Foundations

Foundations MOC

The conceptual bedrock of the credibility revolution: the potential-outcomes framework for defining causal effects, and the role of randomization and assignment mechanisms in identifying them. These papers establish the vocabulary (units, treatments, potential outcomes, estimands, assumptions) that every later method — DiD, RDD, PSM, RCT, SCM — inherits. The throughline is a single obstacle, the Fundamental-Problem-of-Causal-Inference, and the different ways of getting around it.

Papers

• Rubin1974-EstimatingCausalEffects — the seminal statement of Potential-Outcomes; defines causal effects and argues careful nonrandomized studies can estimate them, not only experiments.
• Rubin1977-AssignmentOnCovariate — selection-on-observables identification: when assignment depends only on a covariate $X$, conditioning on $X$ removes bias.
• Rubin1978-BayesianCausalEffects — the Bayesian / missing-data formulation; defines ignorable assignment mechanisms and the central role of randomization.
• Rubin1980-RandomizationAnalysis — names SUTVA and defends Fisher’s randomization test; sharpens the missing-data view.
• Holland1986-StatisticsAndCausalInference — synthesizes and names the “Rubin Causal Model”; states the Fundamental-Problem-of-Causal-Inference and the “no causation without manipulation” doctrine.

Cross-cutting surveys & inference

• ImbensWooldridge2009-ProgramEvaluation — the anchor JEL survey of program evaluation: unconfoundedness vs. design-based strategies, every estimator in one map.
• AbadieCattaneo2018-EconometricMethodsProgramEvaluation — compact Annual Review survey organizing methods by their identifying assumption.
• AbadieEtAl2020-SamplingVsDesignUncertainty — what a standard error means: design-based vs. sampling-based uncertainty.

Mechanisms, mediation & interference

• Keele2015-CausalMediationAnalysis — Mediation requires assumptions beyond the total effect; sequential ignorability and sensitivity analysis.
• VanderWeele2016-MediationAnalysis — practitioner’s guide to natural direct/indirect effects and their confounding assumptions.
• LudwigEtAl2011-MechanismExperiments — testing mechanisms by design (experiments) rather than by modelling assumptions.
• AngelucciDiMaro2016-SpilloverEffects — relaxing the no-interference part of SUTVA; designs that identify spillover effects.

Key Concepts

Potential-Outcomes · Fundamental-Problem-of-Causal-Inference · Causal-Estimand · Randomization · Ignorability · SUTVA · Overlap · Design-Based-Inference · Mediation

Debates & Contradictions

• No causation without manipulation. Holland (1986) argues that only manipulable treatments can be causes, so attributes (race, gender) cannot — a strong, contested boundary the Rubin papers do not themselves draw, and one later authors push back on.
• Scientific vs statistical solution. Holland frames two routes around the Fundamental Problem (unit-homogeneity assumptions vs population averaging); the Rubin papers commit to the statistical/population route.
• Randomization vs observational identification. Rubin (1974, 1977) insists observational data can yield credible causal estimates under assumptions, while Rubin (1978) shows randomized designs are uniquely easy to analyze — the tension that the whole credibility revolution then works to manage.

Next

All method MOCs are now live and inherit the Potential-Outcomes / Causal-Estimand vocabulary defined here: DiD, IV, RDD, PSM, RCT, and SCM. The design-based thread (Design-Based-Inference, AbadieEtAl2020-SamplingVsDesignUncertainty) ties the randomization logic here to RCT, the local-randomization view in RDD, and design-based DiD (AtheyImbens2022-DesignBasedDiD).