Correlation, Causation and Research Design
EVIDENCE LAB · 06
Correlation
& Causation.
Patterns describe what moves together. Causal analysis asks what would have happened otherwise—and demands a credible design for answering it.
TREATMENTCONTROLDESIGN
01 · THE CAUSAL LADDER
Association is evidence.
Not yet an effect.
Description
What happened, to whom, where and when?
E[Y]
Association
How does outcome Y vary when exposure X varies?
Corr(X,Y)
Causation
How would Y change if X were deliberately changed?
E[Y(1)−Y(0)]
Y(1): treated outcome
What happened after the person, firm or region received the intervention.
−
Y(0): untreated outcome
What would have happened to that same unit at the same time without treatment.
02 · THREATS TO IDENTIFICATION
Three routes from pattern
to false conclusion.
Confounding
A third factor influences both X and Y. Ice-cream sales do not cause drowning; hot weather raises both.
Reverse causality
Y may affect X. Growth can attract investment even when investment also influences growth.
Selection bias
Treated and untreated units may differ before treatment in ways that also affect the outcome.
CAUSAL DESIGN STUDIO
Build the
counterfactual.
Define the intervention, outcome, comparison and assignment mechanism. ECONORIA will audit the identification logic.
Define the treatment and outcome.
- Treatment
- Not defined
- Outcome
- Not defined
- Counterfactual proxy
- Not defined
- Design
- Observational comparison
- Key threat
- Unobserved differences between groups
CAUSAL VERDICTA descriptive association can be estimated, but no credible causal claim is yet supported.
03 · IDENTIFICATION STRATEGIES
The strongest design is the one
whose assumptions survive scrutiny.
Randomised experiment
Assignment by chance creates comparable groups in expectation.
Regression discontinuity
Compare units narrowly around a policy eligibility threshold.
Difference-in-differences
Compare changes over time between treated and comparison groups.
Instrumental variables
Use external variation affecting treatment but not the outcome directly.
