A bandpass filter from 0.1 to 100 Hz was applied to continuous MEG data. A sharp discrete Fourier transform filter was applied at 60 Hz to reject line noise. Data were initially epoched to −400 to 2000 ms relative to stimulus onset. Jump artifacts were removed based on deviation from a median filter. Participants were excluded if more than 10% of trials were rejected during preprocessing. The period of −400 to 0 ms was used for baseline correction and 0 to 2000 ms was used to capture dynamics related to the task. MEG and MRI data were coregistered using the fiducial markers and single shell head models were constructed from the segmented MRI (Nolte, 2003).

The covariance matrix used for source estimation was constructed from the longer trial epoch (−400 to 2000 ms). Centroids of each parcel in the Brainnetome atlas, consisting of 246 regions (123 per hemisphere) (Fan et al., 2016), were used in calculation of the leadfield matrix, and the time-series at each position was estimated using a linearly constrained minimum variance beamformer (Van Veen et al., 1997) with 0.1% regularization. Noise estimates for each “virtual sensor” were also projected and used to normalize each estimated time course (i.e., compute the neural activity index). The linearly constrained minimum variance natively produces estimates of activity in the three canonical planes. For this study, we project along the dominant orientation, which is equivalent to taking the first eigenvector of the time series.