Reply With Quote
Response rate is not too small. You can still do the logistic regression.
If you want to increase your response rate , you can think of Biased sampling (Over sampling) technique.
Hi,
How can I do logistic regression of Rare event Data? For my data response rate is 3.1%.
Thanks,
Stata
Response rate is not too small. You can still do the logistic regression.
If you want to increase your response rate , you can think of Biased sampling (Over sampling) technique.
In the long run, we're all dead.
Can you pls. send me some good link for Biased Sampling?
Thanks,
Stata
The following link may help
http://www.gotstat.com/tag/oversampling.aspx
In the long run, we're all dead.
Hi,
Thank you for the link. I saw the example there.But I have two doubts.
1. I have 231 events in development sample.So after applying under sampling I'm getting around 470 observations.I have around 10 independent variable. My doubt is that will this method able to estimate the parameters with this much little data?
2. When I am using logistic regression with "Weight-adjusted Model",the result is showing that no beta is significant,But when I'm using the same data with "Offset-adjusted Model", it is giving some betas as significant.This made me totally confused as same data is giving completely different output. Can you please suggest me which method do I need to follow.
Thanks in advance,
1.Here you are trying to keep the response rate as around 50 % and that is the reason you are short of observations. 470 observations are very less for logistic regression.
So you can make response rate around 10% .. so that you will have more observation for model development.
2. There are lot of variation in the Biased sampling procedure in industry. But I see the "Weght adjusted model" is frequently used.
You can check in the Score card development or credit scoring books for more information.
I will also check the same from my end.
In the long run, we're all dead.
Thank you . I will check credit scoring book.
Hi Vinux,
Can you please suggest me some papers for rare event multinomial modeling?
Actually in my data dependent variable has 3 level, and I have 4% observation for First event,73% observation for second event and 23% observation for third event.When I'm doing multinomial modeling with such data set it is overpredicting level 2 ,underpredicting level 3,and not able to predict the level 1.
Please help me.
Thanks in advance
Posting Permissions
|
|