<p>We investigate the use of Bayesian methods to identify relevant predictors of freshman student attrition using student data from 9 academic years (<i>n</i> = 10,921) and six schools at a private 4-year institution in upstate New York. The proposed framework builds hierarchical binary logistic regression models to compute the posterior probability distributions of models’ parameters and derived metrics using Markov chain Monte Carlo (MCMC) techniques. The paper describes the proposed methods and shows how to formulate hierarchical generalized (Bernoulli) linear models and implement them in a probabilistic programming platform. Tests were conducted to evaluate model fitness and quality, as well as parameter convergence and the significance of the regression parameter estimates. The research identified college academic performance, financial need, gender, and participation in the work-study program as having a significant effect on the log odds of freshmen attrition. The study also found significant heterogeneity in predictor effects across student subgroups when considering interaction effects, as well as fluctuations over time and across schools. We contrasted these results with an equivalent frequentist mixed-effects model and examined interaction effects among key predictors. The Bayesian approach exhibits better parameter stability and offers clearer probabilistic interpretation, especially for groups with limited observations. We believe that the methodology and findings offer valuable guidance for administrators and policymakers, providing a roadmap for applying Bayesian methods to educational data in various institutional contexts.</p>