2.6 Data Processing
Data were processed offline using in-house MATLAB scripts leveraging EEGlab functions (Delorme and Makeig, 2004). Data were high-pass filtered at 0.1 Hz (order: 16,896; cutoff frequency: 0.05 Hz; transition band width: 0.1 Hz) and low-pass filtered at 45 Hz (order: 153; cutoff frequency: 50.625; transition band width: 11.25 Hz) using a finite impulse response filter implemented through EEGlab (pop_eegfiltnew). Bad channels were detected automatically based on joint-probability and kurtosis. A channel was defined as “bad” when recorded data from that electrode exceeded 5 SD from all other electrodes. For eye artifact removal, Independent Component Analysis (ICA) was applied to continuous data high-pass filtered at 1 Hz to identify components corresponding to eye movements and blinks. Based on output from an ICA automated component classifier (ICLabel) (Pion-Tonachini et al., 2019), components with > 80% probability of originating from eye movements and a < 5% probability of origination from a brain source were rejected.
Data were re-referenced offline to the common average. EEG response-locked epochs were constructed from -350 to +300 ms surrounding the button press. A baseline was defined from -350 to -300 ms prior to the button press. An automatic artifact rejection criterion of ±150 microvolts was used to remove trials containing excessive muscular activity and eyeblinks. An additional artifact rejection threshold was calculated based on an array of maximum amplitudes for each trial (the largest absolute value recorded in a given epoch across all channels). Epochs containing values > 3 SD from the median of this array of maximum values were removed. Trials in which participants responded earlier than 150 ms after the cue and trials in which participants responded greater than 1000 ms following the cue were excluded. For TD participants, an average of 4.48% of trials met criteria for removal based on this response profile, while for ASD participants, 7.74% of trials met these criteria (t(166) = -3.963, p < 0.001). In addition, trials with RTs > 3 SD faster or slower than the mean RT of all remaining trials for that individual were excluded from ERP waveforms.