jCeradini wrote:Very recent comment/elaboration from Burnham on summing model weights and relative variable importance. He does couch this approach within the context of all subsets model selection, which is interesting (and also how it is presented in B&A 2002).
http://warnercnr.colostate.edu/~kenb/pd ... urnham.pdf
(posted on his website: http://warnercnr.colostate.edu/~kenb/)
Joe
Been circulating for some time -- good to advertize it. Definitely a step forward.
But, couple of caveats (which I've been discussing with Ken on and off):
1\ Doesn't handle interactin terms (either of continuous covariates, or factor/covariate). As ken obtusely puts it -- "Inferences could change if interaction terms are included". I'd say that was an understatement.
2\ assumes that regressors are strictly orthogonal
3\ assumes that factors and continuous covariates all modeled as, continuous covariates.