group level analysis

I’ve been computing Power of TFR (40 subjects). I’ve encountered a problem in collecting the data for parametric group statistics (ANOVA): for specific frequency bands in a specified predetermined time window, for some subjects I get synchronized activity (positive values; red) while for others I get desynchronized activity (negative values; blue). This mixture of positive and negative values for a specific frequency band is problematic in terms of statistical analysis. Do you have any suggestions to overcome this problem?

Hi soffery. First of all, apologies for your message being held up for 5 days. The site thought it might be spam and waited for me to approve it, but didn’t email me about it. Anyway.

I understand your concern. Without seeing the data, I can’t really give specific advice or say what the issue could be. But there are three possibilities:

  1. Some subjects have noisy data. You can look through the single-subject data more carefully and see if you find noise that you might have missed during cleaning/processing.
  2. The variable effects direction is meaningful and due to some individual difference that you are not (yet) taking into consideration. This happens frequently when trying to interpret the main effect of an ANOVA in the presence of a cross-over interaction.
  3. There is no effect. If the null hypothesis is true, then by chance (sampling variability, noise, etc), you’d expect ~50% of the subjects to have the effect in one direction and ~50% in the other direction.

Hope that helps!
Mike