Time-Frequency Analysis Baseline Correction

Dear Mike,

When running time-frequency analyses I’ve always used a baseline correction (typically dB), for the various reasons outlined in your book (e.g., converting to the same scale across frequencies etc). I’ve recently been looking at some EEG data using time-frequency analysis (wavelet based approach) taken pre and post an intervention as I’m interested in investigating changes in task-related oscillatory power. The thing is, this particular intervention also changes background (rest) activity quite a bit - e.g., it can cause quite pronounced changes, particularly in lower EEG frequencies at rest.

I’m wondering what your thoughts are on baseline corrections in these types of circumstances? I’ve tried looking at the results without comparing to the pre-stimulus baseline, but, as you highlight in your book, I found the results very difficult to visualize/interpret (I’m interested in a wide range of frequencies). I feel like a baseline correction would still be OK, but I suppose the intervention-related changes in rest activity would have quite a large impact on the final result?

Thanks again for your wonderful book and resources!

Yeah, when it comes to baseline normalization, there’s always at least one tricky case :wink:

Considering that you have a within-subjects design, it should be OK to compare the raw spectra pre vs post. It’s probably reasonable to assume that the subjects’ skull/skin don’t change systematically from the treatment. And furthermore, it is reasonable to assume that any differences in electrode preparation and quality are non-systematic. And the equipment is the same (I guess). So it seems that the possible confounds of raw power comparisons are accounted for here.

My advice is to do both pre-stimulus baseline normalization and no normalization. I discuss this in this video.

1 Like

Thanks Mike, that sounds like a reasonable approach, and thank you for the video link!