Panel Paper:
Predicting Demand for Schools: Using Data to Inform School Planning Decisions
*Names in bold indicate Presenter
Indeed, many school planning decisions are currently made with very limited data on how they would affect applications to and enrollment in the corresponding schools, or the effect on other schools nearby. At the school and LEA level, leaders must decide on curriculum, programmatic offerings, and such policies as whether to require uniforms. At the systemic level, LEAs must also make choices about its capital budget, where to improve facilities or replace them, whether to close schools, relocate schools, expand capacity, or re-purpose vacant space. Each of these factors have the potential to affect the number of students applying for enrollment or the number of students the school can enroll. Using quantitative data to inform these decisions, such as whether the market for a particular type of school offering is saturated, or whether a particular neighborhood or area of the city is saturated, can lead to allocations of resources that better align school offerings and capacities with student applications.
Using a combination of data from DC’s unified application and enrollment system and detailed characteristics of schools, neighborhoods, and transportation options, we evaluate the feasibility of simulating schools’ future application volumes in response to various school planning decisions. First, we test the relative performance of different statistical models including random forests, Bayesian hierarchical generative model, and more traditional discrete choice models in predicting demand for enrollment in DC schools of choice. We then describe the accuracy with which these models simulate demand for enrollments in response to changes that schools enacted in subsequent years. Finally, we conclude with a discussion of practical ways for stakeholders to harness data from unified enrollment systems to drive decision making and the additional information gained in doing so.