Measuring Learning Progress via Self-Explanations versus Problem Solving - A Suggestion for Optimizing Adaptation in Intelligent Tutoring Systems

Prior studies have shown that learning by problem solving in an intelligent tutoring system such as the Cognitive Tutor can be even more effective when worked examples are added in comparison to problem solving alone. Introducing self-explanation prompts additionally improves learning. Furthermore, recent findings indicate that fading out worked examples according to students’ performance during learning (i.e., adaptive fading) is even more beneficial than fading worked examples in a predefined procedure (i.e., fixed fading). In this contribution we investigate the relationship between potential indicators for learning progress, which can be used for adapting fading and, thereby, fostering learning outcome. We found a stronger relationship of learning outcomes to self-explanation performance than to problem-solving performance during learning. Additionally, self-explanation performance is a stronger predictor for learning outcome than prior knowledge. Hence, adaptation, not only of the example fading procedure but potentially of other aspects of student learning (e.g., individualized problem selection) might better be based on self-explanation performance and not, or at least not only, on problem-solving performance, as it is typical of intelligent tutoring systems.