All simulators
Raw, ANCOVA & Lin Adjustment
Re-randomise a training experiment and compare the spread of three estimators when slopes differ by arm.
Day 39: Regression Adjustment in Experiments (Lin's Estimator)raw difference
1.355
bias 0.029
ANCOVA
0.722
bias -0.012
Lin (interacted)
0.674
bias -0.008
Values are standard deviations across re-randomisations.
Job training with baseline earnings as a covariate. ANCOVA forces a common slope; when slopes differ by arm and the arms are unequal it can even lose to the raw difference. Lin's fully interacted, centred regression never does worse asymptotically.