multiple regression and adjusted r-squ

smurf

New Member
#1
Hi there,

I just found this forum in a fit of frustration and angst...

Anyway, I'm running a multiple linear regression in SAS for a final project in my SAS computing class. I'm supposed to use backward selection to find the final model. One of the variables has a high p value (.1-ish) for the beta estimate, so I removed it. When I ran the model again without the variable in question, though, the adjusted r-squared actually deceased (only slightly decreased, but still...).

I had thought that when you remove a variable that is not significantly associated from the model, the adjusted r-squared is supposed to increase. Am I mistaken? What is going on, here? This doesn't mean that I should keep the high p-value variable in the final model, does it?

Sorry if this topic was covered in another post that I may have not found. Just give me the link to that thread if that's the case.