topic Not equal volume in Deciles - Logistic Regression in Statistical Procedures

Not equal volume in Deciles - Logistic Regression

sachin01663 — Tue, 23 Jan 2018 09:35:08 GMT

Hello,

I am doing the logistic regression and ranking the final output. In the output, when I rank the observations into deciles, I dont get equal number of obs in each decile. Infact, in some deciles its as low as 35 and highest in a decile is 150. I understand the ranking is done based on predicticted probabilities, but I am wondering is this an indication of wrong prediction or ranking as most of the obs have exactly same prediction score?

Can it be becuase I am using a very few predictors (4 to 5) and all of them are categorical, hence many obs have exactly same predicted probabilities?

Thanks as always.

Sachin

Re: Not equal volume in Deciles - Logistic Regression

PaigeMiller — Tue, 23 Jan 2018 13:17:27 GMT

How are you doing the ranking? Are there ties in the data?

Re: Not equal volume in Deciles - Logistic Regression

ballardw — Tue, 23 Jan 2018 15:43:34 GMT

@sachin01663 wrote:

Hello,

I am doing the logistic regression and ranking the final output. In the output, when I rank the observations into deciles, I dont get equal number of obs in each decile. Infact, in some deciles its as low as 35 and highest in a decile is 150. I understand the ranking is done based on predicticted probabilities, but I am wondering is this an indication of wrong prediction or ranking as most of the obs have exactly same prediction score?

Can it be becuase I am using a very few predictors (4 to 5) and all of them are categorical, hence many obs have exactly same predicted probabilities?

Thanks as always.

Sachin

How are you ranking "the final output"? It might help to show the code used.

Re: Not equal volume in Deciles - Logistic Regression

Reeza — Tue, 23 Jan 2018 15:48:56 GMT

Can it be becuase I am using a very few predictors (4 to 5) and all of them are categorical, hence many obs have exactly same predicted probabilities?

Yes, if you have few predictors that are all categorical you only have a few values.

For example, if you have 3 predictors with 3 levels each, then you have a total of 3*3 = 9 possible predicted values total.

Re: Not equal volume in Deciles - Logistic Regression

StatDave — Tue, 23 Jan 2018 19:36:49 GMT

There is no reason to expect that the decile groups, determined using the predicted probabilities, will have equal sizes. It is not an indication of any problem.

Re: Not equal volume in Deciles - Logistic Regression

sachin01663 — Mon, 29 Jan 2018 15:58:29 GMT

Thanks everyone.

I was expecting the same as I have very few predictors with less levels hence quite a few customers are getting same probability score.

Appreciate your time !!

Sachin