A decision support system to predict student retention in higher education: A multinomial choice logit model

The issue of retention in higher education and the economic impact of lost revenue are of serious concern for administrators. Various studies have been completed in an attempt to understand the variety of causes surrounding why students leave college prematurely. However, efforts to both understand retention and to develop programs to counteract it have not resulted in improving the overall retention rate. This continued loss of students is an economic drain on institutions of higher education and the consequences are a continued spiraling of costs for students and parents.; Mansour (1994) developed a strictly mathematical model that iterates two or more variables simultaneously and identifies student profiles from a mass of data collected by each institution. He found in his study that this model could predict graduation rates within an acceptable statistical error rate for predictor models. This study attempted to verify whether that model is a valid predictor of undergraduate retention in a second study at a four-year public institution of higher education.; Eleven categories of demographic data for 709 student records were collected from two schools within a four-year institution for the cohort year of 1996. The results of this study were similar to those of Mansour's. In this study, the model output resulted in four significant student profiles with a complexity of four variables and at an acceptable error rate of between 2.8% and 6.52%. The error rate was calculated by the difference between the actual graduation rate and the predicted rate. Mansour's (1994) study produced six significant profiles of five variables each and an error rate between 3.35% and 4.44%.; The results of this study confirm that the Mansour-Ferrante model (Mansour, 1994) is an accurate predictor of retention which may be of interest to higher education administrators for early identification of at-risk students. This study also found this mathematical model to be a valid generic predictor.; This mathematical predictor model appears to be a promising tool for higher education in early identification of at-risk students and is a valid generic predictor that could also have applications in other fields.