Predicting state test scores better with intelligent tutoring systems: Developing metrics to measure assistance required

Abstract

The ASSISTment system was used by over 600 students in 2004-05 school year as part of their math class. While in [7] we reported student learning within the ASSISTment system, in this paper we focus on the assessment aspect. Our approach is to use data that the system collected through a year to tracking student learning and thus estimate their performance on a high-stake state test (MCAS) at the end of the year. Because our system is an intelligent tutoring system, we are able to log how much assistance students needed to solve problems (how many hints students requested and how many attempts they had to make). In this paper, our goal is to determine if the models we built by taking the assistance information into account could predict students' test scores better. We present some positive evidence that shows our goal is achieved. © Springer-Verlag Berlin Heidelberg 2006.