*2.3. EEG Recordings*

Scalp electrical activity was recorded through an Enobio device (Neuroelectrics; eight channels, O1, O2, P3, P4, C1, C2, F3, F4; sampling frequency 500 Hz), using the right ear lobe as reference. Electrodes were gel-based passive plates (Ag/AgCl coated; impedance <5 k Ω) and were placed on the scalp by means of an EEG cap. EEG traces were band-pass filtered (0.1–100 Hz).

## *2.4. Data Analyses*

To quantify perceptual responses, we used the subjective estimates of video speed modulation strength (pulsation rating). By separating null (rating = 0) and non-null (rating > 0) responses, we transformed gradual judgments into dichotomous yes/no responses, and we applied a signal detection analysis [6]. Basically, the rating task was treated as a multiple yes/no detection task [7], from which we calculated hit rate (non-null responses in signal trials, i.e., trials in which speed was modulated) and false alarm rate (non-null responses in noise trials, i.e., trials in which speed was not modulated). A correction for extreme values was applied [8]. Given the ten-point rating scale used in this experiment, there were nine possible pairs of hits and false alarms: ratings greater than 0 were first considered to be "yes" responses, while a 0 rating was considered to be a "no" response; next, ratings greater than 1 were considered to be "yes" responses, while ratings less than 2 were considered to be "no" responses, and so on, until encompassing all nine pairs of hits and false alarms. A receiver operating characteristic (ROC) curve was thus fitted and the area under the curve (AUC) computed. For each observer, and for each amplitude and frequency of the two stimuli, the perceptual sensitivity to video speed modulation was obtained by converting the AUC into a corresponding d' index [6].

To quantify cortical responses, we used the Eeglab software ERP averaging tool pop\_averager [9] to obtain the averaged traces of each channel over 1-s time windows. Each averaged trace was then fitted to a sinusoidal model (through the Matlab fit function) to compute the cortical response strength, measured as the peak amplitude of the fitted function.

Both the perceptual and the cortical responses were subjected to generalized linear mixed models analysis (GLMM, with diagonal covariance pattern, normal distribution and identity link). The frequency (Freq) and amplitude (Ampl) of video speed modulation were fixed factors, while the video clip (Clip) and participant (Subj) were modeled as random intercept terms to reduce overfitting [10]. Following [11], the recording channel (Chan) for the EEG analysis was also modeled as a random intercept term. The dependent variables were either perceptual sensitivity (PS) or the amplitude of the fitted sinusoidal function (FA). The models were thus "PS ~ 1 + Freq \* Ampl + (1 | Clip) + (1 | Subj)" and "FA ~ 1 + Freq \* Ampl + (1 | Clip) + (1 | Subj) + (1 | Chan)", respectively, for perceptual and cortical data.
