[R] Why does dredge() rank models differently for lmer() and MCMCglmm()?

[Bert Gunter]

(slightly off topic)

One might also add that different model fitting criteria might produce
rankings that are quite different, but with model predictions that are
very similar. This reflects the inconvenient reality that empirical
models are merely data interpolators and not representations of
"truth." Ergo quite different looking models might fit essentially
equally well, the differences in fits and therefore rankings being
nothing more than noise.

Cheers,
Bert
Bert Gunter

"Data is not information. Information is not knowledge. And knowledge
is certainly not wisdom."
   -- Clifford Stoll


On Sat, Jul 18, 2015 at 10:03 AM, Ben Bolker <bbolker at gmail.com> wrote:
> Corina <itsme <at> CorinaLogan.com> writes:
>
>>
>> Hello,
>> I am running my full model (fm) through lmer() and MCMCglmm() using the
>> default settings:
>>
>> model.lmer <- lmer(fm)
>> model.MCMCglmm <- MCMCglmm(fm)
>>
>
>   [snip]
>
>> However, when I run the models through dredge():
>>
>> dredge(model.lmer)
>> dredge(model.MCMCglmm)
>
>> It ranks the models very differently for lmer() and MCMCglmm() in
>> the model selection tables, even though I used the default settings
>> for dredge() as well (ranked by AICc). I would assume the difference
>> is because lmer() uses REML and MCMCglmm() uses Bayes Rule, however
>> if this is the case then why weren’t the summary outputs different
>> as well?
>
>   [snip]
>
>    This question *might* be better on r-sig-mixed-models at r-project.org ...
> You shouldn't be comparing models with fixed effects that are fitted
> with REML -- try  refitting with REML=FALSE.  The model comparison
> functions in the lme4 package (e.g. anova()) try to stop you from
> making this mistake, but I'm not sure the MuMIn::dredge does.
>
> ______________________________________________
> R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.