Modeling matching error and its effect on estimates of census coverage error - ARCHIVED

Dual system estimators of census undercount rely heavily on the assumption that persons in the evaluation survey can be accurately linked to the same persons in the census. Mismatches and erroneous non-matches, which are unavoidable, reduce the accuracy of the estimators. Studies have shown that the extent of the error can be so large relative to the size of census coverage error as to render the estimate unusable. In this paper, we propose a model for investigating the effect of matching error on the estimators of census undercount and illustrate its use for the 1990 census undercount evaluation program. The mean square error of the dual system estimator is derived under the proposed model and the components of MSE arising from matching error are defined and explained. Under the assumed model, the effect of matching error on the MSE of the estimator of census undercount is investigated. Finally, a methodology for employing the model for the optimal design of matching error evaluation studies will be illustrated and the form of the estimators will be given.