> I do not have access to the appropriate residual file to estimate smoothness with 3dFWHMx for 3dClustSim.

You could use the -resid option in 3dMVM to generate the residuals for ACF estimation, which is arguably a better approach than the individual-level residuals.

> I also have two within-subjects conditions in my task ("fearful faces", "neutral faces").

To use 3dttest++, you already obtained the contrast or interaction effect between the two within-subject factors as input for 3dttest++?

For categorical variables (i.e., Site, Sex), code the two levels as 1 and -1 (or 0.5 and -0.5). See discussion here:

[stats.idre.ucla.edu]

For the centering issue, your specification seems fine.

Gang