Yet another analysis of results from a privately financed experiment with school vouchers in New York City suggests that the program produced significant test-score gains for African-American students.
The findings, unveiled last week by Harvard University researchers Paul E. Peterson and William G. Howell, are the latest in a series of analyses that have yielded different conclusions from the same data.
The program in question, which began in 1997, provides scholarships of $1,400 a year to 1,200 New York public school students so they can attend private schools.
While school choice is a perennial hot topic, the New York effort particularly interested researchers and policymakers because it offered a rare chance to empirically test the effects of vouchers. Since the vouchers were distributed by lottery, researchers could compare the experiences of the winning students with those of students who sought, but did not receive, vouchers.
The full text of Peterson-Howell’s study,is available from the , Harvard University. (Requires .)
The first analysis of the results, which the Harvard researchers conducted with Mathematica Policy Research Inc., showed that while students overall derived no academic benefit from the program, African-American students did make sizable gains. (Feb. 27, 2002.)
That finding was disputed earlier this year by Alan B. Krueger and Pei Zhu, two Princeton University researchers who re-examined the data. Their analysis concluded that the gains found for black students were too small to be significant. (April 9, 2003.)
In the new study, Mr. Peterson and Mr. Howell try to dispel the Princeton researchers’ criticisms of their work by configuring the data dozens of different ways.
“The basic story here is whether you do it one way or another, you get significant, positive impacts for African- Americans,” said Mr. Peterson, a professor of government at Harvard.
By Mr. Peterson and Mr. Howell’s optimal calculations, black students who attended private schools for three years under the program outscored their public school counterparts who weren’t chosen in the lottery by 8.4 percentile points. That gain—equal to about a grade level—is slightly lower than the 9.2 percentile points that the Harvard researchers first reported.
Mr. Peterson said the change reflects some revisions the authors made in their calculations.
Interviewed last week, however, Mr. Krueger said the points the Harvard researchers made in their new study were still “off the mark.” The new findings also failed to sway a group of prominent researchers assembled by the Washington-based Economic Policy Institute via teleconference to answer reporters’ questions about the new study.
Looking at both re-analyses, Cecilia E. Rouse, a Princeton economics and public-policy professor, concludes: “Really, what you take away from this is that the results are basically not as robust as previously characterized.”
A major point of contention between the Princeton analysis and the Harvard study was whether to include students for whom the researchers had no baseline test scores. Adding those students to the mix increases the sample size by 44 percent and reduces some of the original test-score gains.
“The simplest thing to do is to compare those who got vouchers with those who didn’t,” Mr. Krueger said. “The beauty of an experiment is that you should be able to control for anything, and it should not change your results.”
Mr. Peterson and Mr. Howell argue in their new paper, however, for including only students with baseline test-score data—or at least factoring the scores into the calculations. “You’ve got a lot better-quality information if you know kids started out from the same place at the very beginning,” Mr. Peterson said.
The Harvard researchers also took issue with the Princeton scholars’ methods for classifying students’ race and ethnicity. When students were African-American and Hispanic, for example, Mr. Krueger and his co-author classified them as both. The Harvard researchers took the categories to be mutually exclusive.
Mr. Peterson and Mr. Howell also recalculated the race-related numbers several ways, looking at whether a student’s mother, both parents, either parent, or caretaker was black. Results were positive each time, they said.