Abstract
Though controversial, bar graphs of means (BGoM) remain popular for their presumed accessibility to non-experts. This presumption remains, however, largely untested. In this study we present six principles for optimization of graph comprehension assessment measures, utilize these principles to create the Draw Datapoints on Graph (DDoG) method, and deploy the new method to document a common, severe, qualitative error in BGoM interpretation. We label this identified behavior the Bar-Tip Limit Error (BTLE), because adherents represent BGoM data distributions as limited by, rather than distributed across, the bar-tip. BTLE is observed in ~20% of participants: a rate that is consistent across educational levels, ages, and genders. It persists across those who correctly supply the mean value from a BGoM; those who accurately define the word mean; and, critically, despite explicit and repeated instructions that the bars represent means. Those who exhibit the BTLE once tend to repeat it across bar graphs that vary widely in form and content, suggesting that the error results from a fundamental misunderstanding of BGoM. In a survey of bar graphs from several educational and general sources, we found that the majority showed counts or percentages, as opposed to mean values. We therefore hypothesize that the BTLE reflects an acquired, and expertise-associated, conflation of two bar graph types whereby BGoM are assumed to operate like bar graphs of counts (BGoC). The existence of a severe and prevalent misunderstanding of BGoM suggests the need for evidence-based interventions potentially including point-of-contact annotations of BGoM, replacement with alternative graph types, and changes in curricula and statistical software to emphasize distributional understanding.