AUTHOR=Quigley David , McNamara Conor , Ostwald Jonathan , Sumner Tamara TITLE=Using Learning Analytics to Understand Scientific Modeling in the Classroom JOURNAL=Frontiers in ICT VOLUME=4 YEAR=2017 URL=https://www.frontiersin.org/journals/ict/articles/10.3389/fict.2017.00024 DOI=10.3389/fict.2017.00024 ISSN=2297-198X ABSTRACT=

Scientific models represent ideas, processes, and phenomena by describing important components, characteristics, and interactions. Models are constructed across various scientific disciplines, such as the food web in biology, the water cycle in Earth science, or the structure of the solar system in astronomy. Models are central for scientists to understand phenomena, construct explanations, and communicate theories. Constructing and using models to explain scientific phenomena is also an essential practice in contemporary science classrooms. Our research explores new techniques for understanding scientific modeling and engagement with modeling practices. We work with students in secondary biology classrooms as they use a web-based software tool—EcoSurvey—to characterize organisms and their interrelationships found in their local ecosystem. We use learning analytics and machine learning techniques to answer the following questions: (1) How can we automatically measure the extent to which students’ scientific models support complete explanations of phenomena? (2) How does the design of student modeling tools influence the complexity and completeness of students’ models? (3) How do clickstreams reflect and differentiate student engagement with modeling practices? We analyzed EcoSurvey usage data collected from two different deployments with over 1,000 secondary students across a large urban school district. We observe large variations in the completeness and complexity of student models, and large variations in their iterative refinement processes. These differences reveal that certain key model features are highly predictive of other aspects of the model. We also observe large differences in student modeling practices across different classrooms and teachers. We can predict a student’s teacher based on the observed modeling practices with a high degree of accuracy without significant tuning of the predictive model. These results highlight the value of this approach for extending our understanding of student engagement with scientific modeling, an important contemporary science practice, as well as the potential value of analytics for identifying critical differences in classroom implementation.