Causal Inference Methods

Description

Estimates causal effects from observational data using designs such as difference-in-differences, regression discontinuity, instrumental variables and matching when randomised tests are impossible.

The fundamental problem of causal inference is that we never observe the same user both with and without a treatment. Randomised experiments solve this by design; when they are not possible, quasi-experimental methods approximate them. Difference-in-differences compares changes over time between affected and unaffected groups, assuming parallel trends. Regression discontinuity exploits thresholds that assign treatment. Instrumental variables use a factor that affects treatment but not outcome directly. Propensity score matching balances observed confounders. Synthetic control builds a weighted comparison from untreated units. Rigour comes from stating the causal question and assumptions explicitly (ideally with a DAG), choosing a design whose assumptions are plausible, testing them where possible (pre-trend checks, balance diagnostics), and running sensitivity analyses for unmeasured confounding.

Attributes

Loss type
Latent
Stage
Post-launch
Effort
Weeks

Steps

  1. State the causal question precisely, including treatment, outcome, population and time frame.
  2. Draw a DAG of plausible causes to identify confounders and choose an identification strategy.
  3. Select a design (difference-in-differences, regression discontinuity, instrumental variables, matching or synthetic control) whose assumptions fit the situation.
  4. Prepare data and test assumptions, such as parallel pre-trends or covariate balance after matching.
  5. Estimate the effect with confidence intervals.
  6. Run sensitivity and placebo analyses and report limitations alongside the estimate.

Sources

  • Causal Inference
  • Designing Experiments and Analyzing Data
  • Causal Inference (Hernán & Robins)
  • DAGs+SEM Playbook (Gopinath)
  • Nash Equilibrium Playbook (Gopinath)
  • Causal Inference: What If (Hernán & Robins)
  • Mostly Harmless Econometrics (Angrist & Pischke)
  • The Inquiry Series, Ch 45 — The Causal Revolution — Pearl's Ladder & the Fundamental Problem
  • Veal, A. J. & Darcy, S., Research Methods in Sport Studies and Sport Management (2014)

Also known as

  • Quasi-experimental methods
  • Observational causal inference
  • Fundamental problem of causal inference

Categories

  • Analytics & Behaviour
  • Experimentation & Causal Inference