Balanced Subclassification in Observational Studies Using the Propensity Score: A Case Study.
Abstract
The propensity score is the conditional probability of assignment to a particular treatment given a vector of observed covariates. Previous theeoretical arguments have shown that subclassification on the scalar propensity score will balance all observed covariates. The procedure is illustrated in a large observational study of treatments for coronary artery disease. Five subclasses are constructed that balance 74 covariates. Balanced subclassification is combined with model-based adjustments to provide estimates of treatment effects within subpopulations. Two appendices address theoretical issues: propensity scores from incomplete data, and the effectiveness of subclassification on the propensity score.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1983
- Accession Number
- ADA127760
Entities
People
- Donald B. Rubin
- Paul R. Rosenbaum
Organizations
- University of Wisconsin–Madison