Donald Rubin

Overview

Donald B. Rubin is a statistician (Harvard University, retired; now at Tsinghua University). He is one of the most influential statisticians of the twentieth century, known for the potential outcomes framework for causal inference, multiple imputation for missing data, and foundational contributions to Bayesian inference and model checking.

Key Contributions

• Rubin (1984): "Bayesianly Justifiable and Relevant Frequency Calculations for the Applied Statistician." Annals of Statistics 12: 1151–1172. Formalised the posterior predictive distribution as the basis for Bayesian model checking; showed that posterior predictive p-values have Bayesian justification even when they do not behave like classical p-values. Direct predecessor of Gelman-Meng-Stern (1996).
• Discussant of Gelman-Meng-Stern (1996): Identified the conservatism of ppc p-values as a consequence of positive correlation between $y$ and $y^{rep}$ — analogous to the superefficiency phenomenon in multiple imputation (the fraction of missing information is underestimated). Proposed conditioning on fixed features of the data to reduce conservatism. See Posterior Predictive Check.
• Gelman-Rubin (1992): $\hat R$ potential scale reduction factor for MCMC convergence, using multiple chains.
• Potential outcomes / Rubin Causal Model: Framework for counterfactual causal inference; fundamental to modern applied statistics.
• Multiple imputation: Theory and practice for handling missing data via repeated imputation; fraction-of-missing-information concept.

Related

• Andrew Gelman
• Xiao-Li Meng
• Hal Stern
• Posterior Predictive Check
• Gibbs Sampler