The Poisson and negative binomial regression models with zero-inflated count: An application to educational data

Investigation through the number of failed courses and semesters in medical students is a method for assessing students’ performance. Since these amounts have a zero inflated (ZI) value in most students, use of Poisson and negative binomial distributions with ZI, in which can model count data to find the associated factors and estimate the parameters, is recommended. This cross-sectional study was performed between 2008 and 2009 in Iran University of Medical Sciences (IUMS) on 670 sample students who were selected using stratified sampling. The educational and demographic data was collected from University site and related resources. The study was approved in the IUMS Research Council and Board of Ethics and accordingly the students’ data was investigated confidentially. For data analysis SPSS and Stata softwares were used. Up to Poisson regression and negative binomial distributions with ZI for dependent variables of the number of failed semesters, students’ total averages and the enrolment quota had the most roles in the model. Dependent variables of the number of failed course, total average and undergraduate and master levels had the most effect in both models. In ZI regression model for the number of failed semester, the total average and enrolment quota type have important roles so that increasing the average yields the decreasing of the number of failed semester. Moreover, moving from the other quota types to open-for-all enrolment quota yields a decrease in the number of failed semesters. In regressions with the number of failed courses, the average has an important effect to both models and its increase yields a decrease of the number of failed courses. Also, the undergraduate and master levels cause a decrease in the number of failed courses.