Major and Institution Transfer Path Combinations and the Relaxed Lasso Method: Parsimonious Prediction of Vertical Transfer Student Success

Authors

• David Wutchiett, CUNY Central OfficeFollow
• A. W. Logue, CUNY Graduate CenterFollow
• Martin Kurzweil, Ithaka S+RFollow
• Colin Chellman, CUNY Central OfficeFollow

Document Type

Article

Publication Date

2026

Abstract

The majors, institutions, and their combinations, which characterize transfer between higher education institutions relate to student success but introduce high-dimensionality and multicollinearity to statistical modeling. Relaxed lasso regression, a procedure blending regularized lasso models with unregularized refit models, selects variables and facilitates unbiased coefficient estimation in high-dimensional settings. Using a large sample of students transferring from associate to bachelor’s programs (n=28,166), we compared lasso and ridge regression parsimonious models predicting bachelor’s graduation and posttransfer GPA improvement, evaluating associations using unregularized parsimonious predictive models. Relaxed lasso parsimonious optimal models outperformed standard lasso and ridge regression models while including few of 702 predictor variables. This study addresses the critical gap concerning overcoming trade-offs between predictive accuracy and interpretable inference in high-dimensional settings.