Limitations of SAS Enterprise Miner's variable selection method

[frupaul]

Hi everyone,

I was just wondering, does SAS enteprise Miner's stepwise selection methods (Stepwise, Forward, Backward) take collinearity in the model into account? Or does one have to separately handle collinearity first before using the stepwise selection methods?

Thanks,

Paul

[WendyCzika · Accepted solution]

This post from my colleague a few years ago might help:

https://communities.sas.com/t5/SAS-Communities-Library/Tip-How-to-Apply-the-VIF-Regression-Algorithm...

Also, the HP Regression node now outputs a Variance Inflation Factor table when you do a linear regression. The VIF measures how much the variance of a coefficient is increased due to collinearity. A VIF value of 1 indicates that there is no collinearity. Large values, for example 10 or more, indicate that collinearity might be a significant issue in the model.

[frupaul]

Hi Wendy,

Thanks for the answer and article..

Paul

[PaigeMiller]

---

@frupaul wrote:

Hi everyone,

I was just wondering, does SAS enteprise Miner's stepwise selection methods (Stepwise, Forward, Backward) take collinearity in the model into account? Or does one have to separately handle collinearity first before using the stepwise selection methods?

Thanks,

Paul

---

You could use Partial Least Squares regression, which is less sensitive to collinearity than Stepwise/Forward/Backward methods (which also have other drawbacks). With PLS, the issue of variable selection goes away (you use all variables) and you get coefficient estimates and predicted values which have lower mean square error than ordinary least squares regression methods, as shown in http://amstat.tandfonline.com/doi/abs/10.1080/00401706.1993.10485033