Pupillometry: Preprocessing

The diameter of the pupil was recorded at a 1000Hz using an EyeLink 1000 Tower Mount (SR Research). The eye-tracker was calibrated prior to each run. Blinks and saccades were detected using standard EyeLink software with default settings and Hedfpy, a Python package for preprocessing eye-tracking data. Periods of data loss during blinks were removed by linear interpolation, using an interpolation time window of 200ms before until 200ms after a blink. Blinks not identified by the manufacturer’s software were removed by linear interpolation around peaks in the rate of change of pupil size, using the same interpolation time window. The interpolated pupil signal was band-pass filtered between 0.05Hz and 4Hz, using third-order Butterworth filters, z-scored per run, and resampled to 20Hz. As blinks and saccades have strong and relatively long-lasting effects on transient pupil size [40, 95], these influences were removed from the data, as follows. Blink and saccade regressors were created by convolving all blink and saccade events with their standard Impulse Response Function (IRF) [40, 96, 97]. These convolved regressors were used to estimate their responses in a General Linear Model (GLM), after which we used the residuals of this GLM for further analysis. For the subsequent deconvolution analysis, trials were removed in which participants made a saccade towards either of the two presented colored stimuli (i.e. saccades exceeding 3.3° visual angle away from fixation) to ensure that pupil responses were not affected by eye movements (percentage removed trials, mean = 4.8%; SD = 4.5%; range = 0.0%-16.3%).