All of the data analyses were performed with custom-written Matlab code (Mathworks, Natick, MA) using the fieldtrip toolbox ( http://www.ru.nl/fcdonders/fieldtrip/)58. We first performed an artefact rejection procedure. In this procedure, the trials in which the LFP power exceeded 4 s.d. from the average were excluded from the analysis. Then, we performed fast Fourier transformation following hanning tapering of the data59 (a multitaper method with a discrete prolate spheroidal sequence was used to analyse 40–100 Hz frequencies) around the time of the target stimulus onset and the delay period. The analysis window started at 1,500 ms prior to and ended at 1200, ms after the target stimulus onset. The analysis time window did not have any overlap with the reward presentation to avoid contamination of the data with possible reward consumption artefacts. We performed the frequency analysis using a 670 ms sliding time window in 50 ms steps. The power spectra for individual channels were z-score normalized using a baseline of 500–1,000 ms prior to the fixation point onset in both memory-guided saccade task and pro-/anti-saccade task.
