Data on academic performance of students are often hierarchical or multilevel in nature, because education systems have a hierarchical structure (students nested within college, college nested within department and so on). The hierarchical nature of data on academic performance often introduces multilevel dependency that can have implications for model-based statistical inference. Multilevel modeling approach has been suggested for analyzing data with hierarchical nature. This paper examines the predictors of academic performance (measured by grades A, B, C and D) of graduate students of Sultan Qaboos University (SQU), Oman, using multilevel modeling approach. The analysis shows that there is a real multilevel variation among students’ grades. The results indicate that multilevel logistic regression model fits better over the ordinary logistic regression models. The analysis revealed that students’ age, gender, colleges and region of residence are important predictors of students’ grades in SQU. The effect of college variations after controlling the effect of age of students, gender of students and region of residence further implies that there exists a considerable deference in students’ grades among colleges. The study suggests that researchers should use multilevel models rather than ordinary regression methods when their data structure is hierarchal.

multilevel modeling, academic performance, hierarchical, logistic regression, Oman.