Early Warning System Simulations of High School Dropout Propensity
Thursday, November 12, 2015 : 4:30 PM
Tuttle Center (Hyatt Regency Miami)
*Names in bold indicate Presenter
High school dropout rates are a concerning trend for educators, administrators, education policy makers, and communities. A promising effort to address dropout rates is an empirical approach known as Early Warning Systems (EWS). The EWS is a prediction model that determines an individual student’s risk of dropping out of high school. The prediction model is designed to alert educators and parents to the risk at levels at an early stage. Doing so allows for earlier interventions in order to shift the trajectories of students displaying risks for dropping out. To this point, EWS have been limited in their predictive power because they are typically single-level regression models with limited and often out-of-date sample information. In this study, we produce a simulated sample of students consistent with the existing covariance matrix of variables from Delaware's EWS report (Merola & Fernandez, 2010). This work integrates extant research on EWS models and the educational imperative to predict a student’s likelihood of school dropout. Dropout rates have been shown to significantly impact minority populations; however, many statewide dropout models do not incorporate racial identity as a predictor. In this study, we conduct a simulation model with three continuous predictors (attendance rate, number of disciplinary referrals, and English Language Arts (ELA) final score). In the second model, we then include student race along with the three continuous predictors in order to determine whether or not the model fit is significantly different with inclusion of student race. We believe the use of simulated modeling contributes to EWS development because it allows for testing characteristic or program effects in a way that is not directly tied to individual-level data. The experimental design of simulation studies allows for a level of control that results in a more precise extrapolation of population parameters from sample parameters.
- APPAM_Completed_ForSubmission.pdf (183.7KB)