Determining possible gender differences in undergraduate introductory programming courses by applying data analysis techniques in the results of the students’ assignments, to promote gender equality

Abstract

Over three quarters of a century have passed since women gained their right to work in a manner that permits their financial independence, yet the stereotypes regarding the kinds of professions they should prefer continue to exist. One of the domains where this observation still applies is Computer Science, a field traditionally considered to be aimed predominantly towards a male audience. But, does the real-world data provide evidence that could explain the perpetuation of this trend, or does it happen due to reasons of non-empirical nature? The current study aims to contribute by examining the case of an introductory course on Data Structures from the second semester of the undergraduate studies program of the Department of Applied Informatics of University of Macedonia. It was conducted in the years 2021 and 2022 and the city of origin is Thessaloniki, Greece. The chosen course being one with mandatory programming assignments means that it is of very high importance and also one of a certain amount of student failure. Novice programmers are susceptible to facing difficulties with programming courses and these courses often have the highest dropout rates. A per gender analysis of the compilation errors found in students’ assignments along with their assignment grades and the final course grades is performed. Considering the numerical disparity between male and female students in the department, a better understanding of how the results compare between the two genders is interesting, on the basis of it constituting empirical evidence on how or even if a student’s gender and their academic performance relate. The analysis of the collected data is done through learning analytics and more precisely through visualization and statistical analysis tests on the programming errors and grades per student. The aim is to monitor the performance per gender throughout the semester and determine possible differences that may exist. Additionally, association rule mining and clustering are applied in order to reveal potentially contributing factors for a student in passing the course, as well as whether the students can be divided into groups that share specific characteristics. The data is also used for creating models that attempt to predict whether the student will pass the course or not based solely on data from the assignment assessment files. The findings suggest that the students’ gender does not particularly change their performance characteristics, while the results of the two genders usually were deemed not significantly different. Interestingly, whenever any differences emerge, it is women that seem to be the dominant gender in terms of academic performance. Lastly, it turns out that it is possible to predict student success or failure on the course at a level that is fairly acceptable to decent by using exclusively each person’s number of errors and their assignment grade average. This can even be mostly done as early as by the first third of the semester or with higher success by its middle.