Moving Toward Best Practice When Using Propensity Score Weighting in Survey Observational Studies

Area of Science:

• Statistics
• Epidemiology
• Causal Inference

Background:

• Propensity score weighting is standard for observational data analysis.
• Optimal use of survey weights in this context remains unclear.
• Existing methods lack consensus for population-level causal inference with survey data.

Purpose of the Study:

• To provide a unified framework for propensity score weighting with survey data.
• To develop novel weighting and augmented weighting estimators for various target populations.
• To derive robust variance estimators for these new methods.

Main Methods:

• Developed a unified solution within the balancing weights framework.
• Derived weighting and augmented weighting estimators for combined, treated, controlled, and overlap populations.
• Applied M-estimator theory to create closed-form sandwich variance estimators.

Main Results:

• Proposed estimators demonstrated good performance in extensive simulation studies.
• New methods were compared against alternative approaches.
• Case studies illustrated practical application with complex survey data.

Conclusions:

• The unified framework effectively incorporates survey weights for causal inference.
• The developed estimators offer a reliable approach for propensity score weighting with survey data.
• Practical recommendations are provided for analyzing survey observational data.