Hi Jeff,
> I've thus got 1 image for each participant and a mixture of categorical and quantitative predictors.
What kind of categorical variables do you have? Between- or within-subject factors? 3dRegAna can handle between-subjects factors, but it would be quite challenging if you have within-subject factors, and that's why you usually recommend 3dMVM in terms of modeling capability and ease of use.
> Does 3dRegAna test every single voxel in the input dataset?
Yes, it's a voxel-wise (or massively univariate) approach.
> does it perform some sort of auto-masking, so that zero-value voxels across most or
> all of the input images are excluded from analysis?
No, it does not.
> shouldn't the presence or absence of a mask affect multiple comparisons correction
> via FDR, 3dClustSim, or some other method?
For FDR, you can run the following command to set the FDR correction with a mask:
3drefit -addFDR -FDRmask ...
3dClustSim also accepts a mask from the user.
Gang