Herrmann L (2024)
Publication Language: English
Publication Type: Thesis
Publication year: 2024
On their way from school to the labor market, a comparatively large proportion of young adults drop out of post-secondary education, including higher education (HE) or vocational education and training (VET). This is still perceived as problematic by the public and in politics, as the personal consequences of dropping out and the impact on the demand for skilled workers are a matter of concern. However, the term "dropout" is often used, even though it is not a definitive exit from post-secondary education, which may portray the situation too severely. Research has already discovered a lot about the processes involved in dropping out. Furthermore, many characteristics of individuals are known that increase their risk of dropping out. This thesis is dedicated to this topic and contributes by adding an innovative perspective to previous ap-proaches. The focus of the work and the included papers, firstly, is "soft" factors and their influence on dropout from post-secondary education. "Soft" factors are to be understood as less tangible, subjective aspects of a situation or context that are difficult to quantify or measure objectively. Nevertheless, they can significantly impact outcomes. The soft factors are part of a complex structure of dropout reasons and process elements, which is why the thesis, sec-ondly, takes an in-depth look at the potential of methods from artificial intelligence and machine learning. Their ability to recognize patterns in complex interrelations can be valuable when researching dropout, so they are examined for suitability in the context of empirical educational research. Non-traditional students in this thesis serve as examples of specific groups of people in post-secondary education who exhibit certain inherent characteristics regarding dropout.
Paper I (Chapter 2) assumes an interplay of multiple reasons for dropping out of HE and uses data mining to delineate homogeneous groups within this complexity. The cluster analysis is carried out for non-traditional students who drop out disproportionately often. The results show that various reasons for dropping out were particularly decisive for different subgroups. For one group, these were financial reasons, for another, family commitments and a third group had difficulties meeting the requirements. One particularly striking group states all reasons as relevant, while two others showed no clear pattern. This emphasizes the need to provide spe-cific support measures to meet the different needs of students.
Paper II (Chapter 3) covers dropout from HE and VET and analyses comments on YouTube videos on this topic. Text mining methods identify latent topics and sentiments in the texts. The application of Topic Modeling shows that the comments on VET dropouts often discuss finan-cial aspects and working conditions, while less clear topics are discernible in the case of uni-versity dropouts; instead, it is more about the decision to drop out itself and the future perspec-tives. Contrary to the assumption that dropping out sends a negative signal, the sentiments are, on average, neutral to positive, with comments on dropping out of HE being slightly more positive than for VET.
Paper III (Chapter 4) examines the influence of various aspects of information on dropping out of VET, which is analyzed using event history data. The results show that if trainees are well informed about their own training program, they are less likely to drop out. However, knowledge about other training programs increases the risk. Vocational preparation at school can reduce the risk of dropping out if it is perceived to be of high quality. Contrary to the assumption derived from network theory that weak ties, in particular, could be more helpful than strong ones, the various groups of people from whom trainees can obtain information have no significant influ-ence on dropout risk.
Paper IV (Chapter 5) uses a case study to demonstrate how the probability of students drop-ping out can be predicted using examination data and their socio-demographic data. The use of various machine learning methods is already widespread, but the risk of algorithmic bias must be taken into account. The exemplary analyses show that although the algorithms display good predictive accuracy for the student population as a whole, they are less able to predict the failure of non-traditional students.
The main findings suggest that dropping out of HE and VET is influenced by various factors. These range from financial difficulties and family commitments to the information available to trainees. The studies show that different groups of students and trainees exhibit different drop-out patterns, which emphasizes the need for tailored support measures. Furthermore, the pub-lic perception of dropouts may indicate less negative stigmatization than previously assumed. Although machine learning algorithms provide promising results in predicting dropout, predic-tion is less accurate for non-traditional students and requires consideration of algorithmic bias. Overall, the results show the complexity of dropout phenomena and emphasize the importance of differentiated interventions and a comprehensive support strategy to successfully guide stu-dents and trainees through their educational pathways, whereby machine learning can provide support.
APA:
Herrmann, L. (2024). Exploring innovative perspectives on dropout from post-secondary education (Dissertation).
MLA:
Herrmann, Lisa. Exploring innovative perspectives on dropout from post-secondary education. Dissertation, 2024.
BibTeX: Download