-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
On 06/16/2011 01:47 PM, Rob James wrote:> Ben,
>
> Thanks for this. Very helpful and clearly others have tripped over the
> same problem
> I would have supposed that the solution was to ask lrm (or glm) to use
> LR rather than Wald, but I don't see syntax to achieve this.
Typically drop1 or dropterm (MASS package) will drop appropriate terms
from the model and test the difference via LRT (or F test). stepAIC in
the MASS package will do stepwise selection via AIC.
This opens the larger can of worms of why you're doing stepwise model
selection in the first place ... I was surprised to see fastbw() in the
rms package, since Frank Harrell is normally a vociferous opponent of
stepwise approaches, but I assume (my copy of his book is not handy)
that there is some context within which it makes statistical sense.
>
> My challenge is further complicated because I had hoped to use fastbw()
> to generate a parsimonious model, but clearly fastbw also uses Wald, and
> again I see no way to modify this to use LR.
>
> Hmm....
>
> Many thanks,
>
> Rob
>
>
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.10 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/
iEYEARECAAYFAk36RGIACgkQc5UpGjwzenM9jgCfaZLSRdeWnppWpGOlb/BZz/UO
PZAAniF69cKw7qIBI1VJh4+y8pNHfL4T
=X+NT
-----END PGP SIGNATURE-----
