I was doing EEG recordings as part of my project and at certain times stimuli appeared in the form of arrows. As soon as the stimulus arrives, a movement is performed. During signal processing, I analysed my EEG data set as follows: After bandpass filtering, I segmented my data set. I had 30 stimuli and ended up with 30 5-second segments containing 2s before the stimulus and 3s after the stimulus. According to my literature, ERD occurs in the µ-frequency range during movement, so I wanted to analyse the 30 segments as follows: Split each segment into two parts, a 1.5s part before the stimulus and a 3.5s part just before and after the stimulus, and then compare these parts in the frequency domain to analyse possible ERD.
My question is whether the selection of the 1.5s and 3.5s phases makes sense, since the stimulus and the movement occur at time 2s. When I use the 1.5s and 3.5s phases, I see a higher µ concentration in the 3.5s phase, which contains the movement of the hand. And this is not correct according to my literature, because the µ-concentration should be lower during the motor execution.
There are a couple of things to consider here. First, mu ERD appears at contralateral sites, so if you have a right hand movement, you will see mu ERD during movement at and around C3. Second, it takes around 0.5s for the oscillations to go down, so I would recommend a time window starting 0.5s after stimulus onset (until movement offset). The baseline should be a shortish time segment before the stimulus, ideally not including t = 0 to make sure not to contaminate the signal with correlates of the stimulus (e.g. VEP). Third, the exact mu frequency band varies individually, so you might want to inspect t/f maps (ERD/ERS maps) to guide your selection of the mu band (see e.g. Compute and visualize ERDS maps — MNE 1.6.1 documentation).
I would also add that you need to consider your paradigm. It’s not that rare to see the ERD starting during motor preparation (especially if the movement timing is predictable), which may mean that what you consider as baseline in this case might be capturing something like this. Also, depending on the type of movement you may need to consider less than the whole 3 seconds - only the time until movement offset as @cbrnr mentioned.