Proc Glimmix-Multilevel analysis

[pires]

Hello everyone,

I am fitting a multilevel (5 levels , 3 random intercept) of longitudinal data ( binary outcome) using proc glimmix. When I test a random slope ( age) on the lowest level. I got the following message:

‘Estimated G matrix is not positive definite ‘ and the covariance parameter estimate of slope is zero.

It does mean that I don’t need the random slope age? ( that’s one of the suggestions for Proc mixed, not sure if the same applies to Proc glimmix).

Any help  would be appreciated ,

Alda

Code:

proc glimmix data=PENPIG10   method=RSPL noclprint=20 ;;

class   Age FarmID CohortID_C PenID_C PigID Nursc Envc Nursc4 Rxpc Subjpc Mortpc HthP Gender Season PenCat visit Tlw1wk_rtb;

model PigSalmCult (event='Pos')= Tlw1wk_rtb Nursc4  Mortpc

                                  Age FarmID    /solution link=logit dist=binary oddsratio ddfm=Kr; ;

nloptions tech=nrridg;;

random intercept/subject= CohortID_C;

random intercept/subject=PenID_C (CohortID_C);

random intercept Age /subject= PigID (PenID_C CohortID_C)type=VC;

covtest/wald;

Title' MT5.11.-Tlw72h_rtb Trend ';

format  Nursc Envc HthP Tlw1wk_rtb YN. Nursc4 Rxpc Subjpc Mortpc  YNM.  Gender gender. PigSalmCult Cult. Season Season. PenCat PenCat. FarmID FarmID. TNZ12h TNZ.;

run;

[xtc283]

Basically, this is a convergence issue in maximum likelihood estimation...most likely there is a quasi-complete separation of points where one, some or many of the values for Age fall 100% on one or the other of the values for PigSalmCult.  Try grouping Age into buckets.