Local polynomial regresssion estimators in survey sampling

摘要: Estimation of finite population totals in the presence auxiliary information is considered. A class estimators based on local polynomial regression proposed. Like generalized estimators, these are weighted linear combinations study variables, which weights calibrated to known control totals, but assumptions superpopulation model considerably weaker. The shown be asymptotically design-unbiased and consistent under mild assumptions. variance approximation Taylor linearization suggested for design mean squared error estimators. robust sense attaining Godambe-Joshi lower bound anticipated variance. Simulation experiments indicate that more efficient than when function incorrectly specified, while being approximately as parametric specification correct.