Efficiency,Bias,and Classification Schemes

When estimating voucher impacts on test scores in the New York City randomized field trial (RFT) for African Americans (defined either by mother’s ethnicity, parental caretaker, mother and father’s ethnicity, or mother or father’s ethnicity), results remain significantly positive, even when models include students for whom no baseline test scores are available. These results obtain as long as one estimates impacts precisely by controlling for baseline test scores for those students who have them. Positive impacts fall below conventional levels of significance only when analysts needlessly drop baseline test score information or add numerous covariates that neither singly nor together enhance the precision of the estimates. When results differ for those with and without baseline scores, analysts should give greater weight to those for whom one has stronger evidence that the RFT has not been contaminated.

Due to space limitations imposed by the editors, this article replies to the original Krueger and Zhu (KZ) article but not to their rejoinder. Our reply to the rejoinder may be found at www.ksg.harvard.edu/pepg/. There, we discuss the trivial nature of the differences between findings reported below and those in KZ’s replication, our offer to check the replication effort to see how differences could have arisen, the Office of Management and Budget (OMB) definition of “Black (non-Hispanic),” and other issues raised in their rejoinder.