Misrepresenting COVID-19: Lying With Charts During the Second Golden Age of Data Design

Comparing Mortality Rates With Pie Charts

The visualization prepared for Time magazine by Elijah Wolfson (2020) uses four pie charts to compare mortality rates for COVID-19, the seasonal flu, SARS, and MERS. The visualization uses “small multiples” (Tufte, 1990/2011, p. 29) as four similar pie charts combine to create an argument downplaying COVID-19’s severity. Fatal cases are depicted in red whereas nonfatal cases are in teal, echoing the color selections from Nightingale’s rose diagrams depicting deaths during the Crimean War (Brasseur, 2005) and leveraging Western readers’ associations of red with danger or death. The data label is centered below each pie chart, listing the disease name in a bold black font and “Fatal cases: x percent” in red. Wolfson’s visualization places COVID-19 on the far left, followed by the seasonal flu chart. Only 0.1% of flu cases were fatal over the 2018–2019 season, so the pie chart does not contain any red—a thin white line represents its minimal mortality rate. Next, SARS shows a fatality rate of 10%. Finally, MERS is on the right, showing a fatality rate of 34%. Each of these charts is factually accurate and uses effective design strategies; however, Wolfson’s visualization misleads viewers by misusing visual proximity to minimize COVID-19’s death rate as compared with that of SARS and MERS.

Even if the data within a chart are correct, and the designer uses accurate design affordances, a chart can still mislead viewers in two main ways: by displaying deceptive patterns and by providing inadequate data. The Wolfson (2020) visualization is an example of displaying deceptive patterns that create a visual argument that, while technically accurate, downplays the seriousness of COVID-19. Although each pie chart is factually correct, the pie charts positioning within the visualization minimize COVID-19 deaths by leading viewers to equate COVID-19’s fatality rates with the seasonal flu.

The Wolfson (2020) visualization also appears to make a complete argument from insufficient data. Pie charts measure static percentages, not rates of change over time. This visualization does not account for COVID-19’s high contagion rates, 2–13 day incubation period, or disproportionate impact on Black and Native American people in the United States (Laurencin & McClinton, 2020; Shaw, 2020; Yancy, 2020). Reducing COVID-19 fatalities to arguably emotionless pie charts ignores the virus’s long-term impact on survivors; preliminary research shows that adult survivors with mild symptoms live with increased risk of stroke, blood clots, and reduced heart and lung functions (Healy, 2020). Not including enough data can cause viewers to misinterpret visualizations because “a chart only shows what it shows and nothing else” (Cairo, 2019, p. 154). To some extent, all charts simplify real arguments and datasets; some simplification is necessary to make an argument but should not deceive the viewers.

Graphing New COVID-19 Cases

Second, the “New Cases Per Day” (2020) line graph, shown on FOX 31 in Denver, Colorado, on April 4th, 2020, adjusts the logarithm and spacing of its y-axis to minimize the daily number of new cases from March 18 to April 1, 2020. The y-axis starts at 30, serving to visually minimize the 33 cases on March 18th. The first three markers on the line graph follow a predictable scale (30, 60, 90), aligning with the 33, 61, and 86 new cases occurring March 18–20. Next, the y-axis increases from 90 to 100 to 130, appearing to flatten the surge in new cases happening March 21 and March 22 (112 and 116). To make the surge in cases from March 23 to March 24 look smaller, the y-axis jumps to 160 and 190. Finally, to minimize the increase in cases from 174 on March 25 to 344 on March 26, the y-axis skips by 50, except for the 240 to 250 data labels, to accentuate the drop in new cases on March 29th. Visually, this strategy leads viewers to believe that all numbers indicated on the y-axis are measured using the same scale because the y-axis lines are evenly spaced.

Explaining Net Hospital Admissions in New York City

A bar chart used during New York governor Andrew Cuomo’s media briefing on April 9, 2020, misleads because it shows falling net hospitalizations in New York City without contextualizing the record number of hospital admissions still taking place ( Coronavirus Briefing, 2020). Cuomo opened the briefing with a fan chart about predicted possible hospitalizations; due to government action, he explained, New York City did not hit the worst-case scenario during March and April 2020. Cuomo discussed three bar charts representing decreases in hospitalizations (lowest since March 19), ICU admissions, and intubations. Cuomo praises New Yorkers for following the social distancing guidelines and reveals a plan for increased testing and care within African-American and Latino neighborhoods.

Although Cuomo ties the numbers and their displays to specific actions, such as the shelter-in-place order, more detail in the chart’s annotation layer (Cairo, 2019)—a visualization’s title, caption, and data labels—is necessary to contextualize the chart for nonexpert viewers. If viewers see the chart unaccompanied by Cuomo’s briefing, they could reasonably assume that only 200 people were admitted to ICUs in New York City on April 8th.

Along with contextualizing the data, additions to the annotation layer should cite the chart’s data sources. Cuomo’s chart does not indicate what counts as hospital admissions, such as surge hospitals created as emergency care centers, or include the number of people who might have been treated through urgent care or emergency rooms and then sent home. Cuomo’s plan to expand testing and care in minority neighborhoods also suggests that these communities might not have been receiving necessary health care, including hospitalizations. While the net decrease of total hospitalizations shown on April 8th was encouraging data to support arguments for flattening the curve, the visual does not represent the total hospitalizations, only the smaller net decrease in new admissions. Although this visual is accurate, the annotation layer needs to label the chart’s argument and data source more clearly for a more accurate chart. Adding a callout box or caption to remind viewers that hospital admissions are still higher than average would both increase the chart’s accuracy and improve arguments for extended social distancing.