Comparison of Logistic Regression and Generalized Estimating Equations Models for predicting Student Persistence

• In Event: AERA Roundtable Session Sunday 11:45 am - Gold 1
  In Roundtable Session: Roundtable 6

Abstract

This study compares logistic regression (LR) and generalized estimating equations (GEE) in modeling student persistence across seven semesters using institutional data from a Midwestern university (N = 17,058). While LR is commonly used in institutional research, it assumes independent observations, an assumption violated in longitudinal data. GEE accounts for within-student correlations, providing more precise statistical inference. The results showed the estimates from LR and GEE were highly similar, suggesting that LR may be sufficient for exploratory analyses. Findings also revealed that first-generation and Pell-eligible students consistently had lower odds of persistence, while gender differences appeared only in the first term. These findings highlight the need for higher education leaders and policymakers to prioritize support for historically underserved student groups.

Author

• Meng-Jia Wu, Loyola University Chicago