compare covariates

AFNI Message Board
1

Hello,

I am using 3dttest++ to run a model on one group with two between-subjects covariates (A and B). I have shown that covariate A accounts for activity in my area of interest, but covariate B does not. A reviewer would like a statistical comparison of A > B. I am unsure how to do this in 3dttest+. I thought that I could treat the data as groups and have a different covariate for each, but this does not appear to be allowed in 3dttest. I would prefer not to do an ROI based approach due to the difficulty of selection bias (the activation for covariate A covers a substantially smaller volume than the anatomical ROI would). Any advice would be greatly appreciated.

Thanks,

Ian

2

A reviewer would like a statistical comparison of A > B.

I’m not so sure why the reviewer asked for such a test. Does it really make sense to compare the effects of A and B in the first place? What are A and B? And do they have the same physical meaning/scale?

3

What do you (or the reviewer) mean by “A > B”?

That the covariates values themselves are significantly different?
Or that the effect of covariate A is greater than the effect of covariate B in the region of interest?
Or something else?

4

Hi there,

Apologies for not being more clear. Yes, I would like to say the effect of covariate A is higher than the effect of covariate B in a region. Covariate A is a preference rating for a juice, and covariate B is a health rating. Right now I can say that preference, but not health, shows a covariate effect across subjects in region A. I would like to strengthen this by showing that preference has a stronger effect than health ratings, across subjects, in region A.

Thanks,

Ian

5

At present, there isn’t a way to do this in 3dttest++. I would have to derive the formula for t-testing the covariate slope estimates (taking into account covariance between the covariates), and then implement it.

At the present moment, I am deeply involved in developing a response (both in the AFNI code and in a manuscript) to http://www.pnas.org/content/113/28/7900.full and so don’t know when I will get around to this testing subject.

6

Hi Bob,

Thanks for the feedback. I understand the constraints on your time. I will just tell the reviewer it is not currently possible.

Thanks for the help,

Ian