topic Re: Scorecard: using WOE as independent variables in SAS Data Science

Scorecard: using WOE as independent variables

Csands — Thu, 14 Apr 2016 15:01:25 GMT

I use logistic regression in SAS Base to develop Credit Risk models. I did some research about Enterprise Miner and the methodologies it uses to develop Scorecard’s (Scorecard node), and notice that the default methodology is to use the WOE in the logistic regression as the independent variables.

Can anyone share with me some documentation about:

the math’s behind the usage of WOE in the logistic regression
the interpretation of the scorecard points when using WOE as independent variable
the relation between using the grouped variable and the WOE

What I found is similar to this SAS doc https://support.sas.com/documentation/cdl/en/emcsgs/66008/PDF/default/emcsgs.pdf and it doesn't answer to my questions, the focus is not on the statistics but in SAS Miner developing.

Thanks in advance for the help

Re: Scorecard: using WOE as independent variables

JasonXin — Thu, 14 Apr 2016 20:43:08 GMT

Hi, I assume you don't have EM yet. Here is except from EM User Guide that hopefully covers your questions. Jason Xin

Re: Scorecard: using WOE as independent variables

Csands — Mon, 18 Apr 2016 09:59:09 GMT

Hi. Your assumption is correct 🙂 , I don't have SAS EM yet. But when trying to understand how do the scorecard done with SAS EM get like this

Re: Scorecard: using WOE as independent variables

Csands — Mon, 18 Apr 2016 10:30:23 GMT

(Sorry I sent the previous message by mistake, and I don’t know how to delete it. So I will continue in this message) When trying to understand how do the scorecard’s done with SAS EM get like the one in attach I found some difficulties. In the scorecards I develop, I always get a model with a constant and in the coefficients for each of the categorical variables I always get one that is zero (I use param=GLM). My models are always something like this: Constant 2.8 Var1_1 0.5 Var1_1 0.3 Var1_1 0 Var2_1 -1.3 Var2_1 0 The doc you sent me will help me to understand why I don’t have a zero coefficient (I still have to study it in more detail). As the coefficients impact with the model constant, that’s probably why the model doesn’t have a constant, but I don’t understand why. Thanks for taking the time to help me C