Subscribe to DSC Newsletter

Hi,

I have few doubts regarding null hypothesis. Can anyone help me to find the answer?

Q. Your risk analysis team has access to new customer financial data. You want to use this data to improve your prediction of credit default.Previously, the team was using only credit bureau scores,loan size, and customer income to assess risk of default.

What is the null hypothesis that should be used to evaluate the model?

a.Model using the new financial data predicts the outcome just as well as the previous model.

b.New model predicts better than the toss of a coin weighted by the average default rate.

c.Model using the new financial data predicts the outcome better than the previous model.

d.New model predicts as well as toss of a coin weighted by the average default rate.

Thanks.

SS

Views: 799

Replies to This Discussion

Hi Sasmita, I also have some doubts about the choices provided.  Without looking at them I would have posed a null hypothesis in this case as something like 'The model using the new financial data predicts the outcome no better than the previous model' (i.e. its discriminatory ability is not significantly better than could have been achieved by chance).   This statement addresses the issue at hand also accepting the potential for new model to predict the outcome no better than chance, which of course would be a typical null hypothesis, largely irrelevant to the current goal. Best wishes, Ron.

If I fail to reject the Null Hypothesis I am basically saying that I can't tell if A is different from B, therefore I am going to assume that A=B (for all practical purposes)

If I reject the Null I then accept the alternative hypothesis that says that A is not equal to B.  Selection b and c read more like alternative hypothesis.

Selection a and d seem like potential Null hypothesis, but based on the fact that your question is talking about new data and a new model,  I would say that you are trying to determine if the new model is no better than the old one....which would indicate a. as the answer.

you can then calculate the RMSE of each model (or whatever statistic your problem is asking for) and then determine if the new model is statistically different/better from the old model.

If your test statistic says that the new model is better than the old one, you can reject the Null Hypothesis that says the two models are the same, and accept the alternative hypothesis that says the new one is better.

When I had to pound this concept into my head I thought that it was kind of a silly and hard to understand way towards looking at problems.......Since then I have lost track of the number of times I've employed this basic concept into my day to day work....

RSS

© 2019   Data Science Central ®   Powered by

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