Estimating Treatment Effects from Contaminated Multiperiod Education Experiments: The Dynamic Impacts of Class Size Reductions

Abstract

This paper introduces an empirical strategy to estimate dynamic treatment effects in randomized trials that provide treatment in multiple stages and in which various noncompliance problems arise, such as attrition and selective transitions between treatment and control groups. Our approach is applied to the highly influential four-year randomized class size study, Project STAR. We find benefits from attending small classes in all cognitive subject areas in kindergarten and first grade. We do not find any statistically significant dynamic benefits from continuous treatment versus never attending small classes following grade 1. Finally, statistical tests support accounting for both selective attrition and noncompliance with treatment assignment. © 2010 The President and Fellows of Harvard College and the Massachusetts Institute of Technology.

The heterogeneous impact of pension income on elderly living arrangements: evidence from China’s new rural pension scheme

“…To formally test for attrition bias due to observables, we follow Becketti et al (1988) and Ding and Lehrer (2010), and regress the baseline living arrangement on baseline individual characteristics and their interactions with a binary indicator for attrition, using the full sample of wave 2008/09:…”

Understanding the role of time-varying unobserved ability heterogeneity in education production

“…In future research we hope to extend the methodology described in this paper to develop an estimable panel data model in which the individual effect has multiple components and each of 33 Similarly, estimates of causal impacts from Project STAR differ based on the assumptions researchers use to handle violations to the experimental protocol (e. g. Krueger (1999) compared with Ding and Lehrer (2010)). …”

“…Further, since teachers were re-randomized to classrooms each year, we can obtain unbiased estimates of the effects of both current and past teacher characteristics. 17 Third, this data set reduces measurement error from aggregation bias by precisely matching each student to the classroom and the teacher within a school, so that we can focus on estimates of the time-varying impacts of individual unobserved ability. Finally, Project STAR was conducted for a single cohort of children between Kindergarten to grade 3, stages in the lifecycle child development specialists have suggested either the impact or stock of cognitive ability is malleable.…”

A review of research on Project STAR and path ahead

“…In order to address nonrandom noncompliance, Ding and Lehrer (2010) treated the experiment as a multiperiod intervention and accounted for attrition due to observables and the possibility that other forms of noncompliance were due to unobservables. Because of selective attrition and noncompliance with the treatment assignment, they selected only students who had participated in STAR for the entire experimental period.…”