Glimmix repeated measures

swuga — Tue, 26 Mar 2019 16:46:37 GMT

Hi,

I have a problem with repeated measures using Proc Glimmix and wonder if anyone could help me look at it and find a solution. Here is my code:

Proc glimmix data=mortality plots=residualpanel;

class insect treatment rep time;

model mort=insect|treatment|time rep/ d=beta;

random intercept / subject=rep;

random time / residual subject=insect*treatment*rep type=AR(1);

lsmeans insect*treatment*time/ slicediff=(insect treatment time) Adjust=Tukey lines plots=meanplot (cl join);

run;

I am looking at the proportional mortality (mort) of two types of insects at various treatments (5 treatments) at repeated observations (3 time points). Each treatment had 3 experimental units; the experiment was run in 2 repetitions (rep as block). I would like to know the effect of insect, treatment, time, and interactions. I use beta distribution.

The model works without Random statements. But after adding Random for repeated measures, it first showed that "WARNING: Obtaining minimum variance quadratic unbiased estimates as starting values for the covariance parameters failed".

I realized PROC GLIMMIX can only have one observation per level of the repeated effect for each subject, and I have 3 experimental units for each treatment, a total of 2 insects * 5 treatments*2 repetitions*3 experimental units=60 units for each of the 3 observation times. Thus I added a column to label each unit (1-60 for each observation time) and changed the code as:

Proc glimmix data=mortality plots=residualpanel;

class insect treatment rep unit time;

model mort=insect|treatment|time rep/ d=beta;

random intercept / subject=rep;

random time / residual subject=unit type=AR(1);

lsmeans insect*treatment*time/ slicediff=(insect treatment time) Adjust=Tukey lines plots=meanplot (cl join);

run;

Then, there was an error "NOTE: Did not converge". I wonder what step was wrong and how I could fix it.

Many thanks!

Re: Glimmix repeated measures

Rajesh3 — Tue, 26 Mar 2019 19:30:39 GMT

Hello,

You need to be cautious when increasing the random parameters in the model. Based on the error that you got, the variance-covariance matrix might be showing negative values for variance or 0 for covariance. This means a simpler model (without a random intercept/linear trend) is a better alternative in your case. Also, I don't understand why you used 'rep' (i am assuming this to be the id variable) in the model as a covariate.

Thanks,

Rajesh.

topic Re: Glimmix repeated measures in Statistical Procedures

Glimmix repeated measures

Re: Glimmix repeated measures