The study involves a college that evolved from an extensive attrition prediction model with a robust dataset to a trimmed down model with a sparse dataset. This paper the compares effectiveness of this parsimonious model with an extensive model in predicting attrition. Authors find that the trimmed down model possesses improved predictive capability for persisters but lower predictive accuracy for dropouts and students overall than the extensive model. Study results surprisingly suggest that student loan amounts and being commuter students are positively associated with persistence, while SAT verbal/reasoning score is negatively associated with persistence. Authors investigate use of pooled annual data versus split training and confirmation data samples in testing the attrition model. The study offers insight to educational researchers and administrators on the predictive sufficiency of using a relatively small set of easy to attain student background and academic variables in predicting attrition.