The Contribution of Administrative and Experimental Data to Education Policy Research (Forum: Innovations in Education Policy and Education Research‪)‬

INTRODUCTION The last decade has seen a surge in empirical research by economists addressing the impact of school reform policies. This wave of research dates back to Card and Krueger (1992) and a series of analyses that look across the United States to test the effects of school inputs on outcomes such as achievement, educational attainment, and earnings. Many of these studies use either Census data (Heckman, Layne-Farrar, and Todd, 1996) or national longitudinal surveys (Belts 1996; Loeb and Bound, 1996; Grogger, 1996) to estimate input effects. Since then, much of the research has turned from the nation to states and cities, using local administrative and experimental data to focus on specific policy initiatives. (1) The shift in emphasis from the national to the local has enabled researchers to incorporate greater institutional and policy detail into their analyses. Local data often follow individuals over time, allowing researchers to use empirical techniques not possible with pooled cross-sectional Census data. At the same time, the administrative data provide deep coverage of local areas, which national surveys that sample only a small portion of any locality, do not. While local administrative data provide advantages over census and survey data for estimating the causal effects of some policies, experimental data alleviate the perennial concern that correlational analyses have not established causality. However, experiments are not easy to implement in education, they can't be used to address all types of education policy, and they tend to be too small to elicit the responses generated by large-scale reforms.