Extended Data Fig. 8: Chosen value relationship with choice decoding.

(a) Data are taken from the late epoch in Fig. 3b. Solid lines indicate ordinary least squares fit; dashed lines indicate robust fit from MATLAB function ‘robustfit’ with default arguments. Statistics r and ρ indicate Pearson’s and Spearman’s coefficients, respectively. P-values reflect two-sided test against the null hypothesis of r = 0 or ρ = 0. The x-axis gives the per-cell weights that define the choice subspace (see Methods), and the y-axis gives the per-cell weights obtained by fitting equation (2) to the variable ‘chosen value’ – that is the chosen value weights. (b) The choice decoding analysis in Fig. 6c was repeated after first projecting the spiking data into a subspace that is orthogonal to the subspace encoding chosen values. This was done with the same procedure used to project data into the subspace orthogonal to offer values, except that the matrix B has only one column, given by weights obtained by fitting equation (2) to the variable ‘chosen value’ (the same weights as in panel a). This effectively removed information about chosen values from the spiking data, after which the other steps necessary to reproduce Fig. 6c were performed. Conventions are as in Fig. 6c.