Carnegie Mellon University

Searching for mediation models in intelligent tutoring systems data: representational understanding enhances representational fluency - but not vice versa: Martina Rau

Abstract: Conceptual understanding of representations and perceptual fluency in using representations are important aspects of expertise. However, little is known about how these competencies interact: does representational understanding facilitate learning of representational fluency (understanding hypothesis), or does representational fluency enhance learning of representational understanding (fluency hypothesis)? We analyze log data obtained from an experiment that investigates the effects of intelligent tutoring systems (ITS) support for understanding and fluency in connection-making between fractions representations. The experiment shows that instructional support for both representational understanding and fluency are needed for students to benefit from the ITS. In analyzing the ITS log data, we contrast the understanding hypothesis and the fluency hypothesis, testing whether errors made during the learning phase mediate the effect of experimental condition. Finding that a simple statistical model does not the fit data, we searched over all plausible causal path analysis models. Our results support the understanding hypothesis but not the fluency hypothesis. Our findings make predictions that are empirically testable, namely, that instructional materials should support representational understanding before representational fluency.