when to use exact logistic

[rebekah shao]

Dear all,

I would be grateful for your help.

I have a good sample size e.g. 900, and I am running a logistic regression model with N=87 in my 'successful' outcome. Some of my predictors have very small numbers eg. 6 people with marital breakdown, but only 2 people who have that and my outcome of interest. Should I run exact logistic regression in this case because its small numbers? Or is it ok to run logistic regression in my case because of the overall large sample size even though the outcomes of interest/ predictors have small numbers??

Many thanks

[Marcos Almeida]

Exact logistic regression tends to be used with small samples, perfect or near-perfect predictions, rare events, etc.

This text, written by Richard Williams will be helpful to you.

In short, since the sample size is far above 200. you may think about using the user-written (Joseph Coveney's) firfthlogit.