*Names in bold indicate Presenter
We employ a unique dataset that matches the standardized test scores and attendance records of individual Boston Public School students with real estate records indicating whether the student lived at an address involved in foreclosure and whether that student’s parent or guardian was the owner of the property or instead a tenant. We merge that dataset with data on neighborhood and school characteristics from the American Community Survey, the City of Boston, and Boston Public Schools.
The core of our statistical analysis involves estimation of multilevel models in which measures of individual academic achievement are regressed on student characteristics, indicators of whether the student was directly affected by foreclosure, whether the student moved to a new residence, and whether the student changed schools, in addition to neighborhood effects and school effects. Neighborhood and school effects are modeled as functions of observed characteristics and unobserved factors. Among the neighborhood characteristics included in the model is a time-varying measure of foreclosure activity in the student’s neighborhood of residence, and among the school characteristics is a measure of the extent to which foreclosures have affected the school’s student population. Alternatively, we can include a measure of foreclosure activity in the vicinity of the school, which may differ from the student’s home neighborhood under Boston’s school assignment policy. In order to identify a causal effect of neighborhood (or school) foreclosure rates on student performance separately from effects of unobserved neighborhood (or school) characteristics that predict both foreclosure rates and performance, we will rely on time variation in foreclosure activity within schools and neighborhoods, and also explore alternative identification strategies.