Assessing Students’ Learning when Interpreting Histograms: A Gaze-Based Machine Learning Analysis

Students consistently misinterpret histograms. Statistics education literature suggests that solving dotplot items can support correct histogram interpretations. We therefore explore how students’ micro-level histogram interpretations alter during assessment, with the research question: In what way do Grades 10–12 pre-university track students’ histogram interpretations change after solving dotplot items? Students were asked to estimate or compare arithmetic means. Students’ gaze data, answers, and stimulated recall interview data were collected. We used students’ gaze data on four histogram items as inputs for a machine learning algorithm (MLA; random forests). Our MLA can quite accurately classify if students’ gaze data belong to an item solved before or after the dotplot items. Moreover, we found that the direction (e.g., almost vertical) and length of students’ saccades were different on the before and after items. A change in this perceptual form could therefore indicate a change in strategies. Two more indications of actual learning were found. This study is novel in three ways: a novel use of spatial gaze data, use of a MLA for finding differences in gazes that are relevant for changes in students’ topic specific strategy and the first that investigates students’ micro-learning during an assessment. We consider a most likely explanation for the results that the action of solving dotplot items creates readiness for learning and that reflecting on the solution strategy during recall then brings new insights. This study is important for theories on readiness for learning and practice effects, and it has implications for large-scale assessments and homework.