[R-sig-ME] over-dispersed Poisson with lmer -- a trick, with a catch?

David Atkins datkins at u.washington.edu
Thu Jul 16 18:21:23 CEST 2009



Ben Bolker wrote:
>   PS -- if MCMCglmm works, why not stop there?  Do you absolutely need
> it to work in lme4?

Ben--

Thanks for the comments (and, yes, I am very hesitant to ever request 
anything from Doug given the Herculean amount of work he has put into 
developing lmer()... suppose I should take this opportunity to learn how 
to build from sources).

As for the comment above, on the one hand, you're right.  We've got a 
solution that works.  Moreover, a lot of our addictions data is 
zero-inflated, and MCMCglmm() can handle that as well (again, thank you 
Jarrod!).

However, I've been using lme/lmer for 12 years and am really comfortable 
with it.  In addition, it's sssssmoking fast, and some of our problems 
get large.  An alternative model with daily drinking reports for past 90 
days on several hundred participants took a couple hours to fit in 
MCMCglmm().  Now, it's great that we can fit the model, but speed does 
count for something.

[Actually, as an aside, my colleague who uses Stata called me while the 
over-dispersed model was running in Stata, "Eh, Dave, this is going to 
take a while in Stata, any chance you could fit this in R?" ;)]

cheers, Dave

Dave Atkins, PhD
Research Associate Professor
Center for the Study of Health and Risk Behaviors
Department of  Psychiatry and Behavioral Science
University of Washington
1100 NE 45th Street, Suite 300
Seattle, WA  98105
206-616-3879
datkins at u.washington.edu




More information about the R-sig-mixed-models mailing list