Variance estimation in dual registration under population heterogeneity - ARCHIVED

The usual dual system estimator for population size can be severely biased, if there is population heterogeneity in the capture probabilities. In this note we investigate the bias of the corresponding variance estimator under heterogeneity. We show that the usual estimator is conservative, i.e., it gives too large values, if the two registration systems are negatively correlated, uncorrelated, or when the correlation is positive, but small. In the case of high positive correlation the usual estimator may yield too low values. Two alternative estimators are proposed. One is conservative under arbitrary heterogeneity. The other is conservative under Gaussian heterogeneity. The methods are applied to occupational disease data from Finland.