Logistic Regression Model building



I was wondering if it is a good practice to start building a Logistic Regression model using the Backward selection process in SAS. Ultimately SAS will discard the insignificant variables and give us a model with significant variables. This will help us get a sense of the variables that we can concentrate on, given their significance.
Not to say that we just do a backward selection and stop with the iterative model building process. But just to help us get a sense of the significant variables.


If you have data in shape for running models you can try both to validate your question. Over time you will probably understand where to apply which approach.

If you have data in shape i would also recommend running a parallel tree or other classification models that can help you understand significant contributors.

Alternatively WOE or IV can also be used in a way to help with your variable selection.

There is no single approach that will be the best. Whats best with data set 1 may be the worst with data set 2.