|
Abstract:
|
Calibration estimators use auxiliary data to improve the efficiency of survey estimates. Traditionally, the control totals, to which sample weights are calibrated, are assumed to be population values with no sampling variance. Often, however, estimates from other surveys are used because the population controls either do not exist or are not readily accessible. In this situation, many researchers apply traditional variance estimators to cases where the control totals are estimated, thus assuming that any additional sampling variance associated with these controls is negligible. We compare the MSE for linearization and replication variance estimators of proportions when the uncertainty in the control totals is either addressed or ignored. Illustrations are given of the effects of different levels of variability in the estimated controls on the overall variance estimates.
|
