Hello,
I am doing a logistic regression and we have a small sample (438) with a small number of people with the outcome, or counter outcome. There are 15 predictors. I have moved to doing a FIRTH logistic regression for the problem with quasi complete separation. It solves the problem. However, my issue is I do not know what to do with the output. The betas are not the odds ratios, right?! So what do I do with a 1.088 beta for an outcome that is 1/0? I know I could say that we have validated that there is a statistically significant relationship (pvalue) but that's all I can see. I have not seen a single article that uses FIRTH regression and talks about odds ratios or odds of the event.
The logistic regressions show the effect is approximately and odds ratio of 3:1. I know it is unstable though because of the quasi complete separation and I continue to have gender dropped from the model (Even though SPSS keeps the values in the model--drops the name).
Thoughts?! Any resources are appreciated.
You can take exponential of a beta to get the OR as it is done in ordinary logistic regression. Firth regression is interpreted similarly. You can give CIs of odds ratios etc.
Thank you for your reply. Does anyone have any resources they can point me to for how to use the SPSS output that I get (which is from an R add-on)?
Any resources you have on how to interpret it? All I have found is discussion about how the firth regression is ideal in complete separation.
Thank you again.
SPSS is not the ideal program for firth method.Search for Log Exact software in his 30 days trial version in their website.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS