Exact critical values for one-way fixed effects models with random sample sizes (Q2423534)

Consider independent random variables \[ Y_{ij}, \quad 1\le i \le m, j\ge 1, \quad Y_{ij}\sim N(\mu_i, \sigma^2). \] Suppose that we observe $Y_{ij}$, $1\le i \le m, 1 \le j \le N_i$, where the sample sizes $N_i$ are realizations of independent random variables. It is assumed that $N_i\ge n_i^*$, $1\le i \le m$, where lower bounds $ n_i^*$ are given. The goal is to test the hypothesis $H_0 : \mu_1=\dots=\mu_m$. In the paper under review, the following two situations are studied. \begin{itemize} \item[(a)] The sample sizes have Poisson distributions with different and known parameters; an example of this is the collection of observations during a fixed time period in a study comparing, e.g., several pathologies of patients arriving at a hospital. \item[(b)] The sample sizes have binomial distributions with known parameters $r_1,\ldots,r_m$ and $1-p$, where $p$ is the probability of an observation failure; this is applicable if we have observations failure and there exist upper bounds for the sample sizes. \end{itemize} In both situations, precise critical values are obtained. The approach is applied to a real data example on cancer registries. The results suggest that false rejections of the null hypothesis may be avoided by using this technique.