Refining Learning Maps with Data Fitting Techniques: Searching for Better Fitting Learning Maps

Learning sciences needs quantitative methods for comparing alternative theories of what students are learning. This study investigated the accuracy of a learning map and its utility to predict student responses. Our data included a learning map detailing a hierarchical prerequisite skill graph and student responses to questions developed specifically to assess the concepts and skills represented in the map. Each question aligned to one skill in the map, and each skill had one or more prerequisite skills. Our research goal was to seek improvements to the knowledge representation in the map using an iterative process. We applied a greedy iterative search algorithm to simplify the learning map by merging nodes together. Each successive merge resulted in a model with one skill less than the previous model. We share the results of the revised model, its reliability and reproducibility, and discuss the face validity of the most significant merges