Dateline: October, 2021, Issue 2

What conclusions do jurors draw from truncated bar charts?

Many trials are full of charts and graphs intended to communicate numerical information clearly and to persuade jurors of the specific facts or conclusions that are illustrated. The graphs and charts can be technically accurate in that they display correct numerical values, while also being misleading if they invite jurors to draw inappropriate conclusions.

Yang and colleagues (2021) conducted five studies that examined the conclusions people draw from bar charts that violate a fundamental principle of data visualization, specifically the "y-axis truncation" of bar charts. A y-axis truncation occurs when the vertical axis of the bar chart starts at a non-zero value. For example, the exact same data are plotted in both of the bar charts below with the lowest bar on both having a value of 100 and the highest bar on both having a value of 105. The two bar charts differ only in the start value of the vertical axis: the y-axis starts at 97 in the bar chart on the left and at 0 in the bar chart on the right (hover over or touch the bar charts to enlarge them).

Y-Axis Truncation

Yang and colleagues made 40 bar charts communicating information about a range of topics such as public health, geography and technology. Two versions of each bar chart were created: one where the y-axis was truncated and one where the y-axis started at zero. Participants were shown one version of each of the 40 bar charts with 20 being truncated and 20 non-truncated. The researchers then conducted 5 studies to explore various questions related to the conclusions people draw from truncated and non-truncated bar charts.

Across all 5 studies, when bar charts used y-axis truncation, research participants consistently perceived the illustrated differences as larger than when identical data were illustrated in bar charts with a y-axis starting at 0. Fully 83.5% of research participants across the five studies evidenced this truncation effect. In other words, the vast majority of participants consistently judged differences illustrated by truncated bar charts to be larger than differences illustrated by charts where the y-axis started at 0.

The truncation effect was incredibly persistent. People were misled by y-axis truncation even when provided thorough explanations of the technique right before they rated one of the two versions of each of the 40 bar charts created for the research. The explanations of the truncation effect described y-axis truncation, provided an example, warned that some bar charts were created to be misleading, showed participants two bar charts of the same data (one with y-axis truncation, one without), asked participants to identify which bar chart had been designed to be misleading (which 88% answered correctly), and provided feedback about the correct answer. This pre-exposure instruction was not effective. The pre-exposure explanation and warning reduced the extent to which people were misled by y-axis truncation, but did not eliminate the truncation effect. Despite explicit instruction and a warning about misleading design that occurred immediately preceding their judgments of 40 individual bar charts, participants rated the differences in the truncated bar charts as larger than differences in the non-truncated bar charts.

In the first four studies, participants' "graph literacy", that is, their ability to interpret bar charts and their experiences with visual representations of quantitative information, also was assessed. Graph literacy was unrelated to a person's susceptibility to the truncation effect. In the fifth study, Ph.D. students both in quantitative fields and in humanistic fields were found to be susceptible to the truncation effect, with those in quantitative fields exhibiting a modestly smaller truncation effect than those in humanistic fields when no warning was given.

The researchers conclude that bar charts with a truncated y-axis lead to exaggerated understandings of illustrated differences, and education and warnings are ineffective at reducing the truncation effect. However, when truncated and non-truncated bar charts of the same data are shown side by side, the misleading bar chart can be identified by the vast majority of people.

Source Yang, B.W., Restrepo, C.V., Stanley, M.L. & Marsh, E.J. (2021). Truncating bar graphs persistently misleads viewers. Journal of Applied Research in Memory and Cognition, 10(2), pp.298-311.