Data from a moderately-selective state university in the Midwest is used to cross-examine the most appropriate data analytical techniques for predicting versus explaining college student persistence decisions. The current research provides an overview of the relative benefits of models specializing in prediction versus explanation with particular emphasis on estimation methodologies, model specification by estimation technique, and model diagnostics, including classification tables and measurements of the goodness of fit. The predictive validity of a model of college student retention estimated using logistic regression is compared with that of discriminant analysis estimated using cognitive and non-cognitive predictors of retention. Key contributions to the literature include a unique analysis sample, a unique set of independent variables, and statistical estimation methodologies that build upon traditional frameworks, including machine learning techniques. The currents study ends with a discussion that will allow leaders in higher education and policymakers to make better data-informed decisions surrounding prediction and explanation so that they can proactively intervene with the most appropriate attrition-minimization policies.