Using multilevel logistic regression analysis to predict graduation rates at colleges and universitie

This study compared the performance of the Multilevel Logistic Regression (MLogR) procedure to the performances of the Ordinary Least Squares (OLS) regression procedure and the Traditional Logistic Regression (TLR) procedure in terms of predicting college graduation rates, assessing the effects of student- and school-level characteristics on those college graduation rates, and ranking schools based on those unique school effectiveness indicators provided by each of the three statistical methods. The data set, provided by the National Collegiate Athletic Association (NCAA), consisted of graduation, demographic, academic, and school information for a sample of 9,558 student athletes enrolled in 220 NCAA division I schools during the time period of fall 1984 to summer 1990. A cross-validation design was used to compare the predictive powers of each of the modeling techniques.;Results of this research indicate that the multilevel logistic regression analysis procedure is worth considering when the focus of research is on gaining a better understanding of how fixed or random student and school level effects influence student outcomes, or when considering ranking schools in terms of their effectiveness. When the focus of research involves optimizing the prediction of an outcome, the multilevel logistic regression analysis procedure is no better than the more familiar regression techniques of ordinary least squares regression or traditional logistic regression analysis.