Predicting university entry using machine-based models and solutions

Increasingly, governments and grant bodies around the world are funding large databases of longitudinal data on young people as they transition from adolescence into adulthood. They are often put together by multidisciplinary teams including economists, sociologists, educators and psychologists and have led to considerable advancements in theory within these fields. Nevertheless, aspects of these databases remain underutilized. In particular, belying their conception, research flowing from these databases tends to be discipline-specific and consists of a small subset of variables. This is consistent with a dominant focus in social science research on explanatory science at the cost of predictive science. However, advances in machinelearning algorithms mean that there are possibilities to leverage the broad multidisciplinary nature of these databases to build models that can be used to predict important transition outcomes like university entry. We illustrate various approaches, using over 100 variables from the Longitudinal Study of Australian Youth (LSAY) collected when participants (N = 6,363) were 15 years of age to predict university entry three years later. We also consider what insights the various approaches provide to theory. While not a replacement for rigorous testing of causal explanations, machinelearning approaches provide a powerful additional tool for developmental researchers with important real-world applications.