A Covariate{based Coeecient of Source Dependence for Capture{recapture Models February 2000 -final Draft

In capture{recapture modelling, it is necessary to account for possible dependence betwee sources. Along with nominal information, epidemiological capture{recapture studies often involve the collection of a variety of individual covariates. Under the assumption that we can estimate the covariate distribution in the population of interest and that sources are conditionally independent given the covariates, we can estimate a coeecient of source dependence (CSD) for every set of sources. The CSD estimates are based on the estimated conditional covariate distribution given the sources. Such CSD estimates can be used as guides to t joint log{linear models and can also be introduced in a marginal log{linear model to produce a conditional population estimate. We illustrate these applications on data from the Auckland Leg Ulcer Study, 1997{1998.