LAUSR

Digital games can make unique and powerful contributions to K-12 science education, but much of that potential remains unrealized. Research evaluating games for learning still relies primarily on pre- and post-test data, which limits possible insights into more complex interactions between game design features, gameplay, and formal assessment. Therefore, a critical step forward involves developing rich representations for analyzing gameplay data. This paper leverages data mining techniques to model learning and performance, using a metadata markup language that relates game actions to concepts relevant to specific game contexts. We discuss results from a classroom study and identify potential relationships between students’ planning/prediction behaviors observed across game levels and improvement on formal assessments. The results have implications for scaffolding specific activities, that include physics learning during gameplay, solution planning and effect prediction. Overall, the approach underscores the value of our contextualized approach to gameplay markup to facilitate data mining and discovery.