Proceedings of the Annual Meeting of the Cognitive Science Society

We developed and experimentally tested three versions of an interactive statistics simulation designed to foster a qualitative understanding of how sample size influences statistical estimation and inference. Guided by a grounded cognition framework, two versions used perceptually rich, manipulable tokens (Token and Token-to-Rectangle) and were compared to a conventional Histogram version. In a laboratory study with undergraduates (n = 252), all groups improved similarly from pretest to posttest. Contrary to our hypotheses, however, the token-based (grounded) conditions showed less productive in-the-moment engagement during simulation use; that is, interactions were more often directed toward perceptually salient token operations rather than sampling-relevant actions. Grounded conditions also showed weaker performance on within-activity conceptual items. These findings suggest that grounding can be seductive rather than supportive in the moment, even when dynamic and perceptually rich representations are aligned with key concepts and avoid noninformative decoration.