Providing double protection for unit nonresponse with a nonlinear calibration-weighting routine

Abstract

Given a randomly drawn sample, calibration weighting can provide double protection against the selection bias resulting from unit nonresponse. This means that if either an assumed linear prediction model or an implied unit selection model holds, the resulting estimator will be asymptotically unbiased in some sense. The functional form of the selection model when using linear alibration adjustment is dubious. We discuss an alternative, nonlinear calibration-weighting procedure and software that can, among other things, implicitly estimate a logistic-response model.