Empirical Bayes Analysis of Mortality Data I N British Columbia

Estimating and mapping mortality rates is an important step in identifying possible etiologic factors related with deaths, since the spatial distribution of mortality risks can be examined for correlations with risk factors. A main feature of mortality data is the large variability of the death rates in regions with small populations. This hinders the application of standard large-sample approaches such as the use of a normal approximation to the transformed data or generalized linear modeling. An empirical Bayes procedure proposed by Manton, Woodbury, Stallard, Riggan, Creason and Pellom (1989) is applied in the analysis of death rates in the Province of British Columbia. The stochasticity of rates in small-population regions is dealt with by layered random effects modelling. A closed form solution to estimating equations for the parameters is achieved by a two-stage empirical Bayes estimation method. This mortality rate model and its estimation procedure are described in this project and a map of the death rates of B.C. is provided.