Panel Paper: Reconciling Experimental and Quasi-Experimental Evidence On the Impact of Full-Day Kindergarten

Friday, November 9, 2012 : 8:00 AM
Salon B (Radisson Plaza Lord Baltimore Hotel)

*Names in bold indicate Presenter

Chloe Gibbs, University of Chicago

This paper addresses the question of how to interpret competing evidence on the impact of full-day kindergarten, and provides guidance on how this evidence taken in tandem may inform the design of full-day kindergarten policies. Incorporating both experimental and quasi-experimental estimates on program impact, the study capitalizes on student assignment policies that allocated oversubscribed full-day kindergarten slots based on random lotteries and fixed cut-points on kindergarten readiness assessments, testing the causal impact on students’ literacy skills at the end of the kindergarten year. The data employed comes from school districts and schools in Indiana that employed such assignment policies in the 2007–2008 school year.

While nearly all school-age children in the United States attend kindergarten, only 60 percent of kindergarten students are in full-day classrooms. One promising avenue for increased intervention in early childhood is through full-day kindergarten. Experimental findings suggest that students who are assigned to—and those who participate in—kindergarten in a full-day setting outperform their peers in half-day settings (0.31 standard deviations) in the same schools. In particular, I find that nonwhite, predominately Hispanic students benefit (0.52 s.d.) from full-day kindergarten in comparison to their half-day kindergarten peers. These heterogeneous treatment effects have implications for narrowing or closing the achievement gap early in formal schooling, and in fact constitute 117 percent of the control group's race/ethnicity gap. Using rough cost measures, a simple cost-effectiveness analysis suggests a range of effect sizes from 0.07—0.21 s.d. per thousand dollars of spending.

In contrast, results from the districts that employed fixed cut-points to assign the most academically needy students to full-day kindergarten—implemented at the same time and in the same state as the lottery districts—find no program impact. The regression discontinuity design generates an estimate of the local average treatment effect (LATE) by comparing the outcomes of students above and near the cut-score (in half-day kindergarten) with those below and near the cut-score (in full-day kindergarten). Over 70 percent of students participated in the setting corresponding to their pre-test score. I find that students in full-day kindergarten near the margin of interest do not outperform their half-day kindergarten counterparts. Importantly, interpreting these findings in conjunction with the lottery evidence requires consideration of heterogeneous treatment effects across the distribution of kindergarten entry literacy skills, homogeneous versus mixed ability grouping of students, and peer effects under different student assignment policies.

To address these issues, I will discuss the context in which treatment occurs under the different assignment policies and the mechanisms through which peer effects may operate differently. Comparisons of similar students—i.e., those in the middle of the pre-test distribution—across the two designs will be made. I will present evidence from the lottery districts that suggests that full-day kindergarten impact is concentrated in the upper and lower tails of pre-kindergarten literacy skills, and will generate regression discontinuity estimates from the lottery data for comparison. Ultimately, I will focus on lessons for design of full-day kindergarten policies, including targeted versus universal provision.