48. How can measurements from before an experiment improve sensitivity without compromising the randomized comparison?
HardStatistics & experiments
Technical
Show model answer
Prespecified adjustment for predictive pre-treatment measurements can explain some outcome variation, making the treatment estimate more precise. Use covariates that were determined before assignment and cannot have been changed by the treatment.
Check coverage, missingness and the adjustment procedure on suitable historical or validation data. Do not select favorable covariates after inspecting the result, and report the treatment effect with uncertainty from the actual adjusted analysis.
What interviewers listen for
- Use predictive pre-treatment covariates
- Check coverage and adjustment validity
- Avoid outcome-driven selection