A Multiple Imputation Comparison Analysis Approach to Health Surveys Subject to Selection Bias
Conference
65th ISI World Statistics Congress 2025
Format: IPS Abstract - WSC 2025
Session: IPS 890 - Recent Advances in Missing Data Methods for Health Research
Monday 6 October 10:50 a.m. - 12:30 p.m. (Europe/Amsterdam)
Abstract
Beyond traditional personal interviews, survey researchers have increasingly used alternative tools (e.g., web surveys) to collect information for population health research. These data are often referred to as nonprobability samples due to the lack of a well-defined probability sampling structure, or they come from probability samples subject to high nonresponse and coverage errors. Certain statistical adjustments are needed to make proper inferences using these data. One popular approach is to create pseudo-weights and conduct weighted analysis, borrowing the information from a reference probability survey with high quality. When the variable of interest in the nonprobability sample is not collected in the reference survey, we propose to multiply impute the missing variable and analyze the imputed data, combining the information from both data sources. Furthermore, results from different imputation and weighting methods can be compared to aid the inference. We illustrate some main features and performance of the proposed strategy using simulation studies and a web survey based on a commercial probability panel and the public-use data of National Health Interview Survey.