Application of Propensity Score Matching with Complex Survey Weighted Tobacco Smoking Data

• Activity Number: 210 - Contributed Poster Presentations: Survey Research Methods Section
• Type: Contributed
• Date/Time: Tuesday, August 4, 2020 : 10:00 AM to 2:00 PM
• Sponsor: Survey Research Methods Section
• Author(s): Grace H Choi* and Cedric H Bien-Gund and Robert Gross and Kathleen A Brady and Pamela A Shaw
• Companies: University of Pennsylvania Perelman School of Medicine and University of Pennsylvania Perelman School of Medicine and University of Pennsylvania Perelman School of Medicine and Philadelphia Department of Public Health AIDS Activites Coordinating Office and University of Pennsylvania Perelman School of Medicine
• Keywords: propensity score; propensity score matching; survey weights; complex survey data; smoking; HIV

Abstract

Propensity score analysis methods are ubiquitously used in observational studies to balance covariates between two comparative groups in an effort to reduce confounding and biased estimates. Many observational studies are sourced from large nationally representative surveys that have complex survey designs. Unfortunately, there is limited methodology and guidelines on how to best incorporate propensity score analysis methods and the complex survey design of the data to draw appropriate population inferences. Questions arise on how and when the propensity scores should be calculated and evaluated in the outcome model in the presence of survey weights. We compared smoking rates among people living with HIV and the general population in the Philadelphia metropolitan area represented by two health surveillance surveys, the Medical Monitoring Project and the Southeastern Pennsylvania Household Health survey respectively, conducted between 2009-2014, using propensity score matching where the propensity score was defined with and without survey weights. These two methods produced different quantitative results but had the same overall conclusions.