EEG statistical analyses

Hi Mike and others, I hope you are all doing great.

I have a statistics-related question. Regarding analysis type 2a (and 2b), I would like to use the extracted mean power value as Y-variable in a mixed model. However, as I have different mean power values per brain region (3 brain regions) and per frequency band (3 conditions), I am contemplating whether I should use 3 mixed models (per brain region) or 9 mixed models (per brain region, per condition).

Any help would be appreciated


Hi Sybren. Normally it’s better to include all relevant factors into one model. That helps correctly divvy up the variance while also controlling for multiple comparisons (in this case, multiple models).

However, complicated models may be poorly estimated if there isn’t “enough” data (and yeah, how much is “enough”??), which would present an argument for having simpler models with fewer parameters that are easier to fit.

Also keep in mind that you cannot compare effects across models. So if each condition has its own model, then it’s not possible to say whether there is an effect of condition.

So I can’t really give you specific advice, but my general advice is to include as many factors as reasonable into one model, so that you have fewer models overall.

Finally, this question isn’t specific to EEG processing; it’s about statistical modeling more generally. It might be a good idea to ask the advice of a stats person who specializes in mixed-effects modeling.