0

As the title suggests, suppose I want to enter three blocks of variables --according to a theory that I am testing-- and then on the final step, I want to enter few additional predictors (not from the theory), using Stepwise procedure. Can I do that?

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
Simon
  • 21
  • 2
  • 2
    "Can I do that" as can I do that in the software I use or can I do that and still being "statistically kosher?" – Penguin_Knight Feb 05 '14 at 04:21
  • 1
    No doubt you *can* ... but that doesn't mean you *should*. After all, I *can* use a nail gun to try to mend my boots ... while I'm wearing them - the difficulty is it's hard not to shoot yourself in the foot. CV abounds with posts in which reasonably knowledgeable statisticians warn against such an approach. It's worth searching on 'stepwise' (and maybe also on 'all subsets regression', 'backward elimination' and 'forward selection') to see many warnings. That's not to say *never* ... but you should probably learn more about the likely consequences before pulling that trigger. – Glen_b Feb 05 '14 at 05:42
  • 1
    Why not include all the predictors that you're considering selecting using a stepwise procedure? Is the full model overfitting? Do you need a smaller subset giving reasonable predictive performance because it'll be expensive to measure all of them in future? – Scortchi - Reinstate Monica Feb 05 '14 at 11:44

0 Answers0