14.7 Estimation with Matched Data

Doubly robust estimation. An estimation method that identifies the effect even if one of the regression or matching models is misspecified.

1. Estimate propensity score \(p\) for each observation
2. Compute inverse probability weights

1. Treated observations: \(\frac{1}{p}\)
2. Untreated observations: \(\frac{1}{1-p}\)

3. Make regression model on treated observations
4. Make regression model on untreated observations
5. Store predictions for each model
6. Compare means
7. Standard errors? Use bootstrap standard errors