Rubin Causal Model (Potential Outcomes)

Description

Defines a causal effect as the difference between potential outcomes under treatment and control, estimated from matched cohorts when randomisation is impossible.

Each unit has a potential outcome under treatment and under control, only one of which is observed; the average treatment effect (ATE) is the mean difference across units. Without randomisation, unbiased estimation requires that assignment be independent of potential outcomes given measured covariates (unconfoundedness), which motivates matching or weighting on those covariates. Vivek Gopinath's playbook prescribes it for UX when an A/B test cannot be run: match cohorts by user type (novice versus advanced traders), device and session time, then estimate ATE, documenting the unconfoundedness assumption. It is paired with Bradford Hill criteria as a narrative check on the resulting evidence.

Attributes

Loss type
Unmigrated
Stage
Post-launch
Effort
Weeks

Steps

  1. State the treatment (e.g. new IA exposure), the outcome and the unit of analysis.
  2. List confounders that drive both exposure and outcome (user type, device, session time, tenure).
  3. Match or weight treated and untreated users on those confounders and check balance.
  4. Estimate the ATE as the mean outcome difference across matched cohorts with an interval.
  5. Document the unconfoundedness assumption and run sensitivity checks with additional plausible confounders.
  6. Corroborate with Bradford Hill criteria and qualitative mechanism evidence.

Sources

  • Vivek Gopinath (2025), Forward-Thinking UX Research Playbook: A Mixed-Methods, Decision-Ready Approach from Prediction to Validation
  • Lost Methods in UX — A Complete Treatise (Gopinath)
  • Rubin DB (1974) Estimating causal effects of treatments in randomized and nonrandomized studies. J Educ Psychol 66(5):688-701
  • Imbens GW & Rubin DB (2015) Causal Inference for Statistics, Social, and Biomedical Sciences
  • Holland PW (1986) Statistics and causal inference. JASA
  • Rubin, Estimating causal effects of treatments in randomized and nonrandomized studies (1974), Journal of Educational Psychology
  • Holland, Statistics and causal inference (1986), JASA
  • Imbens & Rubin, Causal Inference for Statistics, Social, and Biomedical Sciences (2015)
  • causalgraphicalmodels
  • dowhy
  • causalml
  • econml
  • statsmodels
  • Causal Inference: What If (Hernán & Robins)
  • The Inquiry Series, Vol III, Ch 48
  • The Inquiry Series, C5 — Causal Inference in Python — DAGs, Target Trial Emulation & Sensitivity Analysis
  • Causal Inference — Hernán & Robins
  • Causal Inference: What If (Miguel Hernán & James Robins)
  • DAGs+SEM Playbook (Vivek Gopinath)
  • Hernán & Robins (2016) Using big data to emulate a target trial when a randomized trial is not available
  • Causal Inference (Hernán & Robins)
  • DAGs+SEM Playbook (Gopinath)
  • The Inquiry Series, Ch 48 — Target Trial Emulation — Causal Evidence from Log Data

Also known as

  • Potential outcomes framework
  • Neyman-Rubin model
  • Matched cohort analysis
  • Counterfactual framework
  • Potential outcomes
  • RCM
  • Rubin Causal Model (Potential Outcomes Framework)
  • Emulated target trial
  • Hernan-Robins target trial framework
  • Target trial emulation
  • Rubin causal model
  • Counterfactual Framework for Observational Studies
  • Target trial protocol
  • Target trial specification (7 components)

Categories

  • Analytics & Behaviour
  • Experimentation & Causal Inference