Estimating Bar Graph Averages: Overcoming Within-the-Bar Bias

Abstract

Although most people are not aware of it, bias can occur when interpreting graphs. Within-the-bar bias describes a misinterpretation of the distribution of data underlying bar graphs that indicate an average or where the average estimation point moves inside the bar when the average of several graphs is estimated. This study proposes and tests two methods based on information processing to reduce within-the-bar bias. The first method facilitates bottom-up processing by changing various graph features, such as presenting confidence intervals, placing boundaries around the graph, and showing cumulative bars with different tones. The second method facilitates top-down processing by instructing participants to estimate the mean based on a dot at the end of each bar. Testing of the first method showed that cumulative bars reduced bias, but the other methods did not. The second method was found to reduce bias. Overall, our results demonstrate that the accurate interpretation of bar graphs can be facilitated through the manipulation of specific graph features and instruction.

Two graphs walk into a bar: Readout-based measurement reveals the Bar-Tip Limit error, a common, categorical misinterpretation of mean bar graphs

“…Okan and colleagues ( 2018 ) also replicated both the core asymmetry (this time using health data) and the finding that nonbar graphs did not produce the effect; they additionally found that the asymmetry, counterintuitively, increased with graph literacy. Finally, Godau, Vogelgesang, and Gaschler (2016) and Kang and colleagues ( 2021 ) showed that the asymmetry remains in aggregate—carrying through to judgments of the grand mean of multiple bars.…”