Having One's Cake and Eating It, Too

Abstract

Both true experiments and regression discontinuity (RD) designs produce unbiased estimates of effects. However, true experiments are, of course, often criticized on equity grounds, whereas RD designs entail sacrifices in terms of statistical precision. In this article, a hybrid of true experiments and RD designs is considered. This hybrid entails smaller sacrifices in statistical efficiency than RD designs and also is likely to be judged as being considerably more fair than true experiments and possible even RD designs. Moreover, unlike pure RD designs, the hybrid design considered here can be employed in situations in which the variable on which assignment to treatment condition is based takes on only a discrete set of values.