Reading Rocket: A Game-Based Reading Level Test for Children Based on Stealth Assessment

Assessing a child’s reading level is an essential first step in language arts education, and ideally it would be accurate, fast, and fun. The current test, in which students read increasingly difficult passages aloud and answer comprehension questions, can be boring for students and teaches and requires hours of staff time per student. The PI and team have developed a simple video game called Reading Rocket suitable for middle school classrooms. We hypothesize that, by observing a variety of fine-grained behaviors, supervised machine learning algorithms can accurately predict reading level via indirect—or stealth—assessment. A pilot study to collect initial data is underway. We propose to develop this pilot study into a robust data-driven educational assessment tools framework capable of supporting a large number of students and additional games. Working closely with the PI and a New Orleans education expert, two undergraduate research assistants will analyze the data collected and train a computational model to predict reading level. They will develop a server-side framework to accommodate new players and further improve the model as more data becomes available. Finally, they will critically analyze which metrics had the most significant impact on the model’s accuracy to identify strategies for developing future game-based learning assessments. This project will provide valuable research experience for two undergraduate research assistants, develop accurate, fast, and fun educational assessment tools, and lay the groundwork for research which can be published in leading academic conferences and which will be attractive to competitive federal funding agencies