Normal-approximation tests for a Poisson exposure model Created 2026-09-24 Updated 2026-11-05
In the Poisson exposure model, write with positive total exposure. Under , the total count has the exact Poisson distribution . Thus an exact upper-tail test chooses the smallest integer satisfying ; discreteness can make its type I error smaller than .
When is large, the normal approximation to the Poisson distribution yields the approximate level- rejection rule
The least-squares estimator instead has null variance , giving
This second approximation needs a central limit theorem for the weighted sum , such as a suitable Lindeberg condition. A large unweighted total exposure alone does not establish that weighted approximation. Both displayed thresholds use the null variance, rather than a response-dependent plug-in standard error.
Both alternatives have larger , so use upper-tail rejection regions. Let be the 95th quantile of the standard normal distribution.
For the likelihood estimator,
When is large, the normal approximation to the Poisson distribution gives, under ,
An approximate size- test therefore rejects when
When every is large, independently. A linear combination of independent normal variables is normal, so under ,
The corresponding approximate size- test rejects when
In both cases the null rejection probability is approximately , while values near the alternative mean increasingly fall in the rejection region as the total exposure grows. These are the normal-approximation tests for a Poisson exposure model.