The distribution of the counts Yi is typically assumed to come from a Poisson distribution, as the diseases usually considered in such studies are rare and this distribution gives a good approximation to the underlying binomial distribution that would hold for each risk stratum. The local variability of the counts is thus modeled as follows: