A sample coverage approach to multiple-system estimation with application to census undercount

The concept of "sample coverage" used in animal abundance estimation is modified to evaluate the undercount of a census. Lack of independence between the census and its postenumeration survey leads to correlation bias for the standard estimator of population size. An additional recapture sample (besides the census and postenumeration survey) can be used to estimate the correlation bias due to two types of dependences. This work expresses the correlation bias for the three-sample model as a function of expected sample coverage and measures of dependence between lists. A nonparametric population size estimator that incorporates the correlation bias is proposed. A simulation study investigates the performance of the proposed procedure. Data from a 1988 dress rehearsal study for the 1990 census conducted by the U.S. Bureau of the Census are used to illustrate the proposed three-system estimation procedure and to compare the resulting estimates with those given by Darroch, Fienberg, Glonek, and Junker and Zaslavsky and Wolfgang.