[R-sig-ME] Model selection and averaging

Ben Bolker bolker at ufl.edu
Tue Dec 8 16:01:00 CET 2009


  How different are your results?  Are you using Laplace or AGQ in lmer?
Are the results really very different? (Two possibilities that aren't
"really different": (1) p-values on either side of a magic p=0.05 line
but numerically reasonably close; (2) estimates that are biologically
(or whatever) significantly different, but with confidence intervals on
each that suggest we can't really tell them apart.)
In one small example
<http://glmm.wdfiles.com/local--files/examples/culcita_glmm.pdf> I got
similar results between the two.

  More details?

   Ben Bolker

Marcio M. de Morais Jr. wrote:
> Thank you Luca
> 
> Reading that I interpreted you could choose REML or ML for lmer, but
> glmer would
> use REML. I thought that, mainly because I have got different results
> fitting the same model in glmer and glmmML and differences in REML and ML
> could be the reason.
> 
> Cheers,
> 
> Marcio
> 
> 2009/12/8 Luca Borger <lborger at uoguelph.ca>
> 
>> Hello
>>
>>> is there any
>>> package capable to do model selection using a glmm object fitted by ML?
>> unless I got this wrong, GLMMs are fitted by ML. If you type
>>
>> ?lmer
>>
>> after loading the lme4 library you can see that it says:
>>
>> "REML logical argument to lmer only."
>>
>>
>> HTH
>>
>>
>> Cheers,
>>
>> Luca
>>
>>
>> ----- Original Message -----
>> From: "Marcio M. de Morais Jr." <moraisjr at gmail.com>
>> To: r-sig-mixed-models at r-project.org
>> Sent: Tuesday, 8 December, 2009 13:37:28 GMT +01:00 Amsterdam / Berlin /
>> Bern / Rome / Stockholm / Vienna
>> Subject: [R-sig-ME] Model selection and averaging
>>
>> Dear list members,
>>
>> I am using glmmML package to access the effects of variables on
>> presence-absence data set.  I intend to use IC model selection and
>> averaging, and have looked for a package to do it. I have found none that
>> can work with glmmML objects, but MuMIn works with glmer. Since my models
>> are only different in their fixed effects, I would like to know if it is
>> correct to use model selection for glmer objects that are fitted by REML. I
>> have read that models fitted by REML can be compared only if they are
>> different in their random effect. Is that right? If it is, is there any
>> package capable to do model selection using a glmm object fitted by ML?
>>
>> Thank you very much for your consideration
>>
>> Marcio de Morais
>>
>>         [[alternative HTML version deleted]]
>>
>> _______________________________________________
>> R-sig-mixed-models at r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>>
> 
> 
> 


-- 
Ben Bolker
Associate professor, Biology Dep't, Univ. of Florida
bolker at ufl.edu / www.zoology.ufl.edu/bolker
GPG key: www.zoology.ufl.edu/bolker/benbolker-publickey.asc




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