This article was accepted into the corpus but its outbound wikilinks were never NER-processed — typical at the deepest BFS hop or when the run's entity cap was reached. No expansion funnel to show.
| Rubin causal model | |
|---|---|
| Name | Rubin causal model |
| Discipline | Statistics |
| Introduced | 1974 |
| Key people | Donald Rubin |
| Notable works | "Estimating Causal Effects of Treatments in Randomized and Nonrandomized Studies" (Rubin, 1974) |
Rubin causal model
The Rubin causal model provides a framework for causal inference built on potential outcomes and counterfactual reasoning developed by Donald Rubin and contemporaries. It formalizes treatment effects using potential outcomes, links to design-based inference from Jerzy Neyman and randomization theory, and has influenced applied research across fields including Epidemiology, Econometrics, Political Science, and Public Health. Scholars such as Paul Holland, James Heckman, Guido Imbens, and Judea Pearl have debated, extended, and contrasted its approach with alternative frameworks.
The Rubin causal model frames causal questions by comparing observed outcomes under actual interventions to counterfactual outcomes under alternative interventions, using potential outcomes notation inspired by Jerzy Neyman and formalized by Donald Rubin. It emphasizes design, randomization, and assignment mechanisms as inferences made by investigators such as Ronald Fisher and later formalized in the work of Fisherian randomization. The model interacts with methodological traditions from Biostatistics, Econometrics, and Psychometrics, and sparked discourse with proponents of graphical models exemplified by Judea Pearl.
Potential outcomes are defined for each unit as the outcome that would occur under each possible treatment level, an idea rooted in the experimental designs of Ronald Fisher and Neyman's work on randomized experiments. The model employs notation and estimands such as the average treatment effect (ATE) and the individual treatment effect; influential papers include Donald Rubin (1974) and clarifications by Paul Holland. Connections to likelihood-based inference appear in work by George Box and Leonard Savage, while Bayesian formulations of the model were advanced by Donald Rubin himself and later developed by Andrew Gelman and Dennis Lindley.
Estimation strategies include randomized experiments, instrumental variables, matching, regression adjustment, inverse probability weighting, and doubly robust estimators. Randomized controlled trials trace back to designs championed by Ronald Fisher and institutionalized by National Institutes of Health trials. Instrumental variable methods were popularized in economics by Angus Deaton and James Heckman; matching algorithms draw on algorithmic work by Donald Rubin and computational developments in machine learning by Leo Breiman and Judea Pearl influences. Seminal texts by Guido Imbens and Joshua Angrist synthesize identification via natural experiments and quasi-experiments investigated in empirical studies by Dale Jorgenson and Angus Deaton.
Key assumptions include stable unit treatment value assumption (SUTVA), no unmeasured confounding (ignorability), and positivity (overlap). Critics such as Judea Pearl and James Heckman have argued about the role of structural models and graphical criteria versus potential outcomes. Discussions about external validity and transportability reference work by Miguel Hernán and James Robins and engage debates involving Paul Rosenbaum on sensitivity analysis and design sensitivity. Philosophical critiques have roots in writings by Nancy Cartwright and Donald Campbell regarding causal generalization.
Extensions include principal stratification developed by Frangakis and Rubin, mediation analysis linking to work by Robins and Greenland, and longitudinal causal inference advanced by James Robins. Graphical causal models promoted by Judea Pearl and structural equation modeling associated with Olley and Pakes and Kenneth Arrow provide complementary approaches. Bayesian nonparametric and machine learning integrations involve researchers like Bradley Efron and Susan Athey, while transportability and external validity draw on contributions from Eyal Ben-David and Miguel Hernán.
The framework has been applied to randomized clinical trials coordinated by Food and Drug Administration regulations and World Health Organization trials, program evaluation in social policy studied by Esther Duflo and Abhijit Banerjee, labor economics analyses by James Heckman and Angus Deaton, and education interventions evaluated in large-scale randomized evaluations by John Hattie and Eric Hanushek. Epidemiological studies influenced by Miguel Hernán and James Robins leverage the model for causal effect estimation in observational cohorts curated by Framingham Heart Study and policy impact assessments such as analyses of Affordable Care Act outcomes.
Practical implementation uses software ecosystems shaped by contributions from R Core Team, Hadley Wickham, and packages inspired by Guido Imbens and Donald Rubin teachings; computational methods bring in ideas from Leo Breiman and Trevor Hastie. Design choices reflect institutional review by bodies like Institutional Review Board protocols and regulatory guidance from Food and Drug Administration. Sensitivity analysis techniques and diagnostics follow recommendations by Paul Rosenbaum and applied workflows in public datasets from institutions like World Bank and Centers for Disease Control and Prevention.
Category:Causal inference