Modeling Learning in Online Settings

Over the last decade, digital learning environments such as MOOCs have become a central fixture of the educational landscape. The adoption of such pedagogical approaches has proved highly popular, with millions of enrollments worldwide. While the proliferation of MOOCs has provided fertile grounds for research, the field has often been criticized for being primarily observational and lacking in appropriate rigor. In the literature, this shortcoming has been partially attributed to the lack of theoretically informed approaches employed in the analysis of data.

To address this shortcoming we are developing computational and theoretically grounded methods to analyse both student's interactions with their courses (such as estimating how "engaged" students are) and their interactions with eachother (for instance, to what extent are student's social networks predictive of their final grade).

Student engagement is a frequently used construct in educational research and practice. While engagement is often considered an overarching construct in the field of education, the term has been subjected to a variety of interpretations and there is little consensus regarding the very definition of the construct. This raises grave concerns with regards to construct validity: namely, do these metrics measure the same thing?

To date, the research into engagement in online learning settings has been hindered by the lack of common understanding regarding how engagement should be defined and measured in the context of learning at scale. However, recent theoretical research proposes a quadripartite model, and provides suggestions as to how this may be operationalised. This is a critical first step, and our investigation, using factor analysis, structural equation modeling, and MIMIC modeling, has identified and validated a factor structure that coheres with the predictions of our theoretical model. This work is currently being prepared for publication. However, our analysis was run on a limited dataset, and so a second step will be to conduct similar analysis on a significantly larger dataset, which has recently become available. This subsequent analysis will also expand upon our previous work and use a clustering methodology that we previously published to cluster online courses on the basis of their design, and thus investigate the extent to which such contextual factors mediate student engagement.

Thereafter, we intend to return to social network analysis for which we are currently developing a method for the extraction of structural and neighbourhood related features. Preliminary results from university course enrollment data suggests that up to 16% of the variance in student grades may be explained by these features. Our goal is to apply such methods to online learning environments and, combined with our ongoing investigation of engagement, analyse the association between cliques, faciliation techniques (such as tutor intervention), and learning outcomes.