Analyze Causal Inference

Name the estimand before choosing an estimator. If identification fails, say not identified instead of reporting a number. A narrow interval around a biased estimate is not strong causal evidence.

Open causal inference methods when choosing an adjustment set, design, or sensitivity method.

Use design-product-experiment when assignment can still be designed. Use forecast-with-calibration for a future event. Use analyze-system-dynamics when the question is mechanism over time rather than one contrast.