Association for Public Policy Analysis & Management

Though nationally they now stand at all-time highs, there are significant disparities in U.S. high school graduation rates, with large gaps among student subgroups defined by race, gender and poverty. This paper analyzes how early signs, in terms of course-taking and credit accumulation in beginning high school grades, can be a valid predictor of later high school outcomes. Can early academic momentum propel students to high school success, and what is the potential for using this diagnostic tool in an early intervention? We exploit detailed longitudinal data from New York City Public Schools, the largest school district in the country whose diversity affords us a unique opportunity to analyze heterogeneity in outcomes across all four major racial groups (Hispanics, Blacks, Whites and Asians). Using individual student level data specifically obtained from the New York City Department of Education, we chart out the course-taking history of high school students. The extensive data allow us to control not only for various background variables but also prior academic history, including middle school scores and schools attended. We put particular emphasis on STEM subjects and advanced courses. The presence of Regents examinations provides us with an invaluable validity check in terms of non-graduation outcomes. We exploit both the timing of Regents examinations and scores therein to investigate how course-sequencing patterns interact with these, and how these can be jointly exploited to track students who are on path to graduation and college.

One main motivation here is to investigate the role of individual high schools in mediating this relationship, against the backdrop of a comprehensive high school choice program. The large number of New York City public high schools – during the period of analysis, incoming high school students could choose from 700 individual programs in 400 schools – enables us to study heterogeneity in multiple dimensions. We not only disaggregate high schools by the nature of their programs and admissions rules, but also explore interactions of these with student background. We predict student outcomes in high school and document the significant variance that exists in subsequent outcomes even conditional on past achievement. Separating out students whose actual performance falls below expectations from students whose performance consistently exceeds expectations, we identify school- and individual-level factors which are associated with such divergent experiences. We also run sensitivity analyses in terms of diversity of courses taken in high school and potential peer effects in sequencing of courses and credits earned.

These results have important policy implications. Not only do they highlight important components of human capital acquisition in high school, but by underlining essential ingredients for successful high school outcomes they provide individual schools and families with a concrete roadmap for future academic success. They also have broader significance, pointing towards the role playable by predictive analytics as big data come to the forefront. This is particularly true as the theme of the conference relates to the use of better data for better decision-making - this paper illustrates the potential of the same for improving high school performance.