Dana Goldstein’s final column for The American Prospect looks at some examples of murky evidence in the education policy realm. The point is well-taken, but I think in one specific example she uses — charter schools — the policy solution is pretty clear:
In the past six months, two high-profile studies of charter schools, both out of Stanford University, have attracted significant media attention. The first, a study of charter schools in 16 states conducted by CREDO, an education research group affiliated with the university, found that in math, only 17 percent of charters increase achievement over traditional public schools. The report’s authors called the results “sobering.”
The second, a close look at 75 New York City charter schools by education economist Caroline Hoxby, a Stanford professor and Hoover Institute fellow, couldn’t have drawn a more disparate conclusion. Hoxby’s study, comparing students who win charter school lotteries to those who lose them, found that New York charter students do 31 points better in math and 23 points better in English than their lottery-losing peers, who remain in neighborhood public schools.
Whatever you think of the methodological dispute here (Goldstein explains it well and I guess I side with Hoxby) the crux of the matter is that there’s substantial variation in the performance of different charters. What you need to do is identify schools that consistently perform poorly and shut them down. Then you create space for more effective models to replicate themselves and also for new ideas to be tried out. The promise of charter schools is that by allowing more experimentation we’ll find some good models. But it’s not as if public education in the United States currently achieves some theoretical maximum of badness — with experimentation we’re also discovering bad models. You’ve got to shut those models down, while at the same time curbing state legislators’ tendency to impose arbitrary numerical caps on the total quantity of charter schools. We should let a thousand flowers bloom and then kill 20–30 percent of them if they turn out to look ugly.