I'm using R 2.0.1. I looked in the email archives but didn't see anything on this topic. I've noticed a surprising (to me) difference between AIC and BIC:> methods("AIC")[1] AIC.default* AIC.logLik*> methods("BIC")[1] BIC.gls* BIC.lm* BIC.lme* BIC.lmList* BIC.logLik* BIC.nls* The BIC.gls BIC.lm BIC.lme BIC.lmList and BIC.nls functions appear to have exactly the same definition. (I didn't check with 'diff' though.) Would it make sense to treat BIC similarly to AIC and have just BIC.default and BIC.logLik? Kevin Wright
R does not have a BIC S3 generic function: it is in package nlme! (There is one for S4 classes in package stats4.) So this is a question for the nlme maintainer. On Mon, 18 Apr 2005 kwright@eskimo.com wrote:> > I'm using R 2.0.1. > > I looked in the email archives but didn't see anything on this topic. > > I've noticed a surprising (to me) difference between AIC and BIC: > >> methods("AIC") > [1] AIC.default* AIC.logLik* > >> methods("BIC") > [1] BIC.gls* BIC.lm* BIC.lme* BIC.lmList* BIC.logLik* BIC.nls* > > The BIC.gls BIC.lm BIC.lme BIC.lmList and BIC.nls functions appear to have > exactly the same definition. (I didn't check with 'diff' though.) > > Would it make sense to treat BIC similarly to AIC and have just > BIC.default and BIC.logLik?-- Brian D. Ripley, ripley@stats.ox.ac.uk Professor of Applied Statistics, http://www.stats.ox.ac.uk/~ripley/ University of Oxford, Tel: +44 1865 272861 (self) 1 South Parks Road, +44 1865 272866 (PA) Oxford OX1 3TG, UK Fax: +44 1865 272595
kwright@eskimo.com wrote:> I'm using R 2.0.1. > > I looked in the email archives but didn't see anything on this topic. > > I've noticed a surprising (to me) difference between AIC and BIC: > > >>methods("AIC") > > [1] AIC.default* AIC.logLik* > > >>methods("BIC") > > [1] BIC.gls* BIC.lm* BIC.lme* BIC.lmList* BIC.logLik* BIC.nls* > > The BIC.gls BIC.lm BIC.lme BIC.lmList and BIC.nls functions appear to have > exactly the same definition. (I didn't check with 'diff' though.) > > Would it make sense to treat BIC similarly to AIC and have just > BIC.default and BIC.logLik?Yes. We will do so.