Variable Selection in Regression: Few Statisticians Know They Produce Poor Models

Replies to This Discussion

Permalink Reply by Steffen Springer on October 27, 2009 at 3:10pm

Hm. In my opinion this statement is too general to be true.

I recently used Logistic Regression in combination with genetic attribute construction and selection. It indeed improved the results significantly (yes, variables were removed).

Despite the well known conditions (do the variables have influence on the target variable, do some of them correlate, beware of overfitting etc. bla bla), how did you come to this conclusion ?

▶ Reply to This

Permalink Reply by Bruce Ratner on October 27, 2009 at 3:42pm

do to: http://www.geniq.net/Variable-Selection-for-Regression-Models-Alway...
sorry I left the article unlinked
Bruce

▶ Reply to This

Permalink Reply by Bhaswati Sengupta on October 28, 2009 at 12:57pm

Is there any difference in Variable selection for a linear regression vis-a-vis a logistics regression model?

▶ Reply to This

Permalink Reply by Bruce Ratner on October 28, 2009 at 1:02pm

Your question: For a given variable-selection method would the variables selected for both OLS and LR be the same. Answer, most probably.

▶ Reply to This