ABSTRACT
Exact Poisson regression is presented. It can be used when outcomes are too rare to justify normality assumptions for P-values and confidence intervals. Mid-P values are an option. Median unbiased estimates are discussed. They can be useful for perfect predictors. A limitation of exact methods is that they still require that the data come from a Poisson process. This is usually hard to assess when outcomes are rare.
