To Adjust Or Not To Adjust? Estimating The Average Treatment Effect In Randomized Experiments...

on YouTube ↗32m

" summary

Randomized trials with missing covariates (example: 27 covariates, seven with missingness, ~20% units incomplete) get an easy, provably good estimator: impute missing covariates as zero and augment with missingness indicators, then run OLS and report robust SEs. Under a design‑based framework with missingness fixed before treatment (M(1)=M(0)), the ATE estimator is unbiased, invariant to the imputed value, more efficient than complete‑case or single‑impute approaches; a full missingness‑pattern stratification is even better but requires many units per pattern.

AI-written summary · about

" chapters (17)

1. 0:00Intro
2. 1:02Randomized experiments and covariate adjustment
3. 5:02Missingress patterns in Duflo et al (2011 AER)
4. 5:30The current default covariate adjustment
5. 6:25How to deal with missing x in randomized experiments?
6. 10:42Start from a simple yet reasonable scenario
7. 13:23complete-case (cc) analysis
8. 14:08complete covariate (ccov) analysis
9. 15:19single imputation (imp)
10. 18:16missingness-indicator method (mim)
11. 21:57missingness pattern (mp) method
12. 22:20missingness-pattern (mp) method
13. 23:32illustrating the mp method with 2 missing covariates
14. 24:14Comments on the mp method
15. 25:07Properties of the mp method
16. 27:31Summary of the methods
17. 28:26Discussion of other methods