WARNING: Obtaining minimum variance quadratic unbiased estimates as starting values for the covarian

[Tissue5454]

Hi all, 

I am trying to do logistic regression with random effiect for binary data by GLIMMIX.

However, I get warinig message as follows;

WARNING: Obtaining minimum variance quadratic unbiased estimates as starting values for the covariance parameters failed

Anyone know how I can overcome this problem?

Thanks,

--------------------------------------------------------------------------------

proc glimmix data=logitbank method=quad;
class fid;
model ydat2(event='1') = D1013 fid*D1013
/dist = binary link=logit ddfm=none solution;
random intercept/ subject=fid;
output out =glmout pred =xbeta pred(ilink) =predprob;
covtest;
run;

---------------------------------------------------------------------------------

[Ksharp]

Why you include subject term in the independent variables ?

Try option type=chol.

random intercept/ subject=fid type=chol ;

[Tissue5454]

Hi Ksharp!

Thank you for reply!

>Why you include subject term in the independent variables ?

Because I want to include not only random intercept but also random slope.

>Try option type=chol.

I received same warning message: WARNING: Obtaining minimum variance quadratic unbiased estimates as starting values for the covariance parameters failed...

Tissue5454

[Ksharp · Accepted solution]

Or try PARAM statement . and I quoted here.
"The PARMS statement specifies initial values for the covariance or scale parameters, or it requests a grid search over several values of these parameters in generalized linear mixed models."

[Tissue5454]

Hi Ksharp!

Thank you for reply!

I tried PARAM statement and succeed in getting result!

I think the warning message;

"WARNING: Obtaining minimum variance quadratic unbiased estimates as starting values for the covariance parameters failed" tend to come up when the model is too complicated.

In my case, independent variable "D1013" is nonsignificant .

Thank you again,

tissue5454

[Ksharp]

I am glad that worked and pround of your effort .

[rplum64]

Hey there, tissue5454 - 

Do you mind sharing the syntax of your PARAM fix. I am having a similar code and error log, and would like to know how/where the PARAM fix was inserted. 

Many thanks, in advance.

-R

[Tissue5454]

Sorry... I could not find the above file.

tissue5454

[sld]

Lots of good information in this recent SGF paper by Kathleen Kiernan (2018) of SAS, including possible solutions to the warning message that you are getting

Insights into Using the GLIMMIX Procedure to Model Categorical Outcomes with Random Effects

This older paper by Kiernan, Tao, and Gibbs (2012) is good, too

Tips and Strategies for Mixed Modeling with SAS/STAT® Procedures

[Tissue5454]

Thank you sld !

Your information is valuable for me and rplum64.

These papers containe some examples which use "PARMS" statement.

Thank you again for your help.