3
$\begingroup$

Ok, it is known that stepwise LR isn't valid, and forward LR fails to capture the combination of the variables (sometimes a independent variable will become significant only in the presence of the other variables), when backward elimination resolves very well this kind of problem.

So is it true when we say backward is the best strategy between the three methods?

Improve this question
$\endgroup$
1
  • $\begingroup$ Stepwise regression strategies are basically rustic relics of a bygone age. We have better technology now, the only reason to use stepwise methods is a stubborn refusal to embrace modernity. $\endgroup$
    – Sycorax ♦
    Commented Sep 17, 2015 at 18:07

1 Answer 1

Reset to default
10
$\begingroup$

Backward may be better than forward or stepwise, but only in the sense that it's a lower dose of the same poison. When you do any of these methods, many of the results of the regression will be wrong. I do recall seeing some material saying they are less wrong with backward, but, since there are widely available methods (e.g. LASSO) that are better than any of these three, I don't see why you would use backward.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.