Choose the Estimand Before the Model

Principle: Do not ask a model to decide the question. ATE Use average treatment effect when the decision is about treating the eligible population as a whole. ATT Use effect on the treated when the decision is about whether the intervention helped the people who actually received it. Conditional effect Use segment effects when the decision is about targeting, eligibility, or rollout priority. Why it works Every effect estimate has a target population. If the target population is implicit, the result can be numerically correct and operationally useless.

Sign up free — one personalized lesson every day, matched to your role and goals.

Already have an account? Sign in