Hello,

I need to do an across-subjects mixed effects analysis, in an study for which the within-subject contrast of interest is based on behavioral outcome (correct versus incorrect response.) Therefore, the design matrix is different for each subject, and there are not the same number of trials in each category for each subject. So it seems that just doing a second-level random effects analysis with beta weights as the "summary statistic" would not be valid. If I'm wrong on this, and it is possible to use the built-in ANOVA programs, just stop me here.

I am thinking of using Keith Worsley's MATLAB suite for fMRI analysis just to do the mixed-effects multi-subject analysis, since I am quite happy with the rest of the process done in AFNI. I have to feed this program two things: maps of the beta weights (actually a linear contrast, the subtraction of two beta weights), and corresponding maps of the standard deviation of that weight. I need to know how to generate the standard deviation maps. I think I have figured out a way, but I want to run it by the experts to make sure it's correct.

I am doing regression with basis functions, not FIR deconvolution. Here are my proposed steps:

1) Run the regression with 3Ddeconvolve, using the -vout option to get a map of MSE.
2) Re-run the regression using the -nodata option, to find out the "normalized" standard deviation of the contrast assuming MSE of 1, for each subject separately.
3) Square the number so obtained, and use 3dcalc to multiply it by the MSE value at each voxel obtained from -vout.
4) Take the square root of those numbers using 3dcalc.

And viola, a map of the standard deviations of the regression coefficient.
What do you think of this procedure? Thanks!