A0628
Title: Quantile regression-based data integration for combining probability and nonprobability samples
Authors: Sixia Chen - University of Oklahoma (United States) [presenting]
Emily Berg - Iowa State University (United States)
Cindy Yu - Iowa State University (United States)
Abstract: Researchers often encounter nonprobability samples in practice, including biomedical research, business study, educational research, and other fields. Statistical analysis by using nonprobability samples without further adjustment may lead to biased results due to selection bias. Date integration has been regarded as one of the effective ways to handle nonprobability samples. It combines the information from both nonprobability samples and probability samples to reduce selection bias. Commonly used data integration methods include mass imputation, Propensity score weighting, Calibration, and Hybrid methods. The validity of those methods depends on the underlying model assumptions. To improve the robustness of model misspecification and protect the outliers, a novel quantile regression-based mass imputation method is proposed as a doubly robust method with a nonparametric estimation of the propensity score model. The proposed methods are more robust compared to some existing methods in terms of model misspecification and outliers. Asymptotic theory, including consistency, asymptotic normality, and variance estimation procedures, has been developed. The methods are further evaluated by using a Monte Carlo simulation study and one real data application.
