Subscribe to DSC Newsletter

Why Propensity Scores not working in Campaign Analytics??

Hi All ,

I've recently  build a predictive data model for a campaign to target  group of customer who can switch from  payment method  "A" to Payment method "B".

The idea was to come up with predictive model with payment/usage variables as independent variable and  the event of switch in past is taken as  dependent variable.

i) I used logistic regression model for probability  estimation.

ii) The  data  is sorted in order of descending probability and  customer's were solicited. 

iii)  The results analysed after campaign suggests , the customer have low probability have better switch rate and high probability have  lower switch rate.

iv) Below are the results:

Decile Control Test Total
No Yes No Yes  
1 2784 5 2450 7 5246
2 2871 9 2191 6 5077
3 2861 5 2145 6 5017
4 2795 11 2851 12 5669
5 2924 11 1060 5 4000
6 3018 17 3153 14 6202
7 3200 27 2781 11 6019
8 3142 22 2145 4 5313
9 3079 28 3142 16 6265
10 3161 31 3078 13 6283
Grand Total 29835 166 24996 94 55091

I assumed the customer with high propensity should have high switch rate , so the switch rate in top  decile should have been higher.

Could anyone please suggest

 i ) what mistakes I could have done.

ii) Any similar experience and remedy .

Regards,

Himanshu

Views: 835

Tags: analytics, campaign, decile, model, not, predictive, propensity, socres, working

Comment

You need to be a member of AnalyticBridge to add comments!

Join AnalyticBridge

Comment by Wayne G. Fischer, PhD on January 3, 2015 at 7:37am

Himanshu, if you have set up your model variables correctly, it may something as simple as having the "0" (won't switch) and "1" (will switch) conditions reversed.  In SAS Institute's JMP the user may define which is which.

On Data Science Central

© 2019   AnalyticBridge.com is a subsidiary and dedicated channel of Data Science Central LLC   Powered by

Badges  |  Report an Issue  |  Privacy Policy  |  Terms of Service