Does Competition Among Public Schools Benefit Students and Taxpayers? Reply

摘要

In each case where Rothstein argues that a reasonable change to the specifications estimated in Hoxby (2000) would substantially affect the results, I have shown that the argument is either incorrect or has negligible effects on the results (the latter is usually shown in his Comment as well). He is able to obtain substantially different results only by introducing important errors such as the error-prone zip-code-backing-out method. Similarly, reasonable alterations to the instrumental variables for streams do not change the results. The results change only when Rothstein introduces variables that do not measure what the instruments are intended to measure. Rothstein discusses various "errors" and "corrections" in the data, but all of these have been shown here to have no effect or only a trivial effect on the results. Rothstein's own tables show the same lack of effect. So, why raise these points at all? Rothstein's style of discussion, which raises a long series of issues that he claims to be important but turn out not to be, is consistently misleading. Again and again, his Comment gives the impression that an important error has occurred or an important change needs to be made when, in fact, his own estimates show nothing of the kind. There is not a single case, however, in which he alerts the reader to the fact that an issue he has raised—sometimes at length—has negligible consequences or no consequences at all. Instead, his discussion suggests that he has found a slew of errors and misjudgments in Hoxby (2000)—all of which affect its results to some degree. What his discussion disguises is the sharp discontinuity that really characterizes his results. The reasonable changes that he suggests produce results that differ trivially or not at all from Hoxby (2000). Only when he introduces major errors does he generate results that differ meaningfully. As readers may confirm for themselves (since the replication dataset is available), the results in Hoxby (2000) can be replicated well and are robust to a wide array of reasonable changes in the variables and sample.