Dear all,
I am working on a logistic model where the dependent is a binary variable and independent variables are a collection of continuous and categorical. After removing highly collinear independent variables, outliers i run stepwise models using SPSS. The models i get using forward and backward stepwise models are all okay with hosmer-.. and have high explanatory power. But they have different independent variables?
I would like to ask if it is okay to have different models since hosmer and explanatory powers are okay. Does having two different models a problem despite good model statistics?
Note: The first data set has 1350 observations, the second has 250.
Thanks
Relying solely on stepwise methods is not recommended. You can use subject matter expertise to form a model (namely to decide if a variable is included or not). There are many threads about this topic in RG. I added two links that I could have found.
BTW, you seem to have different dataset. Do you expect to find the same model from those datasets by using stepwise procedures?
I can recommend LASSO for variable selection if subject matter expertise is limited.
https://www.researchgate.net/post/Logistic_Regression_Method
https://www.researchgate.net/post/How_to_continue_after_stepwise_regression
Dear Murat,
Yes, it is ok to arrive at two different viable models. You are carrying out a modelling process of a complex data set so there is often no one right answer. There are a lot of factors to consider in binary logistic regression. Provided you are aware of these and can justify the decisions you have made then presenting two models is completely acceptable.
Dear Mehmet and Peter,
Thank you very much for your targeted responses. They helped a lot!
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS