Exact estimates for a rate ratio.

The incidence rate ratio is a basic measure of association in epidemiology. We present a simple and efficient method for computing exact confidence limits for the common rate ratio in a series of 2 x 2 tables with person-time denominators. The method uses a polynomial multiplication (convolution) algorithm previously described for an odds ratio. We also present two tests, one asymptotic, the other exact, for evaluating rate ratio homogeneity. We extend these homogeneity tests to apply to data where strata have been partitioned into subgroups in which the rate ratio is assumed homogeneous within subgroups, but not necessarily between them. We conclude with a brief description of a microcomputer program that computes exact and asymptotic estimates for both a rate ratio and an odds ratio.