I have a question about the package leaps which I am using for model selection.
I would like to compare 4 different selection methods: forward, backward, stepwise and best subset. I used the code below: library(leaps) forward <- regsubsets(Response ~.,data = mydata, method = "forward", nbest=1) backward <- regsubsets(Response ~.,data = mydata, method = "backward", nbest=1) stepwise <- regsubsets(Response ~., data = mydata, method = "seqrep", nbest=1) best subset <- regsubsets(Response ~.,data = mydata, method = "forward", nbest=1) opt = par (mfrow =c(2,2)) plot(forward, scale = "adjr2", main = "Forward Selection") plot(backward, scale = "adjr2", main = "Backward Selection") plot(stepwise, scale = "adjr2", main = "Stepwise selection") plot(best subset, scale = "adjr2", main = "Best subset selection") Using these commands I obtained figures below: <http://r.789695.n4.nabble.com/file/n4674451/leaps.png> I am wondering why figure A and D are similar to each other (and also figure B aand C). I would expect different algorithms to select models in a different way. For instance models selected with forward selection method should be chosen based on the significance level/ AIC value. On the other hand models selected with best subset selection method should be chosen based on the sample satisitcs. I am also wondering why forward selection does not choose one variable at the time adding it to the exisitng model ? Also Fig B shows that backward selection starts with eight variables in the model. Why it does not start with all the variables and excludes one at the time ? I would be very grateful for these clarifications. -- View this message in context: http://r.789695.n4.nabble.com/Selection-of-regressors-tp4674451.html Sent from the R help mailing list archive at Nabble.com. ______________________________________________ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
