Hi!there. How to perform Stepwise Regression in R or STATA?
Hello Dewan Azmal Hossain. In case you do not already know, please note that there are many problems associated with stepwise regression (and automated, algorithmic variable selection in general). This Stata FAQ item summarizes things nicely.
- https://www.stata.com/support/faqs/statistics/stepwise-regression-problems/
See also Mike Babyak's (2004) nice article on overfitting and the Vanderbilt Biostats Manuscript Checklist section on stepwise variable selection:
- https://people.duke.edu/~mababyak/papers/babyakregression.pdf
- http://biostat.mc.vanderbilt.edu/wiki/Main/ManuscriptChecklist
HTH.
https://www.youtube.com/watch?v=66z_MRwtFJM
https://www.youtube.com/watch?v=2Sb1Gvo5si8
https://www.youtube.com/watch?v=F6m8I1RBWbg
https://www.youtube.com/watch?v=HafqFSB9x70
Hello Dewan Azmal Hossain. In case you do not already know, please note that there are many problems associated with stepwise regression (and automated, algorithmic variable selection in general). This Stata FAQ item summarizes things nicely.
- https://www.stata.com/support/faqs/statistics/stepwise-regression-problems/
See also Mike Babyak's (2004) nice article on overfitting and the Vanderbilt Biostats Manuscript Checklist section on stepwise variable selection:
- https://people.duke.edu/~mababyak/papers/babyakregression.pdf
- http://biostat.mc.vanderbilt.edu/wiki/Main/ManuscriptChecklist
HTH.
As Bruce Weaver says, there are lots of issues with the stepwise approach, particularly the automated versions based on things like shifts in R^2, but if that is what you want and you are using R, the leaps package ( https://cran.r-project.org/web/packages/leaps/leaps.pdf ) finds the set of predictor variables with the "best" fit according to the criterion you choose. If you want an automated procedure lasso regression has benefits over this, but in most cases unless you have tons of predictors and not much knowledge about how they relate, you should based your models on theory.
One of the new features in Stata 16 is a set of lasso tools:
- https://www.stata.com/new-in-stata/lasso/
- https://www.stata.com/features/lasso/
If you have an older version of Stata, there are user-contributed packages you can use. E.g., see this page:
- https://statalasso.github.io/
HTH.
Further info on the lasso is also available at https://web.stanford.edu/~hastie/StatLearnSparsity/ , which can be downloaded there. The book mostly covers information already published in their papers, but it is nice to have it summarized in a uniform bookish way.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression