Identification before estimation

Why causal projects should begin with the estimand, assumptions, and source of variation.

Core Idea

The first question in a causal project is not which model to fit. It is whether the available variation can support the causal question.

A Practical Checklist

• What is the treatment?
• What is the outcome?
• What population is the estimand about?
• What intervention does the treatment correspond to?
• What counterfactual comparison is required?
• What assumptions make the comparison credible?
• What diagnostics can make those assumptions more or less plausible?

Industry Translation

When a stakeholder asks, “What is the impact of this product change?”, the analyst should clarify whether the target is an average effect, a segment-specific effect, or the value of a treatment policy.