mrif Wrote:
-------------------------------------------------------
> Gang Wrote:
> --------------------------------------------------
> -----
> > I can think of the following: use the two
> values
> > of visit2 - visit1 and visit3 - visit2 from
> each
> > subject as input with the two time intervals as
> > covariate in 3dLME. Follow example 2 in the
> 3dLME
> > HELP. Not sure if that would address your
> > question.
> >
> > Alternatively ignore the time intervals, and
> > obtain the intraclass correlation (ICC) value as
> a
> > measurement of reliability using 3dICC_REML:
> >
> >
> [afni.nimh.nih.gov]
>
>
> Gang, thank you for your suggestions.
>
> If we want to stick with the 3dLME approach,we
> want to do it without the subtraction.
>
> So, we are looking at 3 sets of covariates for
> each visit.
>
> Having all zeros for visit 1, is this possible?
> What if I set the values to all 1's instead of
> zeros for visit 1,would that make a difference?
> Also, what is the code for compound symmetry for
> the covariance structure option, -corStr?
>
> Using 3dICC_REML, is there a way to include a
> time-varying covariate?


Gang, I wanted to follow-up regarding this issue.

3dLME:

1) Having all zeros for visit 1, is this possible? Will that cause any problems?
2) Is Compound symmetry the default covariance-variance structure if you don't specify one in the options.
3) Is there a way to extract the variance components (residual, between-subject, within-subject) from the output of 3dLME. Usually, you can calculate the intraclass correlation from those components from linear mixed models. Is this possible since I do not see an option for covariates for 3dICC_REML?

3dICC_REML

1) Is there a way to include a time-varying covariate or time-invariant covariate?

Again, the issue is the impact of the covariate and its effect on variance components. We want to parse out or control for any effects of the covariate when calculating the ICC coefficient.

Thanks,


Michael