The doubly robust moment of Sant'Anna and Zhao (2020) is a weighted
average whose control weights contain pi/(1 - pi), so a single
control unit with a propensity near one can dominate. Kish's
effective sample size ESS = (sum |w|)^2 / sum w^2 equals
n for equal weights and falls towards 1 as one weight
dominates. Stabilisation caps the control weights at their
q-quantile and renormalises. q = 1 leaves every weight
untouched, so the estimate then equals the raw estimate exactly.
Usage
Drvst(y, D, X = NULL, q = 0.95)
Arguments
- y
Outcome change, one entry per unit.
- D
Binary treatment indicator.
- X
Optional baseline covariates.
- q
Quantile at which the control weights are capped, in (0, 1].
Value
List with estimate, tau_raw, ess_raw,
ess_stab, ess_ratio, cap, max_weight,
n.
References
Sant'Anna, P. H. C. and Zhao, J. (2020). Journal of
Econometrics 219(1), 101-122, equation (2.6). Kish, L. (1965).
Survey Sampling, Wiley, section 11.7. Crump, R. K., Hotz, V. J.,
Imbens, G. W. and Mitnik, O. A. (2009). Biometrika 96(1), 187-199.
