 ms boxcar); step of 100 ms.	• Fig. S15
Recurrent Neural Network models (RNNs)  • https://github.com/murraylab/PsychRNN	• 100 units	ReLu nonlinearity	• Fig. S2
Simulated population of PMd neurons	• 350 units	200 Increased, 100 Decreased, and 50 perimovement units	• Fig. S3

Where applicable, links to the toolboxes used are provided. Most analyses were performed using MATLAB.