search

josephdviviano / fslpal

Utilities for running stats using FSL's tools (without using FSL's GUI)
MIT License
0 stars 1 forks source link

pal-randomize and covariate interactions #2

Closed josephdviviano closed 8 years ago

josephdviviano commented 8 years ago

Sonja is running an interaction analysis on a covariate between groups. To do the interaction, the covariate is split into two columns, with zeros entered for the subjects in the other groups. For example:

1 0 -2 0 1 0 -1 0 1 0 -1 0 1 0 1 0 0 1 0 2 0 1 -0 1 0 1 0 1 0 1 0 1

Where column 1 if group1, column 2 is group2, columns three is the covariate in group1, and column four is the covariate in group two.

She is under the impression epi-randomise is currently mean-centering columns 3 and 4 separately. This is not correct. When analyzing an interaction of a covariate, the covariate should be mean-centered across ALL subjects, then split into two columns.

https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM#Two_Groups_with_continuous_covariate_interaction

I didn't check this myself, so maybe it is correct and Sonja misunderstood the code.

josephdviviano commented 8 years ago

@colinhawco there's no case where you would want to mean center each column separately?

colinhawco commented 8 years ago

For a co-variate interaction, I don't think so. If I understand correctly, if they are mean centered independently it could either detect a main effect group difference or, more likely, a difference in the intrcept? Maybe. It's to late for me to be doing statistics, but the FSL wiki is pretty clear that for an interaction they should be centered together before being separated into group columns.

On 21 September 2016 at 13:29, Joseph Viviano notifications@github.com wrote:

@colinhawco https://github.com/colinhawco there's no case where you would want to mean center each column separately?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/josephdviviano/fslpal/issues/2#issuecomment-248683941, or mute the thread https://github.com/notifications/unsubscribe-auth/AI5LmzRseexYZWvkDpBwHHKthaECIoGXks5qsWmBgaJpZM4KDEdL .

Colin Hawco, PhD Neuranalysis Consulting Neuroimaging Research Analysis and Consulting www.neuranalysis.com

josephdviviano commented 8 years ago

Change made. see c1fb4dd