0

I am analyzing a set of clinical data where I try to predict an outcome by using certain covariates. I have already done univariate analysis and now am progressing to binary logistic regression, incorporating the covariates that have a p < 0.1 in univariate tests to the model. In doing binary logisitic regression, which method is better enter or one of the forward or backward elimination methods?

If I use the enter method, should I manually include and exclude different covariates until all covariates with a final significant contribution to outcome are included in the model?

Which ever the method used, what are the factors/statistical measures to consider in selecting the best method?

Andy W
  • 15,245
  • 8
  • 69
  • 191
cb80
  • 43
  • 2
  • 4
  • 2
    It's not only a repost, with 114 threads already created here under the tag "variable-selection," but it's also a huge and very contentious topic without a single correct answer. And it's not something you'll likely decide once and for all based on a few hours' study. I encourage you to read widely on your question, here and elsewhere. Good luck! – rolando2 Jul 24 '12 at 10:31
  • I second the suggestion from @rolando2. Try searching here for answers by Frank Harrell, as he has written a lot of very useful information on this topic. You could also check out this link (which isn't specifically to so with Stata as the concepts are the same irrespective of your software): http://www.stata.com/support/faqs/statistics/stepwise-regression-problems/ – pmgjones Jul 24 '12 at 17:14
  • 1
    +1 to @rolando2. I basically don't think that people should be using model selection techniques like forward or backward elimination *at all*. You might be interested to read the answer I wrote [here](http://stats.stackexchange.com/questions/20836/20856#20856). For a different legitimate perspective, you might want to read [this answer](http://stats.stackexchange.com/questions/32795/32796#32796), which I think might be reasonable (NB the strong caveats he includes), but might also be dangerous to recommend b/c the technique is inevitably abused. – gung - Reinstate Monica Jul 24 '12 at 18:11

0 Answers0