Dear Alex,
> Dear ladies and gentlemen
>
> My name is Dr. Alex Drzezga from Munich in Germany, maybe anybody can
> answer my obviously relatively simple question regarding SPM99
> correlation analysis in our study in 20 subjects. F-18 FDG scans were
> performed at two points of time (baseline and after one year). I want to
>
> do a correlation analysis between metabolic changes within one year and
> STABLE covariates (genetic factors, family history).
> What would be the proper approach for this type analysis? I tried
> "multi-subject with conditions and covariates" and used TWO conditions
> and ONE covariate, however the program always demands covariates for
> every condition. Therefore I entered the same covariate twice for each
> subject at condition 1 and 2 (genetic factors do not change within one
> year). My idea was to enter the following contrasts 1 -1 1 1 and 1 -1 -
> 1 -1 to find out positive or negative correlation between genetic
> predisposition and changes of cerebral metabolism, which did not work.
> The program only allows 1 -1 1 -1 or 1 -1 -1 1 which would calculate the
>
> correlation of changes in cerebral metabolism with changes in the
> covariate, however I want to calculate changes in cerebral metabolism in
>
> dependency of a STABLE covariate only variing between individuals but
> not between conditions.
I guess there is a number of ways in which to do this analysis, but here is
my two pennys worth.
Let us start with a fixed-effects model which should do what you want. If
we estimate separate regression lines for the first and second scans then
the difference between their slopes should answer your question. The
simplest here would be, as you suggested, a "multi-subject conditions and
covariates" design. Specify two conditions (first and second scan), one
covariate (which you enter as [a a b b c c ...], i.e. enter the same value
twice for each subject), covariate-by-condition interaction and centering
around condition means. Now the contrast [0 0 -1 1] will show you areas of
positive correlation between metabolic change and your covariate. As you
see it is very close to your original thought.
A slightly more kosher way would be to do a random-effects analysis, which
is in this case perhaps conceptually simpler. Then you simply enter a
difference image for each subject into Basic Models->Simple Regression.
Good luck Jesper
|