Multistage Modeling from Application Signals to Downstream Success: Predicting Admission, Matriculation, and Retention

The student journey in degree programs can be viewed as a pipeline: applicants must be admitted, admitted students must matriculate, and matriculants must persist and perform. Using administrative records from over 50,000 applicants to an online graduate computing program, we model these three stages as distinct outcomes: admission, matriculation among admitted students, and longer-run success among matriculants. For each stage, we estimate multivariable logistic models using structured application attributes (e.g., age, citizenship, application history, undergraduate GPA) and high-dimensional text features from self-reported academic and employment histories (TF-IDF). We then compare success models based on application data alone versus application data augmented with first-term performance and enrollment patterns. Across stages, models reveal different correlates of advancement through the pipeline: application features strongly differentiate admission and yield, while long-run success is only weakly explained at application time but becomes highly explainable when we include first-term academic momentum. Together, the results clarify what can be inferred at each decision point and highlight the outsized role of early-course performance for retention-oriented interventions.