Association for Public Policy Analysis & Management

Recently, a number of districts have introduced human capital reforms intended to improve the quality of teaching either through compositional change in the workforce or by improving the extant teaching workforce.  Recent empirical work estimates that these compositional changes could have significant effects on students’ long-term economic well-being: in their seminal paper, Chetty, Friedman and Rockoff find that replacing a teacher in the bottom five percent of the performance distribution with an average teacher for one year increases students’ lifetime income by $250,000 per classroom in present discounted value (2014).  Over the past six years, the District of Columbia Public Schools (DCPS) has operated IMPACT, a robust, district-wide teacher evaluation system with uniquely high-powered incentives including large financial bonuses and performance-based dismissal.  Over the first three years of the program IMPACT led to the forced separation of more than 4 percent of the teacher workforce, and hundreds of teachers have voluntarily separated after receiving a low evaluation rating (Dee and Wyckoff, 2015).  Theoretically, increasing the exit rates of poorly performing teachers will result in improved student achievement, provided the average performance of the incoming teachers exceeds that of those exiting.  This paper examines how teacher turnover during IMPACT influences student achievement, and how these effects vary across years of implementation and differing school contexts.

Using longitudinal data on teachers and students in tested grades and subjects from 2009-10 to 2012-13, we examine how different forms of turnover (e.g., voluntary attrition or forced separation) influence subsequent student achievement outcomes within the affected school-grade cells. This approach parallels methods employed by Chetty, Friedman and Rockoff (2014) as well as Rivkin, Hanushek, and Kain (2005), Jackson and Bruegmann (2009), and Ronfeldt, Loeb, and Wyckoff (2013).  In general, we suspect teacher turnover is likely to be non-random: the unobserved school traits that influence teacher turnover are also likely to influence student achievement. However, the variation isolated by our approach – the year-to-year variation in teacher turnover within given school-by-grade cells – may be a plausibly exogenous determinant of student achievement.  To increase our confidence that IMPACT caused the student achievement changes we observe through changes to teacher quality we also estimate the “first-stage” effect of turnover on IMPACT scores as well.  In preliminary analyses we find evidence that IMPACT-induced exits improve math teacher performance in a school-grade cell by more than a standard deviation in the next year and improve student achievement outcomes in math by 20 percent of a standard deviation in high poverty schools.  Results in reading are qualitatively similar, but smaller.  Because teacher turnover during the IMPACT era is closely related to its design features, these results provide encouraging preliminary evidence that IMPACT has shaped teacher quality and student performance through changes to the composition of the teacher workforce in DCPS.