Dear Statalisters,
I'm using Stata 15.1.
I have a dataset with multiple outcomes and three time periods. I have performed some multivariate repeated measure mixed models by following the "gsem" model shown here by Joseph Coveney: https://www.statalist.org/forums/for...del-with-stata to identify the effect of time and other covariates (possibly including interactions) on outcomes.
I've read, however, that the doubly multivariate repeated measure model is more flexible, by relaxing the sphericity assumption. If I'm not wrong, this translates into allowing for unstructured within-group error correlation of the random effect at different time periods (so that the error correlation between time 1 and time 2 doesn't need to be the same as the one between t1 and t3 and t2 and t3).
At least this was my understanding: please correct the possible errors.
My question is: is it possible to perform it in Stata? In particular, is it possible to choose the correlation structure of the error term as we can do in the univariate case (with the "residuals" option of the "mixed" command, or the "corr" option of the "xtgee" command)?
I'm using Stata 15.1.
I have a dataset with multiple outcomes and three time periods. I have performed some multivariate repeated measure mixed models by following the "gsem" model shown here by Joseph Coveney: https://www.statalist.org/forums/for...del-with-stata to identify the effect of time and other covariates (possibly including interactions) on outcomes.
I've read, however, that the doubly multivariate repeated measure model is more flexible, by relaxing the sphericity assumption. If I'm not wrong, this translates into allowing for unstructured within-group error correlation of the random effect at different time periods (so that the error correlation between time 1 and time 2 doesn't need to be the same as the one between t1 and t3 and t2 and t3).
At least this was my understanding: please correct the possible errors.
My question is: is it possible to perform it in Stata? In particular, is it possible to choose the correlation structure of the error term as we can do in the univariate case (with the "residuals" option of the "mixed" command, or the "corr" option of the "xtgee" command)?
