[R] How to compute an exact Poisson model with R?

[David Winsemius]

The point estimates for regression should be the same although the CI's for the MCMC methods might be wider (more conservative if that is what you mean by "more accurate"). I 

I was able to install Brian Caffo's archived 'exactLoglinTest' package for my MacOS system with R 4.6.1 after adding:


CPPFLAGS += -I/Library/Frameworks/R.framework/Versions/4.5-arm64/Resources/include

to my .R/Makevars file. That allows the compilation step to find PrtUtil.h from an earlier version of R since it has been folded into R 4.6.x rather than currently being exposed.

The Notes that Caffo chose not to fix did not cause any errors, at least to my minimal testing with the package's examples.

Best;

David Winsemius, MD, MPH

> On Aug 21, 2026, at 1:21 AM, iagogv via R-help <r-help using r-project.org> wrote:
> 
> Hi all,
> 
> Can an exact Poisson regression be computed with R, instead of using the 
> standard (glm) Poisson using maximum likelihood?
> 
> I am working with a collegue in a project, and I'd like to reproduce his 
> method, if not his results. He uses Stata, particularly the expoisson 
> function (https://www.stata.com/help.cgi?expoisson) in order to compute 
> the exact Poisson regression, relevant for small samples.
> 
> expoisson fits an exact Poisson regression model, which produces more 
> accurate inference in small samples than standard 
> maximum-likelihood-based Poisson regression
> 
> I know about exact Poisson tests 
> (https://stat.ethz.ch/R-manual/R-devel/RHOME/library/stats/html/poisson.test.html), 
> but it does not allow to include offset.
> 
> So, is there some way to reproduce exact Poisson regression with R?
> 
> Thanks!
> 
> Best,
> 
> -- 
> Iago
> [[alternative HTML version deleted]]
> 
> ______________________________________________
> R-help using r-project.org mailing list -- To UNSUBSCRIBE and more, see
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide https://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.