This study, focusing on default avoidance of non-federal institutional (school sponsored) student loans, developed and tested statistical models exploring maximum student access to post-secondary education in context of minimum institutional financial risk. Independent variables were collegiate academic progress, socio-economic status defined by receipt of Federal Pell Grants, and academic preparation as defined by required and completed remedial courses. Nearly 1000 cases were gathered from one community college, fall 1996 and spring 1997. A model building exercise led to clusters of reasonable models. Supporting access while protecting from excessive default was the balance throughout. Independent variables were in some instances highly correlated, requiring judgment regarding collinearity. To understand models, statistics valued in assessment of model fit were calculated. Because all models under-predicted default, ability to predict defaulters was ranked high. Logistic regression analysis was favored. During initial tests, models predicted from 88% to 90%. Using SCORENEW, a SAS ® macro, cross validation was undertaken, fall 1997 and spring 1998 extracted from another near 1000 cases. Using this method, coefficients from first data sets--were used to recalculate second set predictions; overall prediction varied from 88.1% to 94%. Models erred maximizing access, heightening concern that the institution would continue absorbing financial risk. Models were ineffective for new students. They improved default rates restricting borrowing from 0% to 3%. Non-federal loan importance as a component of public policy was implied by providing numbers of students access. Factors including SES, high school preparation, and collegiate academic achievement can be used to predict default avoidance.
