Sample Design for an Evaluation of the Reading First Program
This paper illustrates how to design an experimental sample for measuring the effects of educational programs when whole schools are randomized to a program and control group. Questions addressed by the paper include: How many schools should be randomized? How many students per school are needed? What is the best mix of program and control schools? And how do data on aggregate school-level measures of past student performance or individual student-level measures improve the precision of program impact estimates? Empirical analyses based on extensive data from two urban school districts are used to address each question, and the statistical theory underlying these analyses is presented in an appendix. The paper was prepared to help design the evaluation of a national educational program, and it was circulated by the U.S. Department of Education as methodological background for two recent Requests for Evaluation Proposals.