[R-sig-ME] mgcv gam/bam model selection with random effects and AR terms

[varin sacha]

Hi,
According to Kneib & Greven (biometrika 2010)

« the corrected version of the conditional AIC was developed exactly with the goal of allowing for sensible model selection in mixed models. For the marginal AIC we did not find a proper correction, so we would in general not recommend this in its current form. »

« We have recently developed an R package called cAIC4 (https://cran.r-project.org/web/packages/cAIC4/index.html) that should be a good starting point (also beyond Gaussian mixed effects models). »

Best
Sacha Varin

Envoyé de mon iPhone

> Le 18 juil. 2019 à 17:07, Ben Bolker <bbolker using gmail.com> a écrit :
> 
> 
>   I'm not sure of the answer, but in general I'd say if you're
> interested in out-of-sample predictive accuracy, you should try to find
> something analogous to AIC.  R^2/deviance only tell you how well your
> model fits to a specific set of data ...
> 
>> On 2019-07-17 9:46 a.m., Gi-Mick Wu wrote:
>> Dear Mathew,
>> 
>> I was looking for information on model selection for a bam model with an autocorrelation structure and essentially only found your unanswered post (https://stat.ethz.ch/pipermail/r-sig-mixed-models/2017q2/025566.html).
>> 
>> May I ask if you found any solution for this?
>> 
>> Best,
>> Mick
>> 
>> 
>>    [[alternative HTML version deleted]]
>> 
>> _______________________________________________
>> R-sig-mixed-models using r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>> 
> 
> _______________________________________________
> R-sig-mixed-models using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

	[[alternative HTML version deleted]]