Subscribe to DSC Newsletter

Hi,

I've got a Logistic model built for a particular response-non response event.

The model suggests statistics that don't look like a robust model. I'm sharing those for more clarification..

No. of variables - around 5-8
c = 0.9
concordance = 0.93
H-L Chi square (Goodness of Fit)= 700 (P 0.0001) (rejects Null - bad model characteristics)

Also, a univariate distribution of P(Y=1|X1..Xn) gives me 95% of the probabilities fall within 0.4!!! Which suggests that the model does poorer than a random!!

What are the ways to improve my model? I know of one or two methods that I surfed through recently, but none hands on.. Would like to hear any advice on this!

Thanks in advance.

Arun

Tags: Logistic, Modeling, Models, Predictive, Regression, Robust

Views: 137

Reply to This

Replies to This Discussion

Well, apparently, I've myself been able to make some in-roads into this question that has led to a few answers. Putting them up here for reference...

With Concordance as high as 0.9, the model seems to be near perfect in predicting the '1's'! Yet, the Goodness of Fit stats convey that it is nowhere near to a robust model! All this indicated in checking out some unlikely independent variable contributing to this effect.

RESULT - We ended up finding one variable that seemed to have 99% of non-responders within say 0.3, while almost 90% of redeemers had cutoff greater than 80-90% of non-redeemers!
Hence this variable was biasing the concordance to be that high, yet the model was highly unstable!

Currently, looking for alternate routes to arrive at a solution...

In the meanwhile, my question on
"a univariate distribution of P(Y=1|X1..Xn) gives me 95% of the probabilities fall within 0.4!!! Which suggests that the model does poorer than a random!!"
still needs clarification...
With Concordance as high as 0.9,[url=http://www.edhardykleidungshop.com/]billig ed hardy kleidung[/url] the model seems to be near perfect in predicting the '1's'! Yet, the Goodness of Fit stats convey that it is nowhere near to a robust model! All this indicated in checking [url=http://www.edhardykleidungshop.com/]billig ed hardy schuhe[/url]out some unlikely independent variable contributing to this effect.
RESULT - We ended up finding one variable that seemed to have 99% of non-responders within say 0.3, while almost 90% of redeemers had cutoff greater than[url=http://www.edhardykleidungshop.com/]ed hardy shop[/url] 80-90% of non-redeemers!
Hence this variable was biasing the concordance to be that high, yet the model was highly unstable!

RSS

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