February 14, 2019
HPTP Data Note 10: Transfer Research and Multilevel Models
Structural, financial, and information barriers at both two- and four- institutions complicate the process of student transfer. Current research on the topic does not adequately address the shared responsibility between these institutions, or examine possible differences in transfer success rates by race. In Data Note 10 of the High-Performing Partnerships (HPTP) study, CCRI researchers contribute to filling these research gaps by examining the unique contribution of institutional pairs on transfer student outcomes. The researchers also argue for the use of multi-level models to avoid spurious results in this complex and important area of study.
As part of the HPTP study, researchers Grant Blume and Elizabeth Meza dig into the statistical models used to study college transfer. Empirically, they explore the extent to which hierarchical generalized linear models (HGLMs) allow for control of within-pair correlation and student characteristics. In doing so, their goal is to isolate institutional pairs associated with the highest and lowest odds of baccalaureate attainment.
The researchers focused their exploration on three states included in the national Credit When It’s Due Initiative (CWID), which is the basis of all the HPTP data notes. Their findings suggest that transfer research dependent on linear models may lead to biased results and incorrect conclusions. Spurious results may emerge from a hypothesis that a student-level characteristic such as race or socioeconomic status is associated with baccalaureate attainment if researchers fail to account for how such characteristics vary across pairs and within pairs.
As an example, the researchers present their findings from one of the states that suggest the use of a hierarchical model isolates a statistically significant negative association between Latinx students and the odds of baccalaureate attainment. The generalized linear model fails to show a statistically significant relationship but the fully specified HGLM model suggests the odds of baccalaureate attainment are 22.8% lower for Hispanic students compared to all other students in the sample.
The study presents a robust and viable identification strategy on which future theoretical and empirical work can build. Blume and Meza argue more precision in estimating baccalaureate-degree attainment for institutional pairs within states may be an important step forward in more carefully identifying transfer partnerships that inform policy and practice.
For more information about their models and findings, read the full Data Note below.
This Data Note is part of CCRI’s High-Performing Partnerships Study (HPTP) funded by the Bill and Melinda Gates Foundation. The study focuses on how higher performing transfer collaborations between two and four-year colleges and universities work on the ground. Researchers identified high-performing partnership pairs from a dataset collected for the national initiative on reverse credit transfer called Credit When It’s Due (CWID). Read the full series of Data Notes and more about the project here.