Glossary Entry

Causal Inference

The discipline of estimating what would happen under an intervention, rather than what is merely associated in observed data, using assumptions about how the data came to be.

Statistics Decision Making

Also called: causal effect estimation, identification strategy

Seed source: Causal Inference - The Mixtape (Cunningham)

Causal inference asks intervention questions (“what happens to revenue if we make customers do X?”) rather than association questions (“what is revenue like among customers who did X?”). The two differ whenever the people who received a treatment differ from those who did not, which in any functioning business is nearly always.

Because the counterfactual outcome is never observed, every causal estimate rests on an identification strategy: an explicit argument, such as randomization, measured confounders, an instrument, a threshold rule, or parallel trends, for why the computed quantity equals the causal one. The methods are only as good as the assumptions, which is why refutation tests and sensitivity analysis are part of the craft.