Predicting Early Fall Student Enrollment

Predicting incoming enrollment is an ongoing concern in school districts with school choice, substantial student mobility, or both. Inaccurate predictions can disrupt learning as districts adjust to enrollment fluctuations by reshuffling teachers and students. This REL Mid-Atlantic study for the School District of Philadelphia compared the accuracy of four statistical models for predicting fall enrollment at the school by grade level, using school-by-grade-level data from prior years, to assess which approach might be useful for planning staffing. Key findings include the following: All four models have similar predictive accuracy. If the goal is to increase predictive accuracy at the typical school, the choice of model is not consequential. Of the 259 predictors assessed in the models, four provide the most meaningful contribution to accurately predicting school-by-grade enrollment: prior cohort size, in-school suspensions, out-of-school suspensions, and absences. Details: Predicting incoming enrollment is an ongoing concern for the School District of Philadelphia (SDP) and similar districts with school choice systems, substantial student mobility, or both. Inaccurate predictions can disrupt learning as districts adjust to enrollment fluctuations by reshuffling teachers and students well into the fall semester. This study compared the accuracy of four statistical techniques for predicting fall enrollment at the school-by-grade level, using data from prior years, to assess which approach might be the most useful for planning school staffing in SDP. The predictions differ little in accuracy: predicted cohort size differs from actual cohort size by roughly six students across all methods The statistical techniques leave much student mobility unaccounted for. Even under the best prediction approach, students and teachers in 22 percent of incoming grade levels within schools might have to be reassigned because of unexpected student mobility and district rules on maximum class size. Predictive accuracy is not meaningfully different in schools with larger proportions of Black students, economically disadvantaged students, or English learner students. Of the 259 predictors analyzed, 4 stand out as the most important: prior cohort sizes, in-school suspensions, out-of-school suspensions, and absences.