Dear all,
I have a question regarding how to perform statistical analysis on classifier topography patterns obtained using the Haufe et al. (2014) method.
To provide some background, I have fitted linear classifiers to the data collected in two experimental conditions, and both classifiers were successful. Next, I used the mne example on ‘linear classifier patterns and filters’ to examine the discriminant neural sources in each “condition classifier”. My goal is to determine if the topographies differ significantly between the two classifiers.
My question is whether it is appropriate to use a spatial permutation two-sample test to compare the topographies obtained from the two classifiers. I am treating the spatial patterns obtained from the classifiers as any other activation. Or, is there any type of normalization used in the Haufe et al. (2014) method that should prevent me from directly comparing the two classifiers trained under different conditions?
Additionally, would you see any impediment to testing each condition individually by applying a permutation one-sample test against the null hypothesis of zero?
Thank you for your help.
Sincerely,
Ana