In reading the National Institute on Early Education Research report on the positive effects of pre-kindergarten in Arkansas, I found myself confused by the method of statistical analysis, which is called "regression discontinuity." Steve Barnett of NIEER and the main author of the Arkansas study patiently responded to repeated emails from me over the weekend, trying to help me understand. Basically, it's a way to control for selection bias, when it's not possible to establish an appropriate "control group," as would be needed for a truly scientific experiment. RD, as the statistics pros call it, depends on being able to establish two groups that are similar, based on some sort of deciding factor or "assignment variable." One such variable might be a study participant's birthdate, when used to determine eligibility for a program or a "treatment." In this case, the "treatment" is pre-kindergarten.

Anyway, I mention all of this because RD has been used in a number of key evaluations of pre-k programs and probably will be used in lots more. One such study is the frequently cited work of William T. Gormley and co-authors who evaluated the Oklahoma state pre-k program, and found positive effects. Journalists, as we all know, became journalists because they couldn't handle math. But we are increasingly being called on to report on (or to choose not to report on) program evaluations. We ought to know enough about statistics to at least be able to ask good questions.

Barnett sent me a several good links that clearly explain RD: They are here, here, and here.