Scientific modeling: Using learning analytics to examine student practices and classroom variation

Modeling has a strong focus in current science learning frameworks as a critical skill for students to learn. However, understanding students' scientific models and their modeling practices at scale is a difficult task that has not been taken up by the research literature. The complex variables involved in classroom learning, such as teacher differences, increase the difficulty of understanding this problem. This work begins with an exploration of the methods used to explore students' scientific modeling in the learning sciences space and the frameworks developed to characterize student modeling practices. Learning analytics can be used to leverage these frameworks of scientific modeling practices to explore questions around students' scientific models and their modeling practices. These analyses are focused around the use of EcoSurvey, a collaborative, digital tool used in high-school biology classrooms to model the local ecosystem. This tool was deployed in ten biology classrooms and used with varying degrees of success. There are significant teacher-level differences found in the activity sequences of students using the EcoSurvey tool. The theoretical metrics around scientific modeling practices and automatically extracted feature sequences were also used in a classification task to automatically determine a particular student's teacher. These results underline the power of learning analytics methods to give insight into how modeling practices are realized in the classroom. This work also informs changes to modeling tools, associated curricula, and supporting professional development around scientific modeling.