Competition Among Public Schools: A Reply to Rothstein (2004)

Rothstein has produced two comments, Rothstein (2003) and Rothstein (2004), on Hoxby "Does Competition Among Public Schools Benefit Students and Taxpayers," American Economic Review, 2000. In this paper, I discuss every claim of any importance in the comments. I show that every claim is wrong. I also discuss a number of Rothstein's innuendos--that is, claims that are made by implication rather than with the support of explicit arguments or evidence. I show that, when held up against the evidence, each innuendo proves to be false. One of the major points of Rothstein (2003) is that lagged school districts are a valid instrumental variable for today's school districts. This is not credible. Another major claim of Rothstein (2003) is that it is better to use highly non-representative achievement data based on students' self-selecting into test-taking than to use nationally representative achievement data. This claim is wrong for multiple reasons. The most important claim of Rothstein (2004) is that the results of Hoxby (2000) are not robust to including private school students in the sample. This is incorrect. While Rothstein appears merely to be adding private school students to the data, he actually substitutes error-prone data for error-free data on all students, generating substantial attenuation bias. He attributes the change in estimates to the addition of the private school students, but I show that the change in estimates is actually due to his using erroneous data for public school students. Another important claim in Rothstein (2004) that the results in Hoxby (2000) are not robust to associating streams with the metropolitan areas through which they flow rather than the metropolitan areas where they have their source. This is false: the results are virtually unchanged when the association is shifted from source to flow. Since 93.5 percent of streams flow only in the metropolitan area where they have their source, it would be surprising if the results did change much. The comments Rothstein (2003) and Rothstein (2004) are without merit. All of the data and code used in Hoxby (2000) are available to other researchers. An easy-to-use CD provides not only extracts and estimation code, but all of the raw data and the code for constructing the dataset.