An estimator is a statistic or rule that produces an estimate of an unknown quantity. In a statistical workflow, pose the question, express the target quantity mathematically, then choose an estimator to infer it from data. For example, the sample average estimates a population mean. In causal inference, an estimator targets an identified estimand rather than answering the causal question directly. section 1.5.1 section 10.3
Use estimators as part of a disciplined causal workflow:
- Select an estimand whose identification assumptions are credible; only then choose an estimation method and weigh its statistical and computational trade-offs. section 11.4
- Treat estimator choice as replaceable once the target estimand is fixed; different methods can estimate the same target under different statistical assumptions. section 11.4.6
- Validate first with synthetic data whose causal and statistical structure is known, compare estimates and confidence intervals with ground truth, then refute and stress-test assumptions on real data. section 11.4.6
Estimating a causal effect with DoWhy
After identifying a backdoor estimand, request linear regression as its estimator: model.estimate_effect(identified_estimand, method_name="backdoor.linear_regression", confidence_intervals=True). The method estimates the selected causal target; it does not perform identification. section 11.4
Listing 11.5 Estimating the backdoor estimand with linear regression
causal_estimate_reg = model.estimate_effect(
identified_estimand, #1
method_name="backdoor.linear_regression", #2
confidence_intervals=True #3
)
Inspect the concrete DoWhy call that connects an identified estimand, an estimation method, and confidence intervals. section 11.4
