Thinking with causal models: A visual formalism for collaboratively crafting assumptions

Abstract

Learning Analytics (LA) is a bricolage field that requires a concerted effort to ensure that all stakeholders it affects are able to contribute to its development in a meaningful manner. We need mechanisms that support collaborative sense-making. This paper argues that graphical causal models can help us to span the disciplinary divide, providing a new apparatus to help educators understand, and potentially challenge, the technical models developed by LA practitioners as they form. We briefly introduce causal modelling, highlighting its potential benefits in helping the field to move from associations to causal claims, and illustrate how graphical causal models can help us to reason about complex statistical models. The approach is illustrated by applying it to the well known problem of at-risk modelling.