[R-sig-ME] glmer: Downdated X'X is not positive definite

[Thompson,Paul]

Another approach might be to add an additional observation to each singleton cluster, making them doubletons. You could do so by incorporating a known amount of variance, and accounting for that variance in any variance estimate by subtraction from appropriate terms.

-----Original Message-----
From: r-sig-mixed-models-bounces at r-project.org [mailto:r-sig-mixed-models-bounces at r-project.org] On Behalf Of Ben Bolker
Sent: Sunday, February 24, 2013 8:21 PM
To: r-sig-mixed-models at r-project.org
Subject: Re: [R-sig-ME] glmer: Downdated X'X is not positive definite

Ross Boylan <ross at ...> writes:

 [snip]

> Ben Bolker wrote
> > It is indeed a little surprising (to me) that you're getting 
> > downdating problems with an intercept-only model.
> >    Hard to say without a closer look at the data.
> >    I know it won't solve your problem, but does fitting the model 
> > _without_ the singleton clusters work?

  I didn't mean without the clusters, I meant dropping the singleton clusters from the data ...

> glmer complained that there were no random terms, as I recall (not at 
> the
system now).
> glm fit it without complaints.
> 
> SAS GLIMMIX fit a poisson with random intercepts to the data without incident.

  You could try glmmPQL if you like ... (it's not clear whether you used PQL or Laplace approximation within GLIMMIX ...)

> OTOH, glmer did manage to fit poisson random intercept model to toy 
> simulated data I constructed, even though the simulated data had lots 
> of singleton "clusters".
> 
> >    Do you have a very small number of values >0?
> No.  N > 2,000 and c 1/3 are 1's.
> 

  [snip]

  Did you try with development lme4?  Does it make a difference?

_______________________________________________
R-sig-mixed-models at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
-----------------------------------------------------------------------
Confidentiality Notice: This e-mail message, including any attachments,
is for the sole use of the intended recipient(s) and may contain
privileged and confidential information.  Any unauthorized review, use,
disclosure or distribution is prohibited.  If you are not the intended
recipient, please contact the sender by reply e-mail and destroy
all copies of the original message.