Robustness of population size estimates against violation of the independence assumption

An important quality aspect of Censuses is the degree of coverage of the pop-ulation. When administrative registers are used, undercoverage can be estimated by linking two or more registers and subsequently estimating the number of individuals that is missed by each of the registers. Adding this estimated number to the observed number of individuals yields a population size estimate. This is also known as the capture-recapture method. The standard approach uses loglinear models that most often rely on the assumption that being in the first register is statistically independent from being in the second register. If covariates are available, the independence assumption can be replaced with a less strict independence assump-tion conditional on covariates. However, both independence and independence conditional on covariates are rarely met. In this paper we will investigate the robustness of the population size estimate under dependence. The results show that violation of the independence assumption can lead to seriously biased estimates. Key words: Capture-recapture method, population size estimate, sensitivit