COMPARING POISSON RATES WHEN DATA ARE SUBJECT TO UNDERREPORTING

Count data are often modeled with the Poisson distribution. Comparison of two Poisson rates can be done using the conditional test based on the binomial assumption. When the counts are underreported, the power of the conditional test can be impacted. We propose a large sample normal test and a small sample parametric bootstrap test based on estimators corrected for underreporting via a double sample. An example is given.