Automated variable selection

[Melk]

Hi All,

I am used the automated variable selection (stepwise method) with a SLE of 0.2 and a SLS of 0.05. I am getting variables in my final model that are >0.20 in my univariate analysis. I am wondering why that is, is it because those variables are significant in the presence of other variables and they get entered in one of the middle steps?

[Reeza]

Yes. This is why the univariate method of variable selection is not considered a reliable or acceptable method to determine which variables to include in your regression model.

[Melk]

Is this what they call negative confounding?

[PaigeMiller]

There are a lot of criticisms of stepwise, I assume this is one of the 12,631 reasons why you shouldn't use Stepwise. If you go to your favorite internet search engine and type in "problems with stepwise regression", you will find enough to read for a week.

So, don't use stepwise.

[Melk]

Hm. What do you normally use? LASSO?

[PaigeMiller]

I use Partial Least Squares regression (PROC PLS), which works well in the presence of multicollinearity. According to this article, "Partial least squares (PLS) is a method for constructing predictive models when the factors are many and highly collinear."