Abstract
What I mean to say is that data comes from people. It’s a mark that someone has left behind, or a mark that someone has put their hands on to collect. And in our excitement to harden that data into visualizations we often forget that behind those numbers are human beings.
Sarah Slobin (2014)
Data visualization researchers and designers have explored a range of approaches to ensure that non-expert audiences understand and derive value from their work. Using anthropomorphized data graphics—or anthropographics—is one strategy that can help “create an immediate visual connection between abstract data and actual people” (Boy et al., 2017). Anthropographics have been defined as “visualizations that represent data about people in a way that is intended to promote prosocial feelings (e.g. compassion or empathy) or prosocial behavior (e.g. donating or helping)” (Morais et al., 2020b). However, during the SARS-CoV-2 pandemic, anthropographics were used in data visualizations that had an expanded range of rhetorical goals beyond promoting prosocial feelings and behavior—for instance, informing people about the pandemic, persuading them to adopt certain behaviors, or memorializing those killed by the virus. In particular, anthropographics were used in visualized simulations to model possible futures for audiences, showing the spread and impact of the virus in various scenarios. These simulations used anthropomorphizing strategies in text as well as in graphics, along with interactive options that enabled audiences to explore personal connections with the data. As demonstrated through a close reading of several of these COVID-19 simulations, anthropographics can be viewed holistically as a design strategy that incorporates text and interactivity as well as graphical marks in representing data; anthropographic design strategies can be used by data visualization designers, data journalists, and others who communicate with non-expert audiences for a range of rhetorical goals. Findings from this analysis suggest several additions to the design space for anthropographics described by Morais et al. (2020b).
COVID-19 communications
The COVID-19 pandemic made data visualization an important mode of communication for a general public relying on data to help guide their everyday decisions in the midst of profound uncertainty and fear caused by a deadly global virus. Ben Shneiderman (2020), a leading visualization researcher, wrote that “The complexity and importance of COVID-19 has put data visualization center stage in worldwide discussions,” as specialists in health, science, journalism, government, and the corporate world collected, analyzed, and visualized data in order to understand and effectively respond to the pandemic. While data visualizations were being widely used to facilitate expert analysis in all areas impacted by the pandemic, non-experts were also a target audience for data visualizations that engaged, informed, educated, persuaded, modeled future scenarios, and performed other rhetorical functions. This study focuses on how COVID-19 simulations used anthropomorphized data graphics—or anthropographics—to achieve their rhetorical goals and to reveal what Slobin (2014) refers to above as the human marks on data.
Projects using anthropographics stand in contrast to the more familiar types of charts and graphs that visualized aspects of COVID-19, in which people who were infected, hospitalized, recovered, or dead were represented mostly as the empty space under a line or as an indistinguishable dot of ink or pixel of color in a bar chart or on a map. Anthropographic projects, on the contrary, sought to humanize data by drawing attention in a variety of ways to the human-level impacts of the virus. This visibility of humans in data visualizations was especially important in a global pandemic characterized by invisibility, absence, and loss: the invisibility of the submicroscopic SARS-CoV-2 virus as well as the isolation, social disconnection, death of loved ones, and diffuse but all-encompassing harm caused by the pandemic. Moreover, anthropographic strategies, viewed rhetorically, foreground the subjective and embodied nature of data and its representations and are therefore well suited to evoking the particular, situated experiences of people during the pandemic. More broadly, Kostelnick’s (2019) study of human forms in practical communication informs an understanding of the rhetorical goals of anthropographics: “In visualizing practical information, human forms make that information more understandable, engaging, and meaningful to their audiences” (3).
Anthropographic studies
As noted above, researchers have studied anthropographics primarily as a design strategy intended to increase the empathy and prosocial attitudes and behaviors of an audience that interacts with a data visualization. The recent and still relatively small body of research on anthropographics is part of a broader trend of examining different design approaches and measures of effectiveness for data visualizations that address everyday topics and target a general audience as opposed to those that are utilitarian and designed for data experts and analysts (Hullman and Diakopoulos, 2011; Lupi, 2017; Pousman et al., 2007; Sorapure, 2019; Viegas and Wattenberg, 2006). Some researchers have focused specifically on empathy and other emotional appeals in data visualization (Kennedy and Hill, 2018; Kostelnick, 2016; Zer-Aviv, 2017), while others have studied how data visualizations might encourage civic engagement and activism (Claes and Vande Moere, 2013; Dörk et al., 2013; D’Ignazio and Klein, 2020).
Boy et al. (2017) coined the term “anthropographics” in the context of what they described as the “growing assumption that visually connecting abstract data with iconic representations of people can elicit empathy for, and encourage prosocial behavior toward those people” (5462). Researchers have provided descriptions of the design space of anthropographics that delineate different options for representing humans graphically. Boy et al. (2017) mapped unit visualizations (i.e. where each row of a data table is indicated with a single visual mark) along axes from abstract to realistic and from neutral to expressive; they also categorized unit labels as generic, iconic, or unique, and unit groupings as grid-based or organic. Morais et al. (2020b) built on this work to develop a more detailed design space for anthropographics, consisting of seven dimensions in two broad categories: what is shown and how it is shown. In the first category, addressing the kind of information presented in the visualization, the four dimensions are granularity, specificity, coverage, and authenticity. In the second category, describing how the data is represented, the three dimensions are realism, physicality, and situatedness. Taken together, these seven dimensions present a design space with many options for data visualization designers to represent real people, from abstract and generic (e.g. a dot) to unique and personalized (e.g. a photograph).
Several empirical studies of anthropographics have been conducted thus far and have provided weak or mixed results regarding the “anthropographic assumption” that iconic representations of people can evoke empathy and prosocial behavior. In seven experiments using data visualizations on the topic of human rights (Boy et al., 2017), participants read data-driven stories, some of which included anthropographic elements, about the plight of Syrian child refugees, and then the participants’ empathic responses were measured along with their inclination to donate to the cause they had read about. Contrary to expectations, the results showed that, when compared to standard charts and graphs, the use of anthropographics did not yield increased empathy or prosocial behavior. Boy et al. (2017) concluded that “anthropographics are neither truly beneficial, nor detrimental” (5462) in achieving these goals; they pointed to further avenues for research, such as using graphics with more realistic or detailed depictions of individuals or combining graphics with animations or other multimedia effects to increase affective responses.
In a subsequent study looking at the extent to which physical and situated data visualizations about sexual harassment evoke compassion and prosocial behavior, Morais et al. (2020a) similarly found that anthropographics show only weak or inconclusive effects. Liem et al. (2020), studying visual data storytelling about immigration, did not find any greater effect in the visual data narrative that used anthropographics and was designed to evoke empathy. Finally, Morais et al. (2021) tested what they described as “information-rich” anthropographic data visualizations which showed data about individuals that is granular, realistic, specific, and authentic. In two experiments with larger sample sizes than in previous studies, information-rich visualizations about migrant deaths were compared to information-poor treatments that had aggregated data presented in bar chart format. Morais et al.’s (2021) findings were that the anthropographics had, at best, a small effect on prosocial behavior, and they note that “the overall inconclusive results from the range of experiments [about anthropographics] conducted by researchers so far call for some skepticism.”
The close readings offered below of COVID-19 data visualizations that use anthropographics highlight three substantial differences between these projects and the research conducted to date. First, in the empirical studies, other elements of the data visualization were held constant while the graphics were changed in treatments shown to participants. For instance, in Boy et al. (2017), participants were shown a slideshow in which only the slides with the graphics were different in each treatment; the other slides presenting text and a map were held constant. However, in the examples discussed below, graphics are integrated inextricably with other elements in the project; specifically, the text included in and around the graphics is essential in anthropomorphizing the marks designating humans, engaging the audience in exploring the data, and supporting the connections between the data being presented and the humans behind that data. Especially with more abstract graphical marks, such as dots or lines, contextualizing features in the data visualization project as a whole anthropomorphize these non-human forms and are essential to their being understood as representing humans. An analysis of these examples therefore suggests that an understanding of anthropographics should be holistic, including the entire context in which marks representing humans are presented. In other words, anthropographics should be viewed broadly as a design strategy that incorporates textual and other elements beyond the graphical marks on their own. Prior research does include a consideration of textual elements in anthropographic data visualizations; for instance, Boy et al. (2017) discussed unit labelling of human shaped pictograms and Morais et al. (2020b) noted the information conveyed by annotations in some of the visualizations they discussed. However, the role of text in the anthropographic visualization was underexplored, focused only on labels and annotations rather than on the full range of ways in which text can be integrated in anthropographics.
A second key feature in the examples discussed below is the interactivity of COVID-19 data visualizations. The design space described by Boy et al. (2017) and Morais et al. (2020b) is based on static visualizations, even though approximately 40% of the visualizations in the corpus analyzed by Morais et al. (2020b) included either animation or interactivity. They noted that these dynamic visualizations offer users different ways to explore the data, for instance by enabling them to focus on subsets of data, examine different attributes of data items, and view the data through different representations (e.g. switching between map and grid view). While in some sense, as Morais et al. (2020b) note, “a dynamic visualization can be thought of as a (potentially very large) set of static views,” interactivity is more than the sum of static views and on its own has the potential to support inquiry, lead to insights, and increase users’ engagement (Dimara and Perin, 2019). As they explore the data and customize their view of it, users access interactive options to pursue both lower-level tasks of understanding the information space and higher-level tasks of analysis and discovery (Heer and Shneiderman, 2012; Pike et al., 2009; Yi et al., 2007). Anthropographics is a design strategy that incorporates interactivity as well as text, enabling the exploration of personal connections to data and facilitating the goal of increasing engagement with non-expert audiences.
A third important feature of the anthropographic COVID-19 data visualizations is that they have a range of rhetorical goals beyond those studied in previous empirical research. In the context of the pandemic, a key goal was to explain important concepts such as exponential growth and social distancing. Other visualizations, in genres such as dashboards, explainers, and data-driven narratives, were designed to inform the general public in as clear a manner as possible about the current spread and impact of the virus. In Marco Hernandez and Cate Cadell’s “How coronavirus hitched a ride through China” (https://graphics.reuters.com/HEALTH-CORONAVIRUS/SPREAD/xegpbwexvqz/index.html), for example, anthropographics are used to tell stories about the early spread of the virus, tracing its path out of Wuhan and into the rest of China and beyond. Similarly in Reuters’ “The Korean Clusters” (https://graphics.reuters.com/CHINA-HEALTH-SOUTHKOREA-CLUSTERS/0100B5G33SB/index.html), “everyperson” icons and accompanying text narrate the interactions of a super-spreader in South Korea early in the pandemic. Anthropographics were also used in data visualizations designed to memorialize victims of the pandemic, as in the
Simulation
Simulation as a data visualization genre, with or without anthropographic elements, seems well suited to COVID-19 pandemic communication for several reasons. First, it provides an accessible simplification of a very complex reality. A simulation reduces the complexity of the source system it is modeling by eliminating certain features and homing in on others. For instance, a number of simulations of COVID-19 focused on social interactions as a key factor in the spread of the virus but omitted related factors such as age, gender, race, or cultural context. A second feature of the simulation genre, especially important in the context of COVID-19, is that it is oriented toward the future and can provide valuable speculative information. Frasca (2003) notes that “simulation is the form of the future. It does not deal with what happened or is happening but with what may happen” (233). Living through an unprecedented pandemic, people were naturally anxious about the future; while a simulation cannot directly or decisively answer future-oriented questions, it can provide a clearer sense of possibilities that can then be weighed in the context of other information. Finally, simulations offer opportunities to learn by doing that can significantly engage users. Like playing a game, users interact with simulations in order to understand the system—learning its parameters, figuring out what is and isn’t possible, focusing in on the most interesting or relevant information—in order to accomplish whatever goal has motivated the interaction in the first place (to win, to learn, to be entertained, etc.). During the pandemic, simulations were prominent among the several genres of data visualization used to communicate information to the general public.
Previous studies of anthropographics exclude simulations or hypothetical scenarios in order to focus on measuring the extent to which a data visualization evokes an empathetic response and encourages prosocial behavior toward real people. Indeed, the design space of anthropographics described by Morais et al. (2020b) assumes that the visualized data represents real people; the design dimensions being described are those “that could plausibly promote compassion” specifically toward those people. But the possibility of alternative rhetorical goals for the data visualization means that simulations depicting hypothetical people and scenarios can communicate valuable information to a general audience. Anthropographics can be especially helpful for readers who are envisioning and exploring uncertain futures by explicitly representing humans in speculative scenarios. As Lindgren (2021) observes about simulations, “The aim is not ‘truth’ but usefulness.” In this case, as we see in examples in the next three sections, simulations are made more useful to audiences by being connected more explicitly to them via anthropographic presentations.
Harry Stevens’s “simulitis”
The simulations created by Harry Stevens for his 14 March 2020
Stevens uses geometric shapes as anthropographics, with circles/balls of different colors representing healthy, sick, and recovered individuals (see Figure 1). All of the simulations start with “towns” of 200 balls, one of which is colored red to represent an infected person. A running tally of the numbers of recovered, healthy, and sick designated balls is presented as the simulation unfolds. Throughout the text, Stevens refers to the balls as people; for instance, introducing the fourth simulation, he writes, “instead of allowing a quarter of the population to move, we will see what happens when we let just one of every eight people move.” The balls are set in motion within their towns as the reader scrolls onto that part of the page or opts to run a new simulation. Unlike an animation that readers can replay to become familiar with where the balls move and with the outcome, each simulation presents slightly different results, further enhancing the realism of the scenarios which end with different numbers of infected and dead depending on the “behavior” of the balls; indeed, the point of the simulations is that different behaviors yield different outcomes. Together with the personification in Stevens’s text, the unpredictable motion of the balls gives them a human quality, as they move about, bumping into other moving balls and spreading or avoiding infection. Again personifying the balls, Stevens makes the analogy explicit: “like a ball bouncing across the screen, a single person’s behavior can cause ripple effects that touch faraway people.”
Screenshot from Harry Stevens’s “Why outbreaks like coronavirus spread exponentially and how to ‘flatten the curve’.”
Stevens’s simulations are vast simplifications of the spread of disease; there is no intention to represent the complexities of virus behavior or of real-life interactions. For example, in simulitis, when a sick person touches a healthy person, represented by two balls bumping into each other, the healthy person becomes sick; in real life, the spread of a virus from one person to another is obviously more complicated. But Stevens’s goal is not to show exactly how a virus spreads but rather to explain exponential growth and to demonstrate the effectiveness of social distancing in slowing the spread of infection. Readers are not prompted to respond empathetically to the balls as they move about and become infected in different scenarios of containment or social distancing. However, giving the balls a minimal human characteristic of unpredictable motion and giving them human attributes in the text—individualized treatments different from the use of indistinguishable dots or pixels of color in non-anthropographic charts and graphs—are strategies for increasing readers’ engagement with the visualization and thereby increasing the possibility that they will learn. Reducing the complexity of human movement to the bouncing of balls within a container facilitates insight and intuition about the collective consequences of individual actions during a pandemic. In an interview about the project (Cotgreave, 2020), Stevens explained that “It’s hard to develop an intuitive understanding of dynamic systems like epidemics. Simulations help readers build up their intuition about how diseases work in a way that words and even static charts cannot.” Even when geometric shapes represent hypothetical rather than actual people, as in Stevens’s project, they can be anthropomorphized in a variety of ways to draw attention to the human elements in the data and to enhance the impact and the insights that readers can draw from the visualization.
Kevin Simler’s outbreak
While readers can reset and replay the simulations in Stevens’s article, Kevin Simler’s “Outbreak” (https://meltingasphalt.com/interactive/outbreak/) includes additional interactive options and other anthropographic features intended to enhance engagement and facilitate learning, with opportunities for readers to gain insight into aspects of the pandemic that are most relevant to them and to possibly modify their behaviors as a result. Simler, a writer and software engineer, created “Outbreak” as a playable simulation that demonstrates in general terms how epidemics unfold. Squares in a grid represent people who are susceptible, infected, recovered, or dead. In describing these squares in grids, Simler gives them human attributes: it’s a “set of people who live in neat rows and columns,” with the “poor soul” at the center of the grid who is infected. When all of the squares turn red by infection in the first simulation, Simler writes “Oh no. It looks like everyone sneezed on their neighbors—north, east, south, west—and the whole world got sick.” The text here uses terms and descriptions likely to be more familiar to readers in their everyday contexts, referencing a “poor soul” and a “neighbor” rather than a more distanced and objectified “infected individual.” By anthropomorphizing the squares and grids in his descriptions and by describing infection in accessible terms, like Stevens does, Simler places the focus in this hypothetical scenario on the real human behavior being modeled. Simler also directly addresses readers to increase their engagement with the scenarios. For instance, after discussing incubation periods and noting that asymptomatic carriers are designated by pink squares in his simulation, he writes, “Even as you read this, you may be such a person”; highlighting the word “you” in pink extends the scenario out to the reader.
In addition to these textual features, increasingly complex opportunities for interactivity augment the readers’ engagement with the graphic elements of the simulation. In the first part of the project, readers learn about different factors involved in an outbreak, as Simler introduces the concepts of recovery time, incubation period, and transmission rate. He invites readers to manipulate different parameters to see how the simulation would play out differently with, for instance, higher or lower transmission rates, numbers of encounters per day, travel radius, or hospital capacity. In manipulating these variables, readers are assigning the squares human characteristics, imagining that they interact with other squares, stay close to home or travel further distances, and have local hospitals with a certain number of beds. Experimenting with these anthropomorphized squares, readers may be inclined to simulate their own behaviors and contexts as they set the parameters; they can put in motion changes they are contemplating and see how the simulation plays out these real choices (see Figure 2). In the text, Simler explains that some variables are a function of both the inherent properties of the virus itself (e.g. how naturally infectious it is) and the physical and social environment in which it operates. For instance, the transmission rate is affected by the extent to which people wear masks, wash their hands, and socially isolate. Interaction makes the simulation more relevant because in setting these parameters users are describing familiar human behavior and viewing the projected consequences of that behavior.
Screenshot from Kevin Simler’s “Outbreak.”
Later, Simler challenges readers to find a transmission rate or a travel radius that keeps the virus from spreading to the entire population. At the end of the piece, he gives readers a “final test” to see if they can “flatten the curve” given a fixed fatality rate and hospital capacity. In other words, readers are tasked with modifying variables that are actually within their reach in their everyday lives: encounters per day, travel radius, and transmission rate. They can tweak the parameters, perhaps mirroring their own current or planned behaviors, to see how the simulated epidemic would unfold.
Although Simler states from the outset that he is not modeling COVID-19 and that he is not an epidemiologist, he nevertheless draws connections to COVID-19 at various points in the presentation—for instance, noting that it has a long incubation period—and gives advice and warnings about behavior in the real world as he explains the simulations. The article ends with Simler exhorting readers to take decisive action: “COVID-19 is coming for us, and it won't be stopped by half-measures.” He also notes the shortcomings of just playing with a simulation and acknowledges the complex and unrepresented aspects of human agency: “However this worked out for you in simulation, reality is going to be
R2D3’s “making sense of COVID-19 through simulations”
R2D3’s “Making sense of COVID-19 through simulations” (http://www.r2d3.us/covid-19/) is a set of simulations that uses references to human activities along with annotations and other textual features as a strategy to engage, inform, and influence readers. The authors, Stephanie Yee and Tony Chu, create interactive visualizations; the goal of this project, they write, is to help readers “gain intuition” about what it means to reduce the spread of the virus. Continuing the strategy of direct address that we saw in Simler’s work, Yee and Chu set up an initial scenario for readers to envision: “Let’s say you meet up with two friends in the park. If one is infected but doesn’t know it, that weekend outing can result in the virus spreading.” Immediately following (see Figure 3), a diagram displays hypothetical results for that scenario. As in the previous two examples, the diagram uses color to designate susceptible and infected people, but rather than circles or squares it uses differently styled human icons—described by Harris (2015) as “wee people”—to represent young, middle-aged, and older people. The graphics used here are depersonalized forms—that is, generic figural icons that have no distinguishing features or individual characteristics. As Kostelnick (2019) notes, this “everyperson” approach avoids “the idiosyncracies and emotional distractions that might subvert the picture’s intended purpose” and “[distributes] identity across a large audience, allowing readers to see themselves (or people they [know]) in the picture” (43).
Screenshot from R2D3’s “Making sense of COVID-19 through simulations.”
Arrows and proximity show the connections between people, and textual annotations show generic locations: local park, ER, grocery store, nursing home. The diagram also assigns specific hypothetical characteristics to some of these icons to make them more individualized and relatable: “32 y.o. ER Nurse Healthy”; “68 y.o. Retiree Fighting Cancer” Immunocompromised; “89 y.o. Grandma Hip Transplant Immunocompromised.” Along with the anthropographic icons, these textual strategies work to draw readers in; for instance, generic locations enable readers to envision their own version of these places, while annotations show the kind of challenges that people with specific characteristics and susceptibilities to COVID-19 might encounter there.
The rest of the project is composed of several simulations that continue to use wee people icons, and as in the previous two examples the simulations run when readers scroll into that part of the webpage. The first simulation illustrates uncontrolled spread with exponential growth, and the second shows the impact of social distancing on hospital bed availability. The third simulation invites readers to explore different variations of social distancing by checking or unchecking five options for “connection types”: Social Contact, Non-Essential Work, Essential Work, Family, and Immediate Family. As with the sliding scale parameters in Simler’s simulation, these interactive options give readers an opportunity to assign different human behaviors to the anthropographic icons and then press play to see the consequences of those behaviors as projected by this simulation. With all of the connection types checked—that is, with group socializing, both essential and non-essential work, and contact with both near and extended family—the simulation projects many more infections and deaths and a longer time period of contagion than with only essential work and immediate family selected. The ability to reset, change the options, and replay the simulation means that readers can see the stark differences that result from different choices that they themselves might be making on a daily basis. As in the previous two examples, each time the simulation is run it produces slightly different results, making this more realistic than a predictable animation would be. A fourth simulation allows readers to set different time periods for enforcing social distancing: 40 days of social distancing results in around 250 infections and 20 deaths, whereas 80 days of social distancing drastically reduces those totals to around 40 infections and 1 or 2 deaths. By interacting with simulations that address their most pressing concerns—what type of social distancing should they do and how long should they do it—readers may be more engaged and interested. As in Simler’s visualization, the interactive options increase in complexity as the simulations call on readers to make decisions about more variables, creating an experience where readers might feel an increasing sense of responsibility and interest in learning.
Conclusion
As these examples demonstrate, anthropographic COVID-19 data visualizations use different types of graphics along with contextualizing and interactive strategies to humanize data. While previous anthropographic studies focused fairly narrowly on the graphics themselves and on a limited set of rhetorical goals, this analysis shows that COVID-19 data visualizations demonstrate additional ways in which anthropographic strategies can enhance communication with audiences. In these simulations, the use of anthropographics—anthropomorphized geometric shapes or human-shaped icons with accompanying anthropomorphizing text and interactivity—is a strategy to facilitate readers’ engagement and learning about the pandemic that may also motivate changes in their behavior. Readers are not being asked to empathize—after all, these are hypothetical rather than real people being depicted—but rather to learn about factors in the spread of disease and make connections to their own lives. Although each project explicitly references COVID-19, they all also clearly indicate their own limitations, explaining that these are simulations, not predictions of how this particular virus will actually behave in the real world. Anthropographic features of these visualizations draw attention to human behavior, specifically in order to reduce the spread of infection by projecting what might happen depending on how people act. Here, with the goal of persuading readers to adopt certain behaviors and abandon others, an anthropographic approach connects readers with possible future versions of themselves and their world.
In each project, the authors reveal their own perspectives and, largely through the tone and style of writing, they embrace a situated and subjective presentation that is atypical in data visualization. Textual features like direct address, informal tone, relatable details, and a strong presentation of advice at the end are clearly rhetorical choices aimed at increasing the impact and relevance of the project. The text in these projects serves to engage readers and to humanize both the people being represented by the data and the people presenting the data and its visualizations. The use of stories and anecdotes further conveys the marks of humans as characters, authors, and readers of data. These projects also use a range of interactive options to actively engage and create experiences for readers that highlight the human-level impact of the pandemic. Whether through scrollytelling, hovers or clicks that provide more information, or the selection of variables, readers are doing more than simply looking at the visualization; the interactivity provides opportunities for exploration that can enrich the insights they gain.
For data visualization developers and for researchers interested in exploring the effectiveness of anthropographics as a design strategy, these findings suggest that several elements could be added to the design space for anthropographics outlined by Morais et al. (2020b). As it is, this design space focuses almost entirely on the graphical marks in the visualization, with its two categories of dimensions—
Footnotes
Declaration of conflicting interests
Funding
ORCID iD
