Multilevel analysis -SAS


No cake for spunky
If that doesn't work, read the following, if you haven't already. I remember coming across this back in the day. In particular the example about variability across classrooms and using repeated instead of random option. Though I would first mess with the variance/covariance structure. Even though Jake said he would probably not mess with the convergence criteria, I would perhaps tweak it a little to let the model run a little longer.

Lastly per the classroom example, you may need to also accept there isn't much variation explained by the group variable that isn't picked up in a simple model. Also what is your group variable? is it a geographic location, if so, locations may be getting separated but they are tangential and actually more similar than believed.

I accidently edited you comments rather than quoting them.... :( my apology. I think I put it back in its original form.

My group is a geographical unit. The ICC was 3.7%, I don't know if that is a lot, little or whatever.... :p

My random effects do not show the problem with the Hessian matrix. Only the estimate of the fixed effects.


Less is more. Stay pure. Stay poor.
You are controlling for covariance in the region, though there is also covariance between tangential regions.


No cake for spunky
I have a question. I run the following
proc mixed data=work.test4 method=ml covtest empirical
noclprint ;
class unitid_pri female ;
model dv=female/ ddfm=contain s ;
random intercept /subject=unitid_pri type=ar(1) s ;
parms / ols;
ods output FitStatistics=fm1 SolutionF=SFfm1 ;

and get

Convergence criteria met but final Hessian is not positive definite.

When I run
ods graphics on;
proc mixed data=work.test4 method=ml covtest empirical
noclprint PLOTS(MAXPOINTS= 12000 );
class unitid_pri female ;
model dv=female/ ddfm=contain residuals s ;
random intercept /subject=unitid_pri type=un s ;
parms / ols;
ods output FitStatistics=fm2 SolutionF=SFfm2 ;
ods graphics off;

the problem goes away (I changed the covariance type that was all).

This leads me to suspect that the Hessian matrix is positive definite in both runs, but artificially this is not the case in the first run (because one author suggested that in some cases the code calculates two identical variance terms one of which has an effect size of zero and this causes the warning, which in this case is not meaningful - does not influence the results).

Anyone know if this is true? Also when is it justified to the use the type UN rather than AR(1) - which the documentation I have found does not really cover.


No cake for spunky
I can pop up for many reasons, some serious some not. If the warning is not tied to something entirely artificial (like estimating the same variance twice in different places) then the results you generate are not valid (or at least you don't know if they are or not).

Different scale between variables is one of the reasons it shows up, I forget the others.