Causal inference

Causal inference is the discipline of establishing whether a change in one variable actually produces a change in another, as opposed to the two simply moving together. Correlation alone cannot distinguish cause from coincidence, reverse causation, or a shared underlying driver—causal inference supplies the methods to rule those alternatives out.

The strongest method is a randomized controlled experiment, such as A/B testing, where random assignment ensures that groups differ only in the treatment they receive, so any outcome difference can be attributed to that treatment. When randomization is not possible, analysts use quasi-experimental methods—difference-in-differences, instrumental variables, regression discontinuity, or marketing mix modeling—that try to approximate a controlled comparison using observational data, along with explicit assumptions about what would have happened otherwise.

Causal inference matters wherever a decision assumes an action will produce an effect: attributing a sales lift to a campaign, or a churn reduction to a product change. It underlies incrementality testing and much of rigorous diagnostic analytics. The central pitfall is confusing correlation with causation—concluding a factor caused an outcome purely because the two appeared together in historical data, without accounting for confounding variables or alternative explanations that a proper causal design would control for.