The Effects of the Flint water crisis on the educational outcomes of school-age children

Abstract

In 2014, the municipal water source in Flint, Michigan was switched, causing lead from aging pipes to leach into the city's drinking water. While lead exposure in Flint children increased modestly on average, some children were exposed to high lead levels. Surveys of Flint residents show the water crisis was also associated with increased levels of stress, anxiety, and depression. We use Michigan's administrative education data and utilize synthetic control methods to examine the impact of the crisis on Flint's school-age children. We find decreases in math achievement and increases in special needs classification, even among children living in homes with copper (rather than lead) water service lines. Low socioeconomic status students and younger students experienced the largest effects on math achievement, and boys experienced the largest effects on special needs classification. Our results point toward the broad negative effects of the crisis on children and suggest that existing estimates may substantially underestimate the overall societal cost of the crisis.

Fig. 1.. Timeline of the Flint water crisis.

Fig. 2.. Mean educational outcomes over time in Flint.

This figure displays descriptive trends in the mean academic outcomes for the Flint geographic district from 2006 to 2019. Data are taken from the Michigan Department of Education’s longitudinal administrative database. Black dotted lines represent the start of the posttreatment period (i.e., the first year of data after the FWC had begun). Math and reading achievement (A and B) are observed for only for grades 3 to 8 and are standardized within test subject, grade, and year to the overall state distribution scores. Math and reading achievement observations begin in 2007. Both special needs (C) and attendance (D) are observed in grades K-12, and special needs status observations begin in 2006, whereas attendance observations begin in 2009.

Fig. 3.. Synthetic control sample selection.

Each circle represents a Michigan geographic school district. Data are taken from the Michigan Department of Education’s longitudinal administrative data base. All districts with enrollment greater than 1000 students are displayed, and the size of each school’s circle is proportional to its student enrollment. The highlighted districts represent the three different district samples used in our primary synthetic control model (A) and various robustness checks (B and C). In (A), dashed lines are at X = 0.73 and Y = 0.31, the 90th percentiles of fraction economically disadvantaged and fraction black, respectively. In (B), dashed lines are at X = 0.63 and Y = 0.11, the 75th percentiles. Last, in (C), dashed lines in (B) are at X = 0.8 and Y = 0.54, the 95th percentiles. All variables were measured in the 2013–2014 academic year. While 42 of the districts are the same across the ∪90 and the ∩75 districts samples, there are 23 districts that appear in one but not the other.

Fig. 4.. Synthetic control estimates of the effect of the FWC on student outcomes.

Average treatment effect estimates of the causal effects of FWC on standardized math and reading academic achievement (A and B), fraction special needs (C), and fraction of school days attended (D) are plotted over time. 2015 is the first year post-treatment. A treatment effect was estimated for each year in the post-period (2015–2019). The grayed shaded area represents the 95% confidence interval of the treatment effect estimates. Synthetic Flint was constructed by taking a weighted average of the districts listed in Table 2.

Fig. 5.. Visualizing synthetic control robustness checks.

Average treatment effect estimates of the causal effects of FWC on standardized math and reading academic achievement (A and B), fraction special needs (C), and fraction of school days attended (D) are plotted over time. 2015 is the first year post-treatment. Results from our primary synthetic control model is displayed in dark blue, and the other colored lines represent results from alternative synthetic control models used as robustness checks. See Materials and Methods for a detailed description of each alternative synthetic control model.

Fig. 6.. Mean educational outcomes over time in Flint by service line material.

This figure displays descriptive trends in the mean academic outcomes for the Flint geographic district from 2010–2019. Education data is taken from the Michigan Department of Education’s longitudinal administrative data base. The blue lines display students living in homes with copper service lines, while the red lines display students living in homes with lead service lines. Service line material data was collected by the City of Flint’s service line inspection and replacement program implemented following the crisis. The black dotted vertical line represents time that the FWC begins (i.e., the first year in the post-period). Math and reading achievement (A and B) are observed in only grades 3 to 8, whereas special needs (C) and attendance (D) are observed in grades K-12.