A1105
Title: Efficient data integration of probability and non-probability samples under dual frame sampling
Authors: Kosuke Morikawa - Iowa State University (United States) [presenting]
Abstract: Two semiparametric estimators are proposed for integrating data from independent probability and non-probability surveys. The first estimator achieves semiparametric efficiency under a framework where both survey sources provide complementary information. It employs a parametric model for the non-probability sampling mechanism and remains identifiable even when the sampling process is non-ignorable due to the support of the probability sample. The second estimator is constructed under a dual frame sampling approximation, which, while not fully efficient, attains efficiency within a restricted class of augmented terms and remains robust to the misspecification of the non-probability sampling model. Notably, this sub-efficient estimator achieves consistency and asymptotic normality without modeling the non-probability sampling mechanism. Through simulation studies and numerical examples, the comparative performance and robustness of the proposed methods are highlighted, including scenarios with model misspecification or non-identifiability. Results contribute to the growing literature on data integration and offer practical solutions for combining disparate survey sources under minimal assumptions.
