DataTerminology.comNEW
Also known as: Rubin Causal Model
A causal inference approach defining effects by comparing potential outcomes under different treatments for the same unit.
DESCRIPTION
The potential outcomes framework formalizes causal effects as contrasts between what would happen under treatment versus control for each unit. It underpins randomized controlled trials and observational methods like propensity scoring. Key assumptions include SUTVA and ignorability. Estimands include ATE and ATT. The framework guides design, identification, and estimation with matching, weighting, and regression. It complements Pearl’s structural causal models by focusing on counterfactual contrasts.
ALTERNATIVE DEFINITIONS
Wikipedia
The potential outcomes approach to causal inference (also known as the Neyman–Rubin causal model) provides a framework for defining and estimating causal effects by comparing outcomes under different treatment assignments.
Highlights definitional focus on contrasts across treatment assignments.
Wikipedia
The Rubin causal model (RCM), also known as the potential outcomes framework, is a causal model in statistics in which causal effects are defined by comparing the potential outcomes of the same units under different treatment assignments.
Stresses core idea of unit-level potential outcomes comparisons.
RELATED TERMS
Counterfactual Average Treatment Effect Randomized Controlled Trial Causal Inference Selection Bias Instrumental Variable Propensity Score SUTVA Do-Calculus
COMPETING TERMS
RELATIONS
- Broader: Framework
KNOWLEDGE AREAS
TARGET AUDIENCE
Data Scientist Data Analyst AI Product Manager Information Analyst
EXTERNAL SOURCES
- Wikipedia - Potential outcomes approach
- Harvard T.H. Chan School of Public Health - Causal inference textbook
- Scott Cunningham - Applied causal inference
TOP QUESTIONS
What is the core idea of the Potential Outcomes Framework?
Causal effects are defined as contrasts between potential outcomes the same unit would have under different treatments.
Category: Definition
What assumptions are crucial in the Potential Outcomes Framework?
SUTVA and ignorability/confounder control ensure well-defined counterfactuals and unbiased estimation.
Category: Requirements
How do randomized trials fit into the Potential Outcomes Framework?
Randomization ensures ignorability, making the comparison of potential outcomes identifiable with minimal confounding.
Category: Example
What is ATE in the Potential Outcomes Framework?
The Average Treatment Effect is the expected difference in potential outcomes between treatment and control across units.
Category: Definition
How does the framework handle observational data?
Matching, weighting, and regression adjust for confounding to approximate randomized comparisons.
Category: Overview
How does this differ from Structural Causal Models?
Potential outcomes focus on contrasts; SCMs emphasize graphical structures and interventions via do-calculus.
Category: Alternatives
What is SUTVA and why does it matter?
SUTVA assumes no interference and consistency, ensuring treatment assignment uniquely defines potential outcomes.
Category: Requirements
What mistakes occur when applying the Potential Outcomes Framework?
Violating SUTVA, ignoring confounders, and poor overlap can bias estimates or inflate variance.
Category: Mistakes
Where can I learn more about Potential Outcomes?
Use Hernán & Robins’ book and the Mixtape for practical methods and assumptions.
Category: Overview