Variable selection and raking in propensity scoring

Area of Science:

• Epidemiology
• Biostatistics
• Health Services Research

Background:

• Longitudinal studies with pre-/posttest designs are common for evaluating interventions.
• Propensity scoring is a statistical method used to control for confounding variables.
• High-dimensional covariate spaces present challenges in propensity model development.

Purpose of the Study:

• To discuss practical issues in applying propensity scoring for endpoint analysis.
• To address challenges in pre-/posttest longitudinal designs with ordinal treatment intensity.
• To explore methods for covariate selection, model evaluation, and covariate hypercontrol.

Main Methods:

• Application of propensity scoring techniques.
• Analysis of a health communication program evaluation.
• Consideration of ordinal measures of treatment intensity.
• Management of high-dimensional covariate spaces.

Main Results:

• Practical challenges in covariate selection for propensity models were identified.
• Methods for evaluating the adequacy of propensity models were discussed.
• Strategies for tailoring propensity models for hypercontrol on specific covariates were explored.

Conclusions:

• Effective application of propensity scoring requires careful consideration of covariate selection and model validation.
• Tailoring propensity models can enhance control over key variables in longitudinal health program evaluations.
• Addressing these practical issues is crucial for robust endpoint analysis in complex study designs.