Effective Early Childhood Education Programs: Meta-Analytic Lessons from High Quality Program Evaluations

Co-Principal Investigators: Greg Duncan (University of California, Irvine), Katherine Magnuson (University of Wisconsin-Madison), and Holly S. Schindler, (Harvard University)

Research Methods and Design: Researchers will use meta-analysis, a method of quantitative research synthesis that uses prior study results as the unit of observation. Researchers will draw on a comprehensive meta-analytic database of early childhood education evaluations that has been compiled by researchers at the Harvard Center on the Developing Child, the University of Wisconsin-Madison, the University of California-Irvine, and Johns Hopkins University. Several prior meta-analyses have examined preschool education evaluations. Researchers differentiate this study from earlier meta-analytic efforts in several respects. First, there is a more specific developmental focus. Researchers will distinguish between cognitive outcomes based on their sensitivity to classroom instruction and consider behavioral outcomes separately. Second, researchers screened more candidate studies than prior meta-analyses and included more updated research. Third, researchers will employ rigorous methodological inclusion criteria and develop specific inclusion criteria for different kinds of quasi- or non-experimental studies (e.g., fixed-effects; difference-in-difference; propensity score and other kinds of matching; regression discontinuity; and some other regression-based approaches). Fourth, the project team will not simply average effect sizes within studies, but also retain effect size information in multi-level models and weight for both number of tests within studies and precision of estimates. Finally, researchers will utilize a broad set of robustness analyses. Although the three research questions have differing emphases, they share overlapping data and analytic models. In particular, researchers will work closely to ensure that analyses are coordinated, and findings for one question are incorporated into the analyses related to the other questions.