Author:
Eitel Lauría
Affiliation:
School of Computer Science and Mathematics, Marist University, Poughkeepsie, New York, U.S.A.
Keyword(s):
Bayesian Methods, Probabilistic Programming, Freshmen Attrition, Higher Education Analytics.
Abstract:
The study explores the use of Bayesian hierarchical linear models to make inferences on predictors of freshmen student attrition using student data from nine academic years and six schools at Marist University. We formulate a hierarchical generalized (Bernoulli) linear model, and implement it in a probabilistic programming platform using Markov chain Monte Carlo (MCMC) techniques. Model fitness, parameter convergence, and the significance of regression estimates are assessed. We compared the Bayesian model to a frequentist generalized linear mixed model (GLMM). We identified college academic performance, financial need, gender, tutoring, and work-study program participation as significant factors affecting the log-odds of freshmen attrition. Additionally, the study revealed fluctuations across time and schools. The variation in attrition rates highlights the need for targeted retention initiatives, as some schools appear more vulnerable to higher attrition. The study provides valuabl
e insights for stakeholders, administrators, and decision-makers, offering applicable findings for other institutions and a detailed guideline on analyzing educational data using Bayesian methods.
(More)