158 – Inference with Non-Probability Sample Through Data Integration — Topic Contributed Papers
Inference With Non-Probability Sample Through Data Integration
Phillip S. Kott
RTI International
Non-probability sample happens frequently in practice due to its cost and time efficiency. Sometimes, it is the only way to recruit participants due to difficulty of creating the sampling frame. For instance, it has been used frequently to recruit patients with certain disease such as lung cancer since it is impossible to create a frame of people with lung cancer without diagnosis. However, non-probability sample may suffer from serious sampling, coverage and nonresponse error without additional adjustments. Data integration by combining information from non-probability sample and another probability sample has been shown to be effective in terms of reducing such errors. In this session, parametric, nonparametric and semi-parametric data integration approaches including propensity score weighting, mass imputation or both will be discussed for inference with non-probability sampling design.