[R-sig-ME] comparison between glmm.admb and lme4

Jarrod Hadfield j.hadfield at ed.ac.uk
Sat Nov 27 09:18:53 CET 2010


Hi Brian,

Likelihoods are not necessarily comparable between programs, but  
differences in log-likelihoods between two models should be (i.e.  
likelihoods from different programs should be proportional). For mixed  
models with non-Gaussian data the likelihood has to be approximated so  
programs that use different approximations will give different  
answers, but I guess you would hope the differences are not large.

Cheers,

Jarrod

  Quoting Brian Burke <Brian.Burke at noaa.gov>:

> Hi all,
> I have a situation where I will be using glmm.admb for some analyses  
> (e.g., when I need a negative binomial distribution or zero  
> inflation) and lme4 for others (e.g., when I have crossed random  
> effects).  I would like to be able to compare the log likelihoods  
> for all of the results, regardless of the package I use.  I decided  
> to run a model from each package on the same data set, to see if  
> they resulted in the same (or at least similar) log likelihoods.   
> The data are catch of salmon in over 1100 trawls, many of which are  
> zeros.  Results are below.  Both models result in parameter  
> estimates that seem reasonable (and they are similar, but not  
> identical, to each other).  Did the glmer really fit that much  
> better or are the log likelihoods not technically comparable?
>
> # Compare  glmm.admb and lme4
> pois.admb <- glmm.admb(count ~ temp + I(temp^2) + depth + secchi +  
> chl + month + year, random=~1, group="site", data=stations,  
> family="poisson")
>
> pois.glmer <- glmer(count ~ temp + I(temp^2) + depth + secchi + chl  
> + month + year + (1|site), data=stations, family="poisson", REML=F)  
> # I get the same nll whether I put in the REML statement or not
>
>
> # Results in:
>> logLik(pois.admb)
> 'log Lik.' -2550.45 (df=NULL)
>> logLik(pois.glmer)
> 'log Lik.' -2474.079 (df=20)
>
> Thanks!!
> -Brian
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>
>



-- 
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.




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