R help - Nov 2009 - Variable selection in NLME or LME4

Good morning

I am learning about NLME and LME4, using Pinheiro and Bates and other materials
from Douglas Bates, but I have not seen anything on how to do variable selection
sensibly in this type of model.

In OLS regression, I frequently use the lasso, but googling did not reveal a
method for lasso with mixed models.

Most of the material I've seen on these packages is about models with very
few potential IVs; but what if we have many? In the problem I am working on now,
I have about 40 potential IVs, and a longitudinal DV with 4 time points.

One idea might be to use lasso on an OLS model of the change score and use the
variables selected in further exploration in LME4 or NLME, but that is just an
idea, I have no real backing for it.

Thanks in advance for any suggestions

Peter

Peter L. Flom, PhD
Statistical Consultant
Website: www DOT peterflomconsulting DOT com
Writing; http://www.associatedcontent.com/user/582880/peter_flom.html
Twitter:   @peterflom

On Wed, 4 Nov 2009, Peter Flom wrote:
> Good morning
>
> I am learning about NLME and LME4, using Pinheiro and Bates and other
materials from Douglas Bates, but I have not seen anything on how to do variable
selection sensibly in this type of model.
>
> In OLS regression, I frequently use the lasso, but googling did not reveal
a method for lasso with mixed models.
>
> Most of the material I've seen on these packages is about models with
very few potential IVs; but what if we have many? In the problem I am working on
now, I have about 40 potential IVs, and a longitudinal DV with 4 time points.
>
> One idea might be to use lasso on an OLS model of the change score and use
the variables selected in further exploration in LME4 or NLME, but that is just
an idea, I have no real backing for it.
>
> Thanks in advance for any suggestions
Peter,

This is rather trickier than model selection in fixed effects models.

See the Vaida and Blanchard article on the _conditional AIC_ (Biometrika 
2005 92(2):351-370; doi:10.1093/biomet/92.2.351)and articles that refer to 
it for more on this.

HTH,

Chuck

>
> Peter
>
> Peter L. Flom, PhD
> Statistical Consultant
> Website: www DOT peterflomconsulting DOT com
> Writing; http://www.associatedcontent.com/user/582880/peter_flom.html
> Twitter:   @peterflom
>
> ______________________________________________
> R-help at r-project.org mailing list
> 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.
>
Charles C. Berry                            (858) 534-2098
                                             Dept of Family/Preventive Medicine
E mailto:cberry at tajo.ucsd.edu	            UC San Diego
http://famprevmed.ucsd.edu/faculty/cberry/  La Jolla, San Diego 92093-0901