I ran a 3dttest++ with two groups and one 1D covariate. The data had been ranked within each group prior to this analysis using 3dTsort. Using 3dmaskdump, I then obtained the values for each subject from the data used in the 3dttest++ analysis for 7 voxels surrounding a peak voxel from the 3dttest++. I then correlated these values with the covariate (within each group) to obtain an estimate of the slope. My issue is that, for one group only, when I did this I obtained slightly different values (difference ranging in magnitude from .0036-.0042) than what I get from running 3dmaskdump with the same voxels on sub-brick #10 (slope of setB wrt covariate) of the 3dttest++ (and also from pinpointing the voxels in the AFNI GUI). I've done this same analysis for the same subjects on the same data but with a different covariate without this issue, and with the same covariate but different data and I don't have the issue (the correlation value matches the value from sub-brick #10). I have triple checked all the labeling, scripts, and covariate files and everything seems to be as it should be.

It's a long story as to why I'm doing this analysis in the first place, but my main concern is that there is something weird with the data since the values don't match only for this one group and this one analysis (but they do for all the other ones I've run). Do you have any troubleshooting tips that may help me find what the discrepancy is between my results from calculating the correlation based on individual subjects' data, and the slope value afni provides?

Thanks.