Abstract

Undergraduate student dropout poses academic, social, and economic challenges worldwide. Although this issue is particularly relevant during the first year of undergraduate studies, it has broader implications for institutional effectiveness and students’ academic trajectories. Since dropout rates vary across fields of study and are particularly pronounced in science, technology, engineering, and mathematics (STEM), this study focuses on undergraduate programs in these disciplines. In this work, we investigate the proportion of students who drop out during the first year of undergraduate programs in Brazil. Understanding the behavior of first-year dropout rates may contribute to the development of strategies to improve student retention in these areas. From a statistical perspective, dropout rates present a challenging modeling setting because they are bounded and may exhibit a high frequency of zero values, corresponding to programs with no dropout, as well as occasional values equal to one, corresponding to programs in which all students drop out. Depending on the application, the response variable may therefore take values in [0,1), (0,1], or [0,1]. Inflated models provide a suitable framework for such data by combining a continuous distribution for values strictly between zero and one with discrete components at one or both boundaries.
Recent developments include inflated models based on the beta, Kumaraswamy, unit Lindley, and unit Weibull distributions. These models can be formulated as regression models that incorporate explanatory variables not only for location measures, such as the mean, median, or quantiles, but also for the probabilities of occurrence at the boundary points. We explore these inflated unit regression models to investigate the behavior of first-year dropout rates in Brazilian STEM undergraduate programs. Potential explanatory variables include the proportions of students by age group, race, and gender, as well as the shift and location in which the program is offered and organizational characteristics of higher education institutions. By appropriately modeling dropout rates and their boundary behavior, we aim to identify patterns associated with student dropout and provide statistical evidence that may contribute to more targeted retention strategies and interventions.
This is a joint work with Kállin Miola Albernaz (Universidade Federal de Santa Maria, Brazil) Tatiane Fontana Ribeiro (Universidade Federal do Paraná, Brazil) , Fernando Peña-Ramírez (Universidade Federal de Santa Maria, Brazil), Fernanda Lang Schumacher (The Ohio State University, United States), and Rachel Figard (University of Georgia, United States).