topic Re: Weight of the variables in a credit Scorecard or Proc logistic in Statistical Procedures

Weight of the variables in a credit Scorecard or Proc logistic

kumar_ashwini — Wed, 24 Feb 2021 04:28:49 GMT

I am building a scorecard model using logit approach with default flag ( takes value 0 or 1) as dependent variable, and there are 10 independent variables which are a mix of continuous and discrete variables. Business has asked us to fix the weight/contribution of one type(e.g financial factors) of variables at 40% of the total score. Below are the questions where I need your help:
1. How to calculate the weight of each of the variable in proc logistic. I think weight can only be calculated after the score alignment in line with the scorecard methodology. Weight parameter in proc logistic put the weights on the observation and not on the variables.
2. How to put the constraint on the variables/coefficients to fix the weight of one or many variables. In my case I have to put a 40% weight on one variables and remaining 60% on the others.

Re: Weight of the variables in a credit Scorecard or Proc logistic

StatDave — Wed, 24 Feb 2021 04:45:20 GMT

See this note which describes various methods for assessing the relative importance of the predictors in a model.

Re: Weight of the variables in a credit Scorecard or Proc logistic

Ksharp — Wed, 24 Feb 2021 10:58:13 GMT

I don't understand "weight of each of the variable" . It is just the estimated coefficient beta of logistic model.
If you need know the importance of each variable to model ,Try PROC PLS :

proc pls data=class missing=em nfac=2 plot=(ParmProfiles VIP) details;
class sex;
model age=weight height sex;
* output out=x predicted=p;
run;