Richard DiSalvo



Working papers

Drinking Water Contamination and Infant Health: Evidence from Pennsylvania 2003-2014 [PDF]

We study the relationship between drinking water contamination and infant health in Pennsylvania, using panel data on births and water quality from 2003 through 2014. Our methodology replicates and extends the within-mother approach pioneered by Currie et al. (2013), who focus on New Jersey over the period 1997-2007. We are able to directly compare many of our estimates to theirs. In our replication, we find that drinking water contamination, when measured using public water system water quality violations, has effects on low birth weight that are similar to those from the earlier study, and we estimate effects on pre-term birth that are larger. However, these conclusions are not robust to controlling for plurality. Moreover, and in contrast to their findings, maternal mobility in our sample is very responsive to drinking water violations, and births in our sample that are exposed to an MCL violation are more likely to be from mothers that are less disadvantaged by multiple demographic measures. Given the inconclusiveness of our findings for violations, we turn to the sampling data that underlies these violation, and uncover precise negative effects of drinking water contamination. The sampling data also allows us to study effect heterogeneity by trimester and type of contaminant.

Grade configurations and student performance: evidence from recent national data [PDF] [Github]

Which school district grade configuration is ideal is an important question that school administrators must be prepared to answer. Research to date had provided rigorous evidence on this question by focusing on student test performance and using within-student variation. However, administrators may be reasonably concerned about how generalizable these findings are to their own context, whether the implications for test scores hold for plausibly more important outcomes such as graduation rates, and whether the within-student effects imply comparable across-district effects. This paper uses recent national data to bring new evidence to bear on these three questions. We identify effects on test scores using within-district across-grade variation in student performance and school terminal grades. This has the advantage that the estimates can be obtained using public national data, which maintains replicability and provides substantial statistical power to identify heterogeneous effects. We find that, while grade transitions should be a concern for low-performing districts, there is evidence that transitions are not as important for test performance in the nation’s highest performing districts. Moreover, we find that grade 5 to 6 transitions are more deleterious for student performance than grade 6 to 7 transitions, especially in low-performing districts. Focusing on that transition in particular, we fit propensity score weighted regressions, comparing school districts that used grades 5 to 6 transitions heavily, to those that did not, matching on grades 3 and 4 test performance, demographics, and resources. Our preferred results suggest that having a 5th grade terminal configuration does not have a large effect on graduation rates, dropout rates, or national 8th grade test score percentile rank, and we can reject effects on these outcomes more severe than -0.2pp, +0.1pp, and -0.2pp, respectively.

Work in progress

Suspension reduction and the school-to-prison pipeline: evidence from apparent policy changes

Preliminary and incomplete -- findings may change as we evaluate our data cleaning and modeling choices. Do not cite.

Out-of-school suspension (OSS) is a widely used, though controversial, means of disciplining students in American schools. To inform the debate, we use a new approach to estimate the typical effects of school and school district policy changes that directly target suspension reduction, using national data from the Office of Civil Rights over the period 2000 through 2014. Due to the absence of comprehensive national data on official school policies, we attempt to infer likely policy changes by identifying large, abrupt jumps in reported OSS rates. We study the characteristics of schools and districts that are more likely to have engaged in these “apparent policy changes.” Then, we use these apparent policy changes as an instrument in a difference-in-differences framework to estimate the average impacts of these policy changes on the schools and districts that have made them. We evaluate effects on school test scores, school district graduation rates, and city-level youth arrest rates. We estimate effects on test scores that are similar to the matching literature; estimated effects on dropout and arrest rates may be large, but our estimates are imprecise.