If stepwise regression is being considered then there are not likely to be solid hypotheses for consideration using a model comparison approach and either approach is fundamentally going to be exploratory. Based on my reading of the literature model selection is only going to be more accurate (assuming you mean in regards to parameter estimates) if based on a priori models. Otherwise there are still going to be the same problems with estimates, and--if any critical threshold is used--the same problem with spurious effects. Interestingly some of the discussion of stepwise versus model comparison in the ecological literature has been done based on comparison of all possible models which has the same problems and is equally exploratory in nature. Ned -- Ned Dochtermann Department of Biology University of Nevada, Reno ned.dochtermann at gmail.com http://wolfweb.unr.edu/homepage/mpeacock/Ned.Dochtermann/ -- Today's Topics: 1. Re: Questions - the stepwise selection issue. (Andrew Kosydar) 2. quasi-binomial family in lme4 (T. Florian Jaeger) ---------------------------------------------------------------------- Message: 1 Date: Sun, 7 Nov 2010 20:22:28 -0500 From: Andrew Kosydar <drewdogy at uw.edu> To: John Maindonald <john.maindonald at anu.edu.au> Cc: r-sig-mixed-models at r-project.org Subject: Re: [R-sig-ME] Questions - the stepwise selection issue. Message-ID: <AANLkTikP0JMnChSkWi01Or0dNzcR4ZkVvweYEUfUvuOm at mail.gmail.com> Content-Type: text/plain; charset=windows-1252 Hello All, Instead of using step-wise selection, I would suggest instead using multimodel inference (Burnham & Anderson). The technique avoids having to choose one "right" model and, in my opinion, is a more accurate method than traditional step-wise procedures. Cheers! Andrew
Andrew Kosydar, PhD drewdogy at uw.edu (206) 669-0505 On Sun, Nov 7, 2010 at 7:17 PM, John Maindonald <john.maindonald at anu.edu.au> wrote: > The stepwise model reduction issue is an interesting one. > > My view is that: > 1) One should always begin by looking at the t-statistics for > the coeffs in the full model (assuming that this is a situation > where they are more or less believable!). ?If there is a clear > division into those that are significant and those that are > clearly not significant (p-value > 0.1, maybe), then drop > those that are not significant. ?Check what difference this > makes to the residual SE, and to the coefficients (any large > changes may matter if there is an interest in interpreting > coefficients). ?There are other issues to consider; are some > variables of such scientific consequence that they should > be retained regardless? *******