Abstract
Data visualizations are powerful tools for communicating about quantitative information. Nevertheless, how people interpret graphs has long been studied separately from how people produce effective ones. Here we jointly investigate how people answer questions using graphs and how people judge which graph would be informative for answering those questions. We explored the hypothesis that even non-experts would be sensitive to those properties that make a graph not only interpretable in general, but informative for helping other people answer the specific question at hand. On each trial in the graph-production experiment, participants recruited via Prolific (N=129) were presented with a question about a dataset (e.g., “What is the average petal length of Virginica flowers?” for the Iris dataset). They then judged which of eight graphs varying in several ways (e.g., level of aggregation, ordering of x-axis values) would be most useful to someone else trying to answer that question. We next recruited a separate group of Prolific participants (N=167) to participate in a graph-comprehension experiment. On each trial, they were shown a graph and a question. While the question always referred to variables in the dataset used to generate the graph, sometimes a variable mentioned in the question did not appear in the graph. Participants then judged whether the graph contained the information needed to answer the question and, if so, to type in their response. We found that participants in both experiments could discriminate which graphs contained the minimum information necessary to answer each question (p<0.001), establishing a basic degree of graph literacy in this population. Beyond that, however, production participants were not sensitive to more subtle differences in how well two graphs plotting the same data would actually support comprehension by others (p=0.694), suggesting a potential site for graph literacy interventions.