[R] glmmpql and lmer keep failing

Roel de Jong dejongroel at gmail.com
Wed Nov 2 01:51:27 CET 2005


The word unacceptable was maybe poorly chosen, but I meant it was 
unacceptable for my purposes that a lot of samples needed to be 
discarded. Further, I don't need to give code, data, or other reasons 
why the model may fail because it is well known that PQL can suffer from 
convergence problems, and this issue is already well documented in the R 
help list and articles I skimmed through before posting.  So all I 
wanted to know if how someone else with similar experiences tackled this 
problem, and if there are plans for the implementation of MQL in lme or 
another package.

As of my own contribution, I will soon release a Gibbs sampler for 
multilevel models under the binomial probit link that also can handle 
random slopes.

But I'm sorry I bothered you, guardian of the shining multilevel star.

Roel.





Doran, Harold wrote:
> Sorry to be blunt, but to make the statement that it is unacceptable 
> without providing any reason why the model may fail to converge seems a 
> little presumptious. Your statement really bothers me, especially 
> knowing how hard the developer works on keeping this function at a 
> premier level. Ask anyone working with mixed models and they will all 
> agree that Doug Bates and this function are shining stars in the world 
> of computational statistics.
> 
> R is open source, and therefore you are certainly welcome and encouraged 
> to add to its functionality if you wish. Basically, one may argue that 
> you are in a "put up, or shut up" position. Feel free to write a nice 
> piece of code based on MQL that we can use (and critique).
> 
> But even more importantly, in order to provide any help to you at all, 
> you should provide sample data and examples of your code. We may find 
> that the problem is the user and not the function.
> 
> I generally disagree with rudeness on this list, and agree that my tone 
> is not very agreeable, but please read the posting guide and follow 
> basic protocol by describing your problem, create sample data, 
> illustrate your code, and then ask for help.
> 
> Harold
> 
> 
> 
> -----Original Message-----
> From:   r-help-bounces at stat.math.ethz.ch on behalf of Roel de Jong
> Sent:   Tue 11/1/2005 6:05 PM
> To:     r-help
> Cc:    
> Subject:        Re: [R] glmmpql and lmer keep failing
> 
> Formulated more directly, are there plans for the implementation of the
> crude but more robust Marginal Quasi Likelihood estimation in for
> example LME?
> 
> Regards,
>         Roel de Jong
> 
> Roel de Jong wrote:
>  > Hello,
>  >
>  > I'm running a simulation study of a multilevel model with binary
>  > response using the binomial probit link. It is a random intercept and
>  > random slope model.  GLMMPQL and lmer fail to converge on a
>  > *significant* portion of the *generated* datasets, while MlWin gives
>  > reasonable estimates on those datasets. This is unacceptable. Does
>  > anyone has similar experiences?
>  >
>  > Regards,
>  >     Roel de Jong
>  >
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide! 
> http://www.R-project.org/posting-guide.html
> 
> 
>




More information about the R-help mailing list