RubinKausalmodell

The Rubin Causal Model (RCM), also known as the potential outcomes framework, is a formal approach to causal inference used in statistics, econometrics, and social sciences. It was developed around the ideas of Donald Rubin, building on earlier work by Jerzy Neyman, and emphasizes comparing what would have happened under different treatment conditions for the same unit.

In the RCM, every unit i has two potential outcomes: Yi(1) if it receives the treatment and Yi(0) if it does not. The observed outcome Yi equals Yi(Ti), where Ti denotes the actual treatment received. A causal effect for unit i is defined as the difference Yi(1) − Yi(0). Since a unit cannot simultaneously receive both treatments, these individual effects are generally unobservable, and researchers focus on average effects such as the average treatment effect (ATE) E[Yi(1) − Yi(0)]. The framework requires assumptions to identify these effects from data, with the stable unit treatment value assumption (SUTVA) and the notion of consistency being central.

Key assumptions include consistency, SUTVA, and ignorability (unconfoundedness): given covariates X, the treatment assignment Ti is independent of the potential outcomes, and there is overlap so every unit has a positive probability of receiving each treatment. In randomized experiments, randomization ensures ignorability, making simple differences in means estimators unbiased for the ATE. In observational settings, researchers use methods such as propensity score matching, weighting, stratification, regression adjustment, or instrumental variables to address confounding.

Extensions cover conditional average treatment effects (CATE), average treatment effects on the treated (ATT), and multi-valued or continuous treatments. Limitations include sensitivity to unmeasured confounding and interference between units, which can complicate identification and estimation. The Rubin Causal Model remains a foundational framework for causal inference alongside graphical approaches and modern econometric methods.