So if it takes a long time to converge and the G matrix is not positive definite, can one trust the fixed effects part of the model?

Dr. Paul R. Swank, Professor and Director of Research Children's Learning Institute University of Texas Health Science Center-Houston

-----Original Message----- From: SAS(r) Discussion [mailto:SAS-L@LISTSERV.UGA.EDU] On Behalf Of Dale McLerran Sent: Thursday, April 30, 2009 7:40 PM To: SAS-L@LISTSERV.UGA.EDU Subject: Re: complex var-cov structure in mixed

Paul,

If you fit the models separately for the two groups, then you would be able to estimate different covariance structures for each group. But if group has to affect parameters of the fixed effects part of the design, then you can't use such an approach.

However, you can allow the components of variance (as well as residual error variances) to differ across groups. Refer to the GROUP= option on both the RANDOM and REPEATED statements. When a variable is named to the GROUP= option, then different levels of that variable produce different estimates in the variance/covariance structure. Effectively, if the variance of one of the effects named on the random statement goes to zero for some group level, then you do estimate different covariance structures across groups.

Dale

--------------------------------------- Dale McLerran Fred Hutchinson Cancer Research Center mailto: dmclerra@NO_SPAMfhcrc.org Ph: (206) 667-2926 Fax: (206) 667-5977 ---------------------------------------

--- On Thu, 4/30/09, Swank, Paul R <Paul.R.Swank@UTH.TMC.EDU> wrote:

> From: Swank, Paul R <Paul.R.Swank@UTH.TMC.EDU> > Subject: complex var-cov structure in mixed > To: SAS-L@LISTSERV.UGA.EDU > Date: Thursday, April 30, 2009, 1:20 PM > A colleague has approached me with > the following problem. In a two group design with a growth > curve model, the structure of the var-cov matrix is > different between groups. Unfortunately, in one group, the > variance on the slope and the covariance between the > intercept and slope is close to zero, causing great problems > with convergence and NPD messages about the matrix, using > SAS Proc Mixed. The question is, while one can allow the > parameters to vary between groups, is it possible to allow > the structure itself to vary between groups so that you have > intercept and slope variance, plus covariance, in one group, > but only intercept variance in the other. > > (Apologies for cross-listing). > > Dr. Paul R. Swank, > Professor and Director of Research > Children's Learning Institute > University of Texas Health Science Center-Houston >