Toward the theme of equity-oriented educational data science, I work with novel forms of “big data” (e.g., behavioral traces, online discourse) generated by learners, educators, and institutions in real-world contexts, and conduct three parallel strands of research:
- Combine different data science techniques to generate fine-grained understanding of learning, development and equity;
- Design algorithmic decision support pipelines to improve learning, development and equity;
- Evaluate and improve the fairness, accountability, transparency, and ethics of educational data science applications.
Most of my empirical work focuses on young adults in low-resourced educational contexts, such as minority-serving, fully online, and broad-access institutions.
New projects are always on the way!
Longitudinal modeling of behavioral inequality
We investigate modeling strategies that convert longitudinal digital behavioral traces to a rigorous understanding of how educational inequality accumulates through students' day-to-day experience. We look at different types of digital behavior, such as task completion, forum discussion, and content navigation.
Contextualized predictive analytics
Contexts matter for predictive models, but how? We study theory-informed and data-driven contextualization strategies to improve the robustness and generalizability of educational predictive analytics at scale. We also rigorously evaluate the effects of deploying predictive analytics on achievement, well-being, and equity through field experiments in low-resourced institutional contexts.
Algorithmic fairness in transfer learning
We audit and mitigate algorithmic bias in the context of cross-institutional transfer learning. From there, we build infrastructure for low-resourced institutions to collaboratively benefit from algorithmic decision support in achieving their efficiency and equity goals.