Estimation of relative risk from matched pairs in epidemiologic research.

The matched pairs design is often used in epidemiologic research, both in prospective and retrospective studies. The Kraus estimator of relative risk (1958) has been derived in a number of ways. The development presented here employs an unconditional likelihood function. The estimator is shown to be valid only when disease incidence is low and relative risk is constant over the levels of the covariate. Asymptotic variances are derived.