Research Brief

Graphic Presentation of COVID-19 Data Can Skew Perceptions of Risk

Showing cumulative cases — not day-to-day trends — could nudge people to avoid reckless behavior

Visualizations of COVID-19 data are omnipresent in the media since March 2020 such as the excellent Johns Hopkins dashboard, now one of the most well-known resources for tracking the pandemic across the world. As vaccine hesitancy becomes a greater societal risk given the intense transmissibility of the delta variant, understanding the potential for graphics to encourage or discourage behavior — getting vaccinated, sending kids back to school, wearing a mask, attending large indoor gatherings — is valuable in shaping public health communication policy.

A  paper forthcoming in Journal of Experimental Psychology: Applied by University of Colorado’s Nicholas Reinholtz, University of Toronto’s Sam J. Maglio, and UCLA Anderson’s Stephen A. Spiller investigates whether the format of a chart’s presentation may have different influences on a viewer’s judgment of existing risk of COVID-19 infection and how that may impact subsequent behavior.

The authors previously collaborated on research looking more generally at how data can be presented in ways that lead to different evaluations and forecasts.

A surprise finding in the researchers’ more recent paper suggests that one particular presentation format of data visualizations leaves participants likely to engage in riskier behavior than another format, regardless of whether the format showed the number of new COVID-19 infection cases rising or falling. (More on that below.)

The Dark Art of Data Manipulation

The power of graphic representation to make dense data more accessible has an embedded risk: Data in the wrong hands can be manipulative. This Georgia Department of Public Health graphic, with its nonsensical ordering of days on the x-axis and its daily reordering of the counties to create a downward trend in the data, was particularly troubling.

Another example is this graphic from Fox News, which uses different lengths between days on the x-axis and an inconsistent interval on the y-axis for the count of COVID-19 cases to reduce the slope of the line.

While it’s easy to see how these manipulated charts could sway a viewer’s judgment on the current risk of infection presented by COVID-19, Reinholtz, Maglio and Spiller examine how the format of a data visualization may impact a viewer’s judgment even when the information is appropriately presented and there is no intention of manipulating the viewer’s opinion.

Going with the Flow?

In an experiment, they found that participants’ interpretation of data — and their anticipated behavior based on that interpretation — was swayed by whether they saw a graphic showing the cumulative count of COVID-19 cases or one showing the trend line for daily new cases of infections.

The graphics below illustrate the difference in the two formats representing the same data from the CDC for the period between Jan. 22, 2020, and July 11, 2021. The left graphic shows daily new cases, a format known as “flow,” while the right graphic displays the cumulative number of cases, a format known as “stock.”

The cumulative graph on the right rises steeply when the left side showing daily cases rises. The cumulative graph then flattens as the daily cases return to lower levels. While it‘s possible to convert between the two formats, past research shows that even highly educated people have a difficult time doing so. In 2009, MIT graduate students were given the task to convert data between situations representing the two chart formats; fewer than one-third of the students made the conversions correctly.

For this reason, viewers usually interpret data based on the chart format presented to them. Reinholtz, Maglio and Spiller investigate whether viewers’ interpretation of the data is inconsistent when the trend of the graphs in the two formats moves in different directions. Since the cumulative number of COVID-19 cases is always increasing, divergence between the formats only happens when the daily number of new cases is decreasing. This diverging state occurs repeatedly over time as the number of new cases tends to rise and fall.

To examine this idea, the researchers conducted an experiment showing 20 days of COVID-19 data to 596 participants recruited online. The participants were first split into two groups; those who would be shown the data as a cumulative number of cases and those who would be shown the data as daily new cases of infections.

Both groups were further split with some shown one state’s data from March 16, 2020, to April 4, 2020, when the number of new cases was decreasing. And the rest were shown data from May 7, 2020, to May 26, 2020, when the number of new cases was increasing.

On the left side are the graphs of the daily changes in new COVID-19 infections, while the right side graphs the cumulative cases. The two graphs at the top represent the period when the number of new cases was increasing and the bottom two represent the period when the number of cases was decreasing.

After viewing their assigned graph, participants were asked to fill out a survey in which they indicated their opinion on the current level of risk and then were asked a series of questions to indicate how likely they would be to engage in certain behaviors given the level of risk. These questions included outdoor and indoor dining, using public transportation, having visitors in one’s home, sending their children back to school and various questions around shopping and using other services.

The researchers found that when daily new cases of COVID-19 were decreasing, participants perceived greater risk when viewing the data as cumulative cases of infections (showing an increasing trend) than when viewing the data as daily cases of new infections (showing a decreasing trend).

When daily cases were increasing there wasn’t much of a difference in judgments of risk as both charts showed an increasing trend. This was not a great surprise to the researchers given their past research in the area.

However, they were surprised to find that the participants’ behavioral intentions did not necessarily follow their risk judgments. Participants indicated a stronger intention to engage in risky behavior when they viewed a chart with the format as daily changes in new cases. They indicated a stronger likelihood to engage in risky behavior regardless of whether the trend of new cases was increasing or decreasing.

These findings suggest that what seems to be a minor choice in how to present data may lead to different conclusions about risk among viewers and even impact their intended behaviors. Based on these results, authorities may want to consider presenting data as cumulative cases, rather than daily changes, as a type of behavioral “nudge” to increase appropriate risk responses during a pandemic.

Featured Faculty

  • Stephen Spiller

    Associate Professor of Marketing and Behavioral Decision Making

About the Research

Reinholtz, N., Maglio, S. J., & Spiller, S. A. (in press). Stocks, Flows, and Risk Response to Pandemic Data. Journal of Experimental Psychology: Applied.

Related Articles

Two staffers wearing face masks hand dishes to women in the senior living center Research Brief / COVID-19

More Staffers: More COVID-19 Exposures and More Nursing Home Deaths

Reliance on part-timers raises headcounts and dangers to patients

Research Brief / Technology

Can Algorithms Help Reduce Hospital Readmissions?

Measuring the utility of increased care and testing, inputs that aren’t always immediately available

Empty Times Square in New York in blue (left side); Research Brief / COVID-19

How Clinton and Trump Voters Behave in — and Spread — a Pandemic

Estimates are based on smartphone data and precinct-level 2016 vote results

Illustration of a guillotine Research Brief / Cultural History

Beyond Angry Mobs: Intellectuals in the French Revolution

History’s Encyclopédie subscribers are matched to grievances against the monarchy