Rachel Suppok, Research and Instruction Librarian, rachel.suppok@pitt.edu
Data visualizations, such as charts and graphs, are a way to explain and summarize complex phenomena and vast amounts of data. These visualizations are often circulated on social media sites, where users can share and comment on the posts as a way of collective sense-making. However, these valuable insights can be compromised by perpetuating misinformation. Misleading visualizations are primarily defined as those that interfere with the viewers' abilities to read off and compare the values from the visuals. The visual deceptions utilized in misinformation are rooted in the true value of the data used for the chart and the different values perceived by the viewer, referred to as the "Lie Factor."
There a number of ways for graphs and charts to be altered in order to present data in a misleading format. Axes are manipulated to change how the data is displayed. When the scaling of the Y-axis is manipulated, the differences between data points can be exaggerated or minimized. For example, a truncated Y-axis can make a small change appear significant, or stretching the Y-axis can make the difference between data points look smaller than they actually are. If the X-axis has been manipulated through the use of inconsistent intervals, the uneven spacing between the data points may result in the illusion of a trend or pattern that does not actually exist within the data. Additionally, graphs may manipulated so that the baseline does not begin at zero. If the graph begins at a non-zero value, the differences between the data points can be exaggerated, resulting in the differences appearing far more significant than they actually are.
Data can be manipulated by selection, something often referred to as "cherry-picking." A biased representation of the data can be displayed by selecting specific data points or time periods to be included in the graph rather than the entirety of the data or time. The omission of certain data can lead to trends being either amplified or suppressed in order to support a specific narrative. False trends can be established through other methods, such as data interpolation. Without sufficient evidence, the interpolation of data between points can lead to inaccurate conclusions or tends.
How data is displayed can also contribute to misleading conclusions. Choosing an inappropriate chart type to exhibit the data can distort it. For example, a 3D chart makes it more difficult for the viewer to interpret the differences between values accurately. The chart can also be created to intentionally confuse viewers by including overlapping data series. By including data series that are overlaid, the viewer has a more difficult time distinguishing between them, resulting in confusion or misrepresentation of the relationships between the variables. The data series could contain data that is entirely unrelated in order to create false associations or imply relationships that simply do not exist. Charts and graphs may be intentionally created to have misleading labels, like ambiguous or deceptive axis labels, causing viewers to misinterpret the data.
Lisnic et al. investigated data visualizations created to spread COVID-19 misinformation. Through this examination, they determined that the vast majority of the charts and graphs in their data set did not actually violate common visualization design guidelines. Visual design violations were helpful for exaggerating effects and supporting a specific narrative. However, visuals that still conform to the design guidelines were also successful in supporting effective misinformation arguments. Often, the visualizations from their data set were from reputable sources, such as government and media outlets. The intent behind the creation of the data graphics were not to be deceptive, but their presentation caused them to be misleading. Depending on how the visual is framed, the conversation context, and anticipated audience, data visualizations can be vulnerable to misinterpretation even when created with the best intentions.
Answers to Exhibit Examples:
1. Covid Rate Maps, Georgia (source: Georgia Department of Health)
Although these maps at first glance appear to show that the rate of COVID infections remained relatively stable between the two timepoints, take a closer look at the bins the data are sorted into on each map. The bins on the second map have different parameters than the first, causing the map to appear quite similar while rates may actually have increased between the two dates.
2. Gun Deaths in Florida (source: Florida Department of Law Enforcement)
At first glance, it may look like this visualization is implying that firearm murders in Florida decreased after the passage of a "stand your ground" law in 2005. However, the y-axis has been inverted, and the number of these events actually increased after the passage of this law. The use of a dark red color can also give the impression that the red is a background, further emphasizing the apparent "decrease."
3. Percentage of Union Workers in Electricity Generation Industries in the US and Canada (source: unknown)
This graph appears to show that the percentage of union workers are nearly the same in the US and Canada in most industries, as the respective bars are typically very close to the same height. However, the Canadian data and the US data have different y-axis scales, and the union membership rates for each industry are actually dramatically different between the US and Canada.