Hi, I am currently working on clinical data in which the some of the variables were cannot be analysed using conventional logistic regression method due to limited number of reported cases to estimate (rare event). Therefore, I tried to analyse using exlogistic and firthlogistic. However, I cannot progress to variable selection as the conventional logistic method is not working on exact logistic regression. Is the anyway I can perform the multivariable analysis for exact logistic regression?
Thank you very much
exlogistic and firthlogistic may not support the command stepwise in STATA. In any case, I would not recommend using automatic stepwise for variable selection!
http://www.stata.com/support/faqs/statistics/stepwise-regression-problems/
You may want to consider relogit, which is a user-written program for Stata by Gary King. It is specifically designed for rare events. I agree with Lapo that stepwise approaches to variable selection are not recommended. Instead, you should first and foremost rely on content expertise.
Model selection of logistic regression
I was users of SAS, SPSS and Statistica. Now, I use JMP. Therefore, my answer is restricted within JMP. JMP support all possible combination and stepwise procedures. But, JMP does not support those.
And although SAS support Firth and exact logistic regression, JMP support only Firth. But I can analyze many linear separable data by Firth. I do like to know the differences of Firth and Exact.
I evaluated seven linear discriminant functions (LDFs) such as logistic regression, Fisher's LDF, H-SVM, S-SVM and my 3 LDFs using over 10 data sets. The number of independent variables are from 4 to 19 variables. I evaluated 7 LDFs by all possible combination (P