Hi everyone! I am finalizing the same analyses I posted about a few weeks ago and am driving myself crazy trying to select the best covariance structure for my mixed model. I have gone back and fourth and changed all my tables multiple times. I read many articles and blog posts and I'm still torn. Hoping to draw on some of the ever-helpful advice from this forum.
I am running a simple mixed effects model, examining a RCT (N=600; 150 in each of four groups) with four equally spaced time points (baseline, 6 month, 12 month and 18 month follow-up). My basic command is as follows:
tmixed mCD4 i.group##i.time c.D2_a_0 c.QOL c.AM7__0 c.Tot_MET c.ISscale c.FItotal_0 i.educ4 || Subject:, var reml
That sets the covariance structure to the default (independent). The AIC and BIC are lowest in this covariance structure.
The identity covariance structure has a little bit higher AIC and BIC but not by much.
Of course the other option is unstructured, which has a higher AIC and BIC.
tmixed mCD4 i.group##time c.D2_a_0 c.QOL c.AM7__0 c.Tot_MET c.ISscale c.FItotal_0 i.educ4 || Subject:, var noconst residuals(unstr, t(time)) reml
Any ideas on how to select would be beyond appreciated.
Thank you!
I am running a simple mixed effects model, examining a RCT (N=600; 150 in each of four groups) with four equally spaced time points (baseline, 6 month, 12 month and 18 month follow-up). My basic command is as follows:
tmixed mCD4 i.group##i.time c.D2_a_0 c.QOL c.AM7__0 c.Tot_MET c.ISscale c.FItotal_0 i.educ4 || Subject:, var reml
That sets the covariance structure to the default (independent). The AIC and BIC are lowest in this covariance structure.
The identity covariance structure has a little bit higher AIC and BIC but not by much.
Of course the other option is unstructured, which has a higher AIC and BIC.
tmixed mCD4 i.group##time c.D2_a_0 c.QOL c.AM7__0 c.Tot_MET c.ISscale c.FItotal_0 i.educ4 || Subject:, var noconst residuals(unstr, t(time)) reml
Any ideas on how to select would be beyond appreciated.
Thank you!
Comment