Figure 3.

Combining information across subjects compensates for inaccuracies in localization due to multiple experimental factors. Simulations were run covering different regulation parameters λ (A), noise covariances used in inverse estimation (B), errors in coregistration alignment (C), and a range of signal-to-noise ratios (D). The centroid error (solid lines) and the point spread (dashed lines) were both reduced as information across subjects was combined in all of these scenarios, suggesting that the simulation results from Figure 2 generalize to many situations. Mean ± 2 s.e.m. (across 20,424 cortical locations) is shown by the lines and shaded backgrounds. Note that s.e.m. are mostly small enough to be masked by the mean lines; standard deviation values across cortex are provided in Table 1. Here for simplicity only the V = 25 point case is shown, the V = 25 lines in Figures 2A,D equivalent here to the $\sqrt{(1/\lambda )}=3$ case in A, the ERM case in B, and the 0 mm case in C. The infinite SNR case in D corresponds to the task-based covariance case from B, since the evoked covariance was used in the inverse solutions for that simulation.