Multiple random statements in GLIMMIX

[Kastchei]

Hi all,

I have hopefully a quick question.  Do these two sets of random statements from GLIMMIX yield the same results?

random intercept slopeVar / subject = ID type = cs;

vs.

random intercept / subject = ID type = vc;

random slopeVar / subject = ID type = cs;

I'm thinking that the single statement applies the constant correlation between the intercept and the levels of slopeVar, whereas the two individual statements have 0 correlation between the intercept and the levels of slopeVar.  However, in my data, for both sets of statements, one covariance parameter estimate is always 0, and so they are resulting in the same V matrix in the end.  I'm guessing that isn't always the case though; or is it?

Thanks!

Michael

[SteveDenham · Accepted solution]

Not always the case, as you surmise.  It will depend on the actual data, and how much of it there is (how many IDs are being fit).

Steve Denham

[Kastchei]

Thanks.

So the latter would only be used if you have reason to believe that there is or should be no correlation between the intercept and the slope?  If I didn't know or suspect that, for example, a higher intercept results in a higher slope, then I'd want the former, single statement?

[SteveDenham]

I suppose you could make the assumption of no correlation, especially if you were running into convergence problems.  My inclination would be to allow for the possibility, and if it went to zero, that's fine.  Forcing it to zero in the face off no prior information seems like not such a good idea.

Steve Denham