Longitudinal relationship between academic staffs’ evaluation score by students and their characteristics: Does the choice of correlation structure matter?

Each semester, students are asked to evaluate the academic staff through an online questionnaire. Generalized estimating equations model (GEE), taking into account the correlation between scores, is the established tool to analyze longitudinal data. The aim of this manuscript is to identify characteristics that influence staff score and to address the importance of selection of appropriate correlation structure. We analyzed scores of 336 staff in six consecutive semesters applying GEE with three correlation structures: exchangeable, autoregressive, and unstructured. We also compared the performance of these correlation structures via simulation study. Three normally distributed outcomes with exchangeable correlation structure were simulated. Four independent variables (two continuous and two binary) of which only one was related to the outcome were generated. In the empirical data set, time and academic degree were positively correlated with staffs’ score. Our simulation study showed that the probability that autoregressive and unstructured correlation structures select wrong predictors as being significant is 1.3% and 3.7%. We concluded that evaluation of staff by students improved the quality of education. In addition, selection of inappropriate correlation structure affects the significance of variables.