Investigating an Alternative for Estimation from a Nonprobability Sample: Matching plus Calibration

Abstract Matching a nonprobability sample to probability is one strategy both for selecting the units and weighting them. This approach has been employed in past select subsamples of persons from large panel volunteers. One method weighting, introduced here, assign unit weight its matched case sample. The properties resulting estimators depend on whether weights are inverses selection probabilities or calibrated. In addition, imperfect matching can cause estimates be biased so that need adjusted, especially when size volunteer small. Calibration combined with correct bias reduce variances. We explore theoretical matched, calibrated respect quasirandomization distribution assumed describe how observed, superpopulation model analysis variables collected sample, randomization Numerical studies using simulated real data 2015 US Behavioral Risk Factor Surveillance Survey conducted examine performance alternative estimators.