I think it is better to be fully Bayes if you can figure out how
to do so. I've modified Venables and Ripley's stepAIC to use Burnham
and Anderson's AIC.c for stepwise regression, but Ripley is negatively
impressed with that; he referred me to his (1996) Pattern Recognition
and Neural Networks (Springer). I need to modify this further to use an
informative prior, because a limited simulation study demonstrated that
Burnham and Anderson's "Akaike weights" showed an inappropriate
preference against the null hypothesis when zero correlation was
simulated. With normally distributed residuals, a fully Bayesian
solution is feasible, but so far as I know has not yet been programmed
(at least for S-Plus or R). My inadequate attempt at doing so is
downloadable from "www.prodsyse.com", and you are free to use this as
a
starting point if you would like.
hope this helps. spencer graves
Roy Sanderson wrote:
>Hello
>
>I've been investigating using multi-model inference, based on
calculating
>AIC and AIC weights, using the techniques outlined in Burnham and
>Anderson's (2002) book. However I notice a couple of emails in the
R-help
>archive stating that there are errors in the technique. Are these errors
>associated with the particular implementation that B & A propose in
their
>text, or is the whole approach flawed in some way?
>
>Many thanks
>Roy
>
>______________________________________________
>R-help at stat.math.ethz.ch mailing list
>https://www.stat.math.ethz.ch/mailman/listinfo/r-help
>PLEASE do read the posting guide!
http://www.R-project.org/posting-guide.html
>
>