Hi FSLers,
I have longitudinal data and was interested in analyzing the association between change scores in a variable of interest and brain changes. In order to use randomise and TFCE, I just substracted participants' scores at time 2 minus time 1, did the same with the imaging data, and then ran randomise to look at the correlation between change (difference) scores and brain changes.
However, a reviewer complained about this procedure and suggested implementing instead a linear mixed-effects regression (a procedure that I believe can't be implemented in randomise?).
It is true that difference scores are often unreliable and researchers in the field of psychometrics avoid using them... yet, I often see that substracting time points is recommended in the forum. I guess these are similar scenarios since the error variance of time 1 and time 2 are compounded? Anyways, since you recommend substracting, I was wondering if you could help me argue that this is actually an appropriate procedure.
