Predicting retention of first-year college students

The purpose of this study was to investigate demographic and academic variables that may contribute to the persistence of students at a mid-size, public, four-year institution in southeast Florida. A discriminant analysis was performed using 16 demographic (gender, age, ethnicity, permanent address location, residential or commuter status, and types/amounts of financial aid received) and academic variables (major, high school GPA, SAT verbal scores, SAT quantitative scores, ACT composite scores, total credits attempted first semester, and type of orientation program attended) as predictors to differentiate between classification as a persister or a leaver. The predictive accuracy of the analysis was assessed using a one-tailed z test that compared the results of the analysis to the proportional chance expectation (Huberty, 1994). Analyses were also completed to investigate the contributions of individual and subsets of variables on predicting the criterion. The research questions considered were: Can a predictive model based on student demographic and academic variables known to an institution of higher education prior to a student's matriculation correctly classify students as potential persisters or leavers with greater accuracy than chance? Is there a significant relationship between the criterion and any of the individual variables or subsets of variables in the model that correctly classify students as potential persisters or leavers? The findings indicated that the model failed to classify students as persisters better than chance; however, it was able to provide some predictive accuracy in the classification of leavers. In the most parsimonious model, it was found that persisters were more likely to receive moderate student loan amounts and attempted greater numbers of credit hours than leavers. Since a large number of students at this institution expressed an intention to transfer from the onset of their studies and because student intent could not be used as a variable, the predictive accuracy of the model was affected. A recommendation of this study is to link student intent to the data to create a more successful model.