Drawing Valid Inferences From The Nested Structure Of Engineering Education Data: Application Of A Hierarchical Linear Model To The Succeed Longitudinal Database

Although hierarchical linear models are seldom used in engineering educational research, the nested structure of students in various colleges of engineering and the longitudinal nature of student records supports the use of such models. Hierarchical linear models account for the nested structure and can test hypotheses on both the schools and the students within the schools simultaneously, thereby eliminating aggregation bias and misestimated standard errors that result when the nested structure is ignored. In the present study, a hierarchical linear model is fitted to the SUCCEED longitudinal database using only students that graduated. As an example, cumulative GPA is regressed on Carnegie school classification, school setting, degree received, gender gap, and citizenship gap with SAT total score and number of terms attended as covariates. The results indicate that there is significant cumulative GPA variance between schools, accounting for 19% of the variance. Additionally, the gender gap and citizenship gap accounted for 6% of the within school cumulative GPA variance, but school setting accounted for 61% of the between school citizenship gap variance. In particular, students that receive their degree in engineering had the highest cumulative GPA. Non-citizens tended to have higher cumulative GPAs than citizens. Another finding is total SAT score is more predictive of cumulative GPA in urban schools than suburban schools. Finally, urban and/or research schools had the strongest relationship between number of terms until graduation with cumulative GPA in that longer times to graduation are associated with lower cumulative GPA.