Murdoch University Research Repository

Welcome to the Murdoch University Research Repository

The Murdoch University Research Repository is an open access digital collection of research
created by Murdoch University staff, researchers and postgraduate students.

Learn more

Predicting student attrition with "big data": Considering demographic, study-based and psychological factors using two large datasets

Williams, C., de Freitas, S. and Dziurawiec, S. (2018) Predicting student attrition with "big data": Considering demographic, study-based and psychological factors using two large datasets. Australian Psychologist, 53 (Supp. 1). pp. 68-69.

Free to read:
*No subscription required


Aim: Given the enormous range of detrimental consequences that student attrition poses (e.g., reduced job prospects for individuals and increased levels of societal inequality), it is essential to understand what factors increase the risk of university withdrawal. However, even though one-in three students attrite from their studies, research in this field remains methodologically and conceptually limited (e.g., focused on single disciplines). Furthermore, these studies are generally restricted to a consideration of demographic variables, without considering a wide-range of psychological factors that may be even more predictive. Ultimately, this presentation will draw upon two large databases to better understand which demographic, study-related, and psychological factors increase attrition risk. Design/ Method: This research employed “data-mining” techniques (e.g., regression/ decision-trees) with an institution-wide record of 24,928 students (n = 7,058 complete cases), spanning across nine disciplines. This data allowed for a large-scale analysis to determine which demographic and study-based variables predicted actual graduation/attrition outcomes. However, as this institutional record does not consider psychological factors (e.g., individual differences), data from a convenience sample of 2,451 students (representing 15 disciplines/40 universities) was also employed. While this dataset is cross-sectional in nature (and thus does not include actual attrition/retention outcomes), it has identified psychological variables that predict self-reported withdrawal intentions. Results: A wide array of findings will be discussed (with each result offering an individual set of implications). For example, within the institution-wide dataset, thirteen variables were assessed (and the predictive model offered 80% accuracy). Only two demographic variables were significant predictors, alongside three of the four study-related variables (e.g., GPA). Surprisingly, students' year-12 entrance exam scores (tertiary rank) did not reach statistical significance. From the national dataset, ~20 demographic and psychological predictors of withdrawal intentions were entered into a multiple-regression model. Almost two-thirds of these variables reached statistical significance (in accordance with theoretically-based hypotheses), and ultimately, burnout, low levels of reward/recognition, and perceived stress were found to have the largest predictive effects (all p < .001). Conclusion: Alongside the practical implications of each separate result, this research highlights the benefits of analysing existing institution-wide datasets. Furthermore, the benefits of collecting regular/ongoing psychological-based data (e.g., burnout levels) should not be overlooked. This psychological data should also be collected alongside actual retention/attrition outcomes (rather than solely withdrawal intention, to account for this study's limitation). Such an approach may ensure that “at-risk” students are offered appropriate support (e.g., instruction- and/or counselling-based), but this must be done ethically (without discriminating against students).

Item Type: Journal Article
Murdoch Affiliation(s): School of Psychology and Exercise Science
Publisher: Taylor and Francis
Copyright: © 2018 The Australian Psychological Society.
Item Control Page Item Control Page