Step and stepAIC

Peter-Heinz Fox <peterheinzfox <at> yahoo.de> writes:

I???m using "step" and "stepAIC" for stepwise regression. After each step, 
I would like to make an additional calculation based on the independent
variables that have been selected until this step and
their corresponding weights. Where do I have to add this calculation?

dropTerm is the toolbox function stepAIC uses; you could use it to build 
your own stepAIC version.

But also make sure you have read

http://markmail.org/thread/gk7ysg6tclz4vtdi

and consider some lasso variant method instead (this partly 
answers your second question).

Dieter

An embedded and charset-unspecified text was scrubbed...
Name: not available
URL: <https://stat.ethz.ch/pipermail/r-help/attachments/20090507/a7596301/attachment-0001.pl>

Jun Shen <jun.shen.ut <at> gmail.com> writes:

I also wonder if stepAIC can work with an lme model. I tried. it doesn't for
now. Appreciate any comment.

It can, and it displays a nice error message telling you exactly why it
did not work if you try out of the box. Check parameter "method" in your
lme call. 

Dieter

Jun Shen <jun.shen.ut <at> gmail.com> writes:

I also wonder if stepAIC can work with an lme model. I tried. it doesn't for
now. Appreciate any comment.

It can do it, and it displays a nice error message, as I remember, why
it won't do it in the default setting. Check parameter method of your 
lme call.

Dieter

An embedded and charset-unspecified text was scrubbed...
Name: not available
URL: <https://stat.ethz.ch/pipermail/r-help/attachments/20090507/4c24c393/attachment-0001.pl>

Jun Shen <jun.shen.ut <at> gmail.com> writes:

I changed the "method" to ML for my lme call. Now I get a different error
when running the stepAIC

Error in terms.formula(formula, data = data) :
  'data' argument is of the wrong type

What does that mean?
The same dataset was used for an lm call and everything is fine to run
stepAIC.

Show sessionInfo() (do you have the recent version?)

Show str(yourdataset)

Show lme(.... your formula)

Dieter

An embedded and charset-unspecified text was scrubbed...
Name: not available
URL: <https://stat.ethz.ch/pipermail/r-help/attachments/20090507/5c4a338f/attachment-0001.pl>

Jun Shen <jun.shen.ut <at> gmail.com> writes:

lme(biomarker~Cmax+AGE,data=cov,method='ML',random=~1+AGE|ID,keep.data=T)
Warning message:
In lme.formula(biomarker ~ Cmax + AGE, data = cov, method = "ML",  :
  Fewer observations than random effects in all level 1 groups

That's why the posting guide says: tell the whole story. After
that warning, expect everything. You want too much from your data.


Dieter