UC Irvine

As educational institutions increasingly adopt digital tools for daily operations, unprecedented amounts of data are generated at different levels of the education system. The granularity of these big data makes it possible to understand and support educational processes in a data-informed, easy-to-scale manner, and educational data science (EDS) has emerged as a nascent field to realize this potential. This dissertation specifically focuses on the promise of EDS to address issues related to educational equity, a central theme of education research. To begin with, a two-dimensional taxonomy is presented to characterize equity-oriented EDS research -- whether the work is focused on explanation or prediction, and whether the problem of interest takes a micro- or macro-level perspective of education research. The interaction of these two dimensions partitions the research space into four quadrants, and one empirical study in higher education contexts is presented to illustrate each quadrant. The first two explanatory studies leverage novel data sources (i.e., digital behavioral traces) to understand systematic sociodemographic gaps in 1) peer interaction experience in virtual learning environments (micro-level), and 2) academic engagement over time at the institutional level (macro-level). The latter two prediction-oriented studies investigate algorithmic fairness from the perspectives of 1) choice of data sources in online learning performance prediction (micro-level), and 2) use of sensitive attributes in early warning systems (macro-level). These studies highlight how EDS research can advance the understanding of existing educational inequalities and guide preventive action against potential inequities. Finally, recommendations for future research on equity-oriented EDS are discussed.