An experiment tested statistically lay decision makers’ use of error bars in a graphreading task. Participants viewed two-point dot plots, with each point representing asample mean. Across conditions, means were accompanied by error bars of differentsizes. Graphs were described as plots of consumer product ratings, and participants wereasked to judge whether the products represented in each graph were rated differently orabout the same. Signal detection analysis showed no influence of error bars onparticipants’ sensitivity or bias, and on average performance was poorer and far moreliberal than predicted by a simple decision rule that classified the two data points asdifferent if their error bars did not overlap. Results suggest that non-expert graph readersmake little use of error bars in drawing conclusions from visualized data.
展开▼