Graphs of averages exaggerate and sow disagreement

Scientific communication must be both valid (accurate) and reliable (consistent). We show that visualized averages fail on both counts, producing severely exaggerated and inconsistent interpretations. Using a novel drawing-based response measure, participants sketched the data points that they believed underlay plotted averages. Across 40 replications, exaggeration and inconsistency proved pervasive—occurring across plot types, data content, and education levels. Exaggeration levels were so large as to far exceed the bounds of established science, and variability in interpretations of single graphs was so large as to surpass that seen across entire scientific fields. In three of four tested cases, accuracy was systematically worse than blind guesses. We document a simple, effective remedy: display individual data points along with the mean. Contrary to concerns that this added detail would confuse viewers, it yielded greater accuracy, stronger consensus, and improved subjective evaluations. We therefore recommend the routine inclusion of raw data as a practical solution to the serious flaws of plotted averages that we have documented repeatedly here.