Estimation of the Conditional Variance in Paired Experiments

ABADIE, IMBENS
2008 Annales d Économie et de Statistique  
In paired randomized experiments units are grouped in pairs, often based on covariate information, with randomized assignment within the pairs. Average treatment effects are then estimated by averaging the within-pair differences in outcomes. Typically the unconditional variance of the average treatment effect estimator is estimated using the sample variance of the within-pair differences. Conditional on the covariates the variance may be substantially smaller. Here we propose a simple way of
more » ... e a simple way of estimating the conditional variance by forming pairs-of-pairs with similar covariate values and estimating the variances within these pairs-of-pairs. Even though these withinpairs-of-pairs variance estimates are not consistent, their average is consistent for the variance of the average treatment effect estimator and leads to asymptotically valid confidence intervals.
doi:10.2307/27917244 fatcat:xj5gsos6nvbydiuw6c6qihurqi