Feeling Numbers: the Rhetoric of Pathos in Visualization

Feeling Numbers

The Rhetoric of Pathos in Visualization

SARAH CAMPBELL

Feeling Numbers The Rhetoric of Pathos in Visualization

Thesis presented by Sarah Campbell

To The Department of Art + Design

In Partial Fulfillment of The Requirements for The Degree of Master of Fine Arts in Information Design + Visualization

Advisor: Dietmar Offenhuber Reader: Alberto Cairo Respondent: Pedro Cruz Chair: Nathan Felde

Northeastern University Boston, Massachusetts

May, 2018

Acknowledgements

To my mom and dad, for their never-ending support. To Alex, for being my rock through this journey. To my cohort, for the critiques, laughs, shared pain, and inspiration. To Phelps, for being such an entertaining turtle.

Abstract

Rhetoric is a powerful tool used to influence and persuade. Due to their inherent subjectivity, data visualizations are a form of communication that employ persuasion techniques. Therefore, the rhetoric of visualizations deserves deeper investigation. Drawing from Aristotle’s three modes of persuasion, this thesis explores the rhetoric of pathos, or appeals to emotion, within data visualizations. In this thesis, I develop a taxonomy of pathos techniques applied to visualizations and empirically measure the emotional effect of pathos techniques that relate to data proximity. This research improves the visualization community’s understanding on how certain design decisions can add meaning and relevance to data. Contents

11 17 37

Introduction Rhetoric and Techniques for Emotional Appeals Appealing to Emotion: in Visualization A Taxonomy

Rhetoric Engage

Emotional Appeals Humanize in Visualization Personalize 49 73 77/83

Evaluating the Conclusion References Emotional Effect of Proximity Techniques Appendix

The Value

Proximity

Emotion

Study Methodology

Analysis & Results

Discussion

Introduction

Data and Emotions. These two words are an unlikely pair. Data are typically described as rational, objective, cold, black and white. Emotions, on the other hand, are subjective, colorful, and expressive. It may seem like the two do not have a common meeting place, but they potentially do: in the visual representation of the data.

Today data are ubiquitous, so it is not surprising that visualizations of data are increasing in ubiquity as well. With that, visualizations have broadened their scope to act as a communication tool to convey powerful messages. Data journalism has popularized visualizing information for the general audience. Newspapers like The New York Times, The Guardian, and The Wall Street Journal employ their own graphics teams that visually convey information on issues we care about: politics, the environment, health care, education, and so on. Advocacy groups are also taking advantage of increased accessibility of data, harnessing the power of information and technology for activism by appealing to audiences rationally, ethically, or emotionally (Tactical Technology Collective 2014).

Appeals to logic, ethics, and emotion lie at the foundation of Aristotle’s Figure 1.1 system of rhetoric (2004). The existence of persuasion in visual Items: Is Fashion Modern? A representations of data is not consensually accepted in the visualization fashion exhibition visualized community. When persuasion is discussed, it is typically discussed by Accurat for MoMa through negatively connoted terminology such as bias (Hullman and (Bassan et al. 2017). 12 INTRODUCTION

Diakopoulos 2011) or deception (Pandey et al. 2015). The study of rhetoric in this way has tremendous value for developing a visual literacy to recognize forms of deception, but I argue that rhetoric has other facets to consider with respect to data visualization.

If we look at persuasion as an umbrella term for influence, whether it is to influence one’s beliefs, attitudes, or behaviors, the concept is not completely negative. However, efforts to influence by visualizing data have been traditionally looked down upon due to visualizations being considered as objective, neutral representations of the data. This viewpoint is fueled by the scientific, analytical lens that visualization has typically been looked at through. Additionally, influence in visualizations is typically focused on in terms of unethical design decisions that lead to deceptive representations of the data, where the message of the data does not match the message of the visualization. In an effort to minimize deceptive uses of data, the visualization community has emphasized data-centric designs, where clarity of the data is stressed. Therefore, any visual elements that do not aid in maximizing the data shown are considered unnecessary.

However, in several ways, this has led to false thinking that there is an objective way to visualize data, when indeed there is not (Dörk et al. 2013; Hullman and Diakopoulos 2011; Viegas and Wattenberg 2007). If this were true, designers given the same data would visualize it in the exact same way, making identical design decisions. In reality, one dataset can be visualized in many different ways. By merely going through the process of visualizing data, the designer is creating a persuasive message, since the act indicates that the designer believes the data can be influential in some way. In this light, it is far more ethical to acknowledge the subjective nature visualizations have instead of masking it under the ideals of neutrality.

Visualization practitioners first need to acknowledge the subjective, persuasive nature of visualizations and look at the practice as a form of communication. By better understanding the subjective and influential aspects of a visualization, we can better scrutinize design decisions made that deviate from the message of the data. We can go a step beyond that and embrace the subjective nature of visualizations. In doing so, we can craft visualizations that are more human-centric for our audience: people. A closer look at persuasive efforts in this way focuses on pathos: the appeal to emotions of the audience.

In Aristotle’s three established appeals, pathos is the mode that is the least agreed upon in the visualization community. It is understandable how an accord was found in the use of visualizations to speak to reason and ethics. But to admit, or even celebrate, the potential for visualizations to bring an emotional capacity to data? This is a concept the community as a whole has only recently begun to talk about. With the growth of using visualizations INTRODUCTION 13

as a form of communication about important issues relevant to humanity, designers are seeking ways to add the human element to the data and impact how the audience feels about the information the visualization conveys. Through recent literature, conference talks, blog posts, and podcasts, practitioners are engaging in a conversation about emotions in visualizations. Some even offer techniques that may appeal to the emotions of the audience. However, little research has evaluated the emotive impact these techniques have in visualizations. Do they evoke emotion at all? If so, how powerful are they?

This thesis builds upon existing discourse and research on the role of pathos in visualizations. Through collecting and categorizing recommended techniques, I propose a working taxonomy of techniques that are said to appeal to emotions. More importantly, I present an empirical evaluation of three specific techniques: proximity of interests, temporal proximity, and spatial proximity. These techniques rise from the concept that we care more about people and events that are near us in time, space, and importance (Campbell 1776). Through this evaluation, I begin to identify the emotive impact these techniques have when implemented in visualizations. The goal of this work is to further the research on how visualizing data can appeal to emotions. By improving the community’s understanding of emotional appeals, we improve our understanding on the communicative power of visualizations as a whole and, therefore, enhance our ability to visualize data to connect with the people who explore it. 14 INTRODUCTION INTRODUCTION 15

If I look at the mass I will never act. If I look at the one, I will.

Mother Teresa

Rhetoric and Emotional Appeals in Visualization

This chapter serves to highlight related work around rhetoric and emotional appeals in visualization. While rhetoric has established theories throughout history, the visualization community as a whole has yet to cozy up to the idea of visualization as rhetorical communication. However, visualization literature points to the subjective and rhetorical nature of visualizations. When it comes to appealing to emotions, psychology research validates the struggle to affect people with numbers. Data visualization practitioners acknowledge this difficulty and, at the same time, many see value in arousing the audience’s emotions through visualizing data. A few practitioners identify techniques that are suspected to appeal to emotions in visualization, but very little research evaluates the emotional impact of these techniques.

RHETORIC

Aristotle is credited with developing the foundational system of rhetoric and Figure 2.1 defines it as the study of means of persuasion (2004). Within this system are Off the Staff. a series three modes of persuasion. The first is logos, which appeals to reason. Since of scores visualized by rhetorical efforts were primarily delivered in the form of speeches in his time, Nicholas Rougeux for Aristotle refers to logos as the substance of the speech itself “when we have the OpenScore initiative proved a truth or an apparent truth” through evidence. Ethos, the second (Rougeux 2016). mode of persuasion, depends on the personal character of the speaker. Aristotle notes that this is likely to be the most effective mode of persuasion. The third mode and focal concept in this thesis is pathos, which depends on the ability to stir the emotions of the audience. 18 RHETORIC AND EMOTIONAL APPEALS IN VISUALIZATION

These three modes of persuasion relate to one another and together increase the strength of an argument. They are also commonly referred to together as the rhetorical triangle, shown in Figure 2.1. When the modes are properly balanced within the argument, Booth defines this as the rhetorical stance (1963). If the modes are unbalanced, the stance can be corrupted. One example of this is the pedant’s stance, where the relationship between the audience and speaker is ignored, minimizing pathos for the sake of maximizing logos. At the same time, undervaluing logos for the sake of maximizing the impact on the audience is argued as a more serious threat to society. The argument is a combination of these three modes, and it is the most successful when each mode is valued (Booth 1963).

ETHOS character of the speaker

LOGOS PATHOS logic of the message emotions of the audience

Rhetoric in Design Figure 2.2 Buchanan (1985) tethers the greater field of design to traditional rhetoric The rhetorical triangle, theory in two ways: by asserting design as a form of communication and comprised of Aristotle’s three by applying logos, ethos, and pathos to design. While communication is modes of persuasion. traditionally tied to the spoken or written word, Buchanan identifies design as a form of communication in that it attempts to persuade potential product users that 1) the product is useful and 2) the designer’s values regarding practicality are important. Therefore, designers are constructing a persuasive argument when designing a product because design is being treated as “a mediating agency of influence between designers and their intended audience” (Buchanan 1985).

Buchanan identifies three elements that form a design argument, elements of which are Aristotle’s three modes of persuasion applied to design studies (1985). The logos of design, technological reasoning, is argued to be the foundation of the design argument. Products persuade in this way when RHETORIC AND EMOTIONAL APPEALS IN VISUALIZATION 19

they meet needs in a reasonable and convenient way. The second element is the ethos of design: the character of the product. Products have character because they reflect their designer in some fashion. Lastly, the design argument includes pathos, or emotions: “the problem for design is to put the audience of users into a frame of mind so that when they use a product they are persuaded that it is emotionally desirable and valuable in their lives” (Buchanan 1985). Through this appropriation of traditional rhetorical theory to the greater field of design, Buchanan not only recognizes the rhetoric of designed products, but the triad of elements that form the design argument.

Rhetoric in Data Visualization

“Similar to a photograph’s relationship to reality, visualizations do not capture reality as found in data but rather present a particular angle on it. Depending on the intention of the designer, visualizations can be used to influence, manipulate, and empower viewers in many ways” (Dörk et al. 2013).

All visualizations are subjective and interpretive to some extent. Additionally, traditional boundaries in visualization are expanding to include voices from other perspectives, like those that are artistic, causal, and narrative (Dörk et al. 2013). Therefore, Dörk et al. (2013) outlined a critical approach that examines the intentions of visualizations and possible implications. The approach, which promotes disclosure, plurality, contingency, and empowerment, takes a critical step towards community acknowledgment and assessment of the subjective nature of visualizations and holds designers accountable for their design decisions.

From the artistic perspective, Viegas and Wattenberg (2007) acknowledge the shift in treating visualization as a neutral, analytic tool to one that expresses a particular point of view. They propose a radical change in attitude by visualization designers to embrace the ability of visualizations to be both persuasive and analytical.

Kostelnick (2008) outlines several methods of visual rhetoric in data visualization over the past half-century, using clarity as a touchstone. He starts with the rhetoric of sciences, which embodies the rational, cognitive- perceptual approach to visualization design with a universal idea of clarity. He then moves from this position of rhetoric to describe the rhetoric of adaptation, social rhetoric, and the rhetoric of participation, which considers other facets of clarity and different considerations about the audience.

Rawlins and Wilson (2014) focus on the rhetoric of agency in interactive data visualizations. With levels of interactivity now embedded in the majority of digital visualizations, the forming of the communicated message shifts 20 RHETORIC AND EMOTIONAL APPEALS IN VISUALIZATION

from full authority of the designer to a collaborative effort between user and designer, creating a co-created rhetorical space. This elevates the role of the user, who is no longer just receiving the message but has agency to actively participate and create. To better grasp the continuum of interactivity that digital visualizations provide, Rawlins and Wilson identify several levels of interactivity, from static to unconstrained. For each level, Rawlins and Wilson determine the balance of agency between designer and user across three rhetorical elements: the data, design, and message.

While literature on the rhetorical nature of visualizations is expanding, there lacks empirical research on the persuasive power of visualizations. Pandey et al. (2014) began to fill this gap by conducting a study comparing the representation of data in tabular form to its graphical form for a variety of topics. The results of the study suggest that visualizations have the potential to be more persuasive when the person does not already have a strong opinion on the topic.

Deception More common than the term rhetoric in visualization literature is the negatively connoted term bias (Hullman and Diakopoulos 2011). This is in part due to early theory emphasizing visualization as an analytic tool, such as Card et al. (1999), which stresses instant clarity and minimal intervention by the designer (Tufte 2001). Under this perspective, any bias introduced by the designer is expected to hinder the effectiveness of the visualization.

Also, with the increase in use of visualizations as persuasive devices, the greater the potential is for misuse and deception (Pandey et al. 2015). Pandey et al. define deceptive visualizations as “a graphical depiction of information, designed with or without an intent to deceive, that may create a belief about the message and/or its components, which varies from the actual message” (2015). A plethora of literature educates and warns against the deceptive use of visualizations. In his book, How to Lie with Statistics, Huff (1954) brought to light several ways in which statistics can misinform. Tufte (2001) developed the concepts graphical integrity and lie factor to describe how visualizations can distort data and deceive the audience. In a similar light, Jones (2011) and Monmonier (1996) focus on deception for particular uses and types of visual representations of information.

While a fair amount has been said on the topic of deception in visualizing data, very little research has empirically evaluated deception techniques. Pandey et al. (2015) contributed research to the topic by testing two categories of deception in the message of a visualization: exaggeration or understatement of the message, and reversal of the message. Techniques tested that exaggerate or understate the message 1) truncated the axis, 2) represented quantity as area size, or 3) distorted the angle of a trend in a line chart. The fourth technique tested inverting the axis of a chart, leading to a RHETORIC AND EMOTIONAL APPEALS IN VISUALIZATION 21

perceived reversal of the message. Unsurprisingly, their results confirmed that these techniques lead to a significant amount of misinterpretation.

Storytelling Practitioners are recognizing data visualization as a communication tool for telling a story. Offenhuber (2010) explores the concept of visualizations as visual anecdotes, from stories connected to a single data point to a story told by the visualized data collectively. In his conversation with designer Mauro Martino, Martino explained that “usually there are several stories in the data – you have to select one. You can change the visualization by putting emphasis on a different story.”

With the increased use of data visualizations to tell a story, Segel and Heer created a design space for better interpretation of narrative visualization techniques (2010). Their framework consists of segmenting narrative visualizations by three distinctions: genre, narrative tactics that are visual, and narrative tactics that are structural. Building on Segel and Heer’s work, Hullman and Diakopoulos (2011) devise an analytical framework to better understand the rhetorical techniques in narrative visualizations. They identify four editorial layers where rhetorical decisions are made: data, visual representation, textual annotations, and interactivity. Technique categories in their framework include information access rhetoric, provenance rhetoric, mapping rhetoric, linguistic rhetoric, and procedural rhetoric.

Cartography A great deal has been said on the persuasive nature of cartography. Robinson and Petchenik (1975) called for cartography to be examined as a communication system, emphasizing the relationship between map maker and map consumer. Harley argued for establishing a cartographic ethics, stating that “each map is a manifesto for a set of beliefs about the word” (1991). Kent (2016) emphasizes the design of maps to convey a political message. Tyner (1982) identifies persuasion techniques utilized to manipulate cartographic elements: 1) the distortion of scale and shape, 2) the selection of information displayed, 3) symbolization, and 4) the choice of textual language.

Muehlenhaus identified four rhetorical styles of maps (2010) and studied their impact on recall, opinion, interpretation, and trust (2012). Muehlenhaus (2013) goes on to argue for academic cartographers to “start rigorously exploring how persuasive maps construct and articulate their messages in different contexts.” He leads by example, presenting initial findings on what components of a persuasive map make it persuasive. Among his findings were that 1) inappropriate use of visual elements were only found in 7.4% of the persuasive maps in the sample and 2) emotive symbols were among the top five important variables in a persuasive map, most often found in dynamic maps. 22 RHETORIC AND EMOTIONAL APPEALS IN VISUALIZATION

EMOTIONAL APPEALS IN VISUALIZATION

One point in the rhetorical triangle is pathos, the appeal to the emotions of the audience. In data visualization, this point is not only the least emphasized of the three, it is also the point that practitioners are typically encouraged to minimize. Lately, a rich discussion has been unfolding within the visualization community as to the role of emotions, particularly empathy, in visualizations. While there may be inconsistencies with the use of empathy, sympathy, and compassion (Boy et al. 2017), the conversation draws attention to the idea of eliciting an emotive response by visualizing data. To do so, however, is not an easy feat; research in psychology has found that a large barrier exists on the ability of data to bring about feeling. This section provides recent psychological research regarding the numbing effect of numbers, the current conversation in the visualization community, techniques proposed to evoke emotion in visualizations, and existing visualization research around these techniques.

Numb-ers Mother Teresa famously said “If I look at the mass I will never act. If I look at the one, I will.” This quote epitomizes the struggle between emotions and numbers; it is easier to feel for one individual than feel for the masses. Psychologist Paul Slovic calls this a “fundamental deficiency in our humanity” (2007).

Understanding this deficiency first requires looking into how humans think. Psychologists distinguish two systems of thinking: the experiential system and the analytics system (Epstein 1994). The experiential system is the feelings system; it is the intuitive, fast-processing system that helped us survive throughout evolution. The experiential system is associated with affect: the feeling that something is good or bad. The analytical system, on the other hand, is relatively new in comparison. This system is associated with logical thinking and slow processing. The theory that these two ways of thinking are fundamentally different begins to shed light on why we struggle to feel for large numbers of people.

When considering how we as human beings value saving human lives, we might expect that we give equal value to each saving each human life. This perspective, likely attributed to the analytics system, is represented visually in Figure 2.3. Unfortunately, research by Slovic and others grimly supports a perspective tied to the experiential system.

In one experiment, Small et al. (2007) allowed participants to contribute $5 to Save the Children. Participants were either shown an identifiable victim, statistical victims, or the identifiable victim with statistical information. Unsurprisingly, the resulting donations were far greater for the identifiable victim. Surprisingly, however, the incorporation of statistical information with the identifiable victim significantly reduced contributions. RHETORIC AND EMOTIONAL APPEALS IN VISUALIZATION 23

Kogut and Ritov (2005) found that people are more likely to donate more money to one person than eight people. Building upon this experiment, results from a study by Västfjäll et al. (2014) found that the same effect, coined as psychic numbing (Lifton 1967), was found when the number of children was two. This research highlights how limited our capacity to feel is, which immediately decreases when considering more than one person. Reconsider then how we value saving human lives, which looks more like Figure 2.4. This model helps explain what Slovic calls our “collapse of compassion” (2007). VALUE OF LIFE OF VALUE LIFE OF VALUE

NUMBER OF LIVES NUMBER OF LIVES

The collapse of compassion with groups of people confirms what people Figure 2.3 have generally known for a while: numbers representing lives do not A model for the logical value communicate the importance of those lives. Simply put, numbers numb. of human life (Slovic 2007). Those advocating for a strictly rational approach to convey information may be completely content with this knowledge, but it is somewhat hollow: “without affect, information lacks meaning and won’t be used in judgment Figure 2.4 and decision making” (Slovic 2007). Where numbers fail, images succeed; A model for the value of it is the image that holds the key to conveying meaning and affect (Slovic human life, accounting 2007). In light of this disconnect between numbers and feeling, can visual for psychic numbing representations of data help bridge the gap? (Slovic 2007).

Discourse among Practitioners Given the psychological research, it is understandable why practitioners are skeptical about visualizations evoking empathy and related emotions (Cairo 2016; Mushon 2015; Harris 2015). The first main reason for the skepticism is that visualizing data is a naturally rational endeavor, therefore visualization and empathy do not naturally go together. The second main reason is that rational thinking is more fair than emotional thinking, where eliciting empathy can ultimately mislead (Cairo 2016). This is seconded by psychologist Paul Bloom who says “our public decisions will be fairer and more moral once we put empathy aside” (2014). While practitioners cite Bloom’s argument against empathy, he ultimately argues for embracing an alternative emotion: non-empathetic compassion. 24 RHETORIC AND EMOTIONAL APPEALS IN VISUALIZATION

Zer-Aviv (2016) and Rost (2017) acknowledge the tension between empathy and rationality. However, Zer-Aviv (2015) wonders if a desensitized, quantified view of data prevents a deeper understanding of human problems. In her talk, Rost advocates for a dual approach: “We should make people care about a topic and then explain how to do and what to do in rational terms. Making decisions with your feelings is a bad idea. But starting to care about a topic, here feelings can be really, really important” (2017).

Several voices in the discussion offer the U.S. Gun Deaths Visualization by Periscopic (2013) as an exemplar for displaying the quantitative and evoking emotion simultaneously. The visualization represents those in the US killed from gun violence, in 2013 and 2010 respectively. In this visualization, Periscopic goes beyond showing the number of people that died by showing the age of those that died and their potential lifespan. The difference between the two is phrased as “stolen years”.

The visualization begins with a thin horizontal axis, representing years, stretched across a black background. A bright orange arc animates out of the 0 point on the axis, but abruptly stops before reaching its peak. At this RHETORIC AND EMOTIONAL APPEALS IN VISUALIZATION 25

point, a label appears and reads ‘Alexander Lipkins, killed at 29’. Where the arc stopped short, a dot falls to the axis, resting just below it. The arc then continues its trajectory in gray, and upon reaching its completion another label appears: ‘could have lived to be 93’. The animation accelerates, drawing four more arcs for four other gun victims with different ages of death and projected stolen years. Then, overwhelmingly, a burst of arcs shoots out of Figure 2.5 the 0 point. Their dots quickly rain down from the age of death, collecting U.S. Gun Deaths at their final resting place under the axis. Through the animation, two large Visualization (Periscopic numbers rapidly tally on the screen: people killed and stolen years. 2013).

Zer-Aviv (2015) argues that the empathetic and poetic framing of this chart is the source of its power and effectiveness. Cairo (2013) describes the visualization as a “triumph of emotion made functional”, however he cautions the representation of uncertainty; Periscopic calculated expected age of death through statistics, but represented stolen years as precise numbers, which may mislead readers. By choosing to represent stolen years, Persicopic traded a level of certainty for affect. Despite this, their gun deaths visualization often sits in the center of the discussion as an exemplary visualization for evoking empathy. 26 RHETORIC AND EMOTIONAL APPEALS IN VISUALIZATION

In a discussion with Zer-Aviv, Periscopic co-founder Kim Rees (2016) states that it is almost mandatory to need the audience to react. Both designers reframe data as another part of language. In doing so, visualizing data immediately becomes a means to convey an opinion. When the community understands this, Rees and Zer-Aviv argue that practitioners will be more comfortable with the concept of incorporating emotions in the language. The two also emphasize ethos, since designers use data in their arguments and therefore should be clear about their arguments; Zer-Aviv finds it more ethical for the designer to be explicit about their opinion rather than to hiding it under a “shred of objectivity” (2016).

Lambert (2016) advocates for visualizations that go beyond affecting the reader by providing them a way to act. In this way, visualizations can promote social change. Lambert celebrates the ability of the U.S. Gun Deaths Visualization to combine numbers with emotional resonance but points out the missed opportunity of not providing agency for the audience, subjecting them to compassion fatigue. Rost seconds the importance of a call to action, adding that designers and data journalists have a responsibility to give people an action to take after being moved (2017).

In response to Lambert, Rees (2016) explains that the U.S. Gun Deaths Visualization was meant to be more inclusive rather than prescriptive. Her problem with a deliberate call to action is that they often exclude those in the other camp of thought, resulting in “preaching to the choir”, and therefore does not see the need for every affecting visualization to include one. Zer- Aviv (2016) considers whether designers can get the best of both worlds: bringing people into the conversation that may not have otherwise, but also veering them towards a desired action.

This discussion among practitioners is drawing attention to the growing desire for humanized, personalized visualizations. Lupi (2017) calls this data humanism, and it applies the concept of pathos from the first steps of data collection through design implementation. Lupi pushes for a new wave of visualization that connects numbers to what they stand for: knowledge, behaviors, and people. To move towards creating more thoughtful and meaningful visualizations, Lupi states that we need to embrace complexity, move beyond standard visual forms, sneak context in, and remember the imperfect nature of data. Lupi embraces complexity, favoring rich and complex storytelling over immediate clarity and simplification. To move beyond standards, Lupi calls for customization of visual form to uniquely work with the data at hand, rather than “throwing technology at the problem” with “one-size fits all” visual forms. In order to introduce novel ways of thinking and designs, Lupi encourages sketching with data. When incorporating context, Lupi encourages subjectivity, intimate stories, and spending time with data. Lastly, when data visualization embraces imperfection and approximation, it allows visualization designers “to RHETORIC AND EMOTIONAL APPEALS IN VISUALIZATION 27

envision ways to use data to feel more empathetic, to connect with Figure 2.6 ourselves and others at a deeper level” (Lupi 2017). Data Humanism (Lupi 2017).

The Dear Data project (Lupi, Posavec, and Popova 2016) demonstrates this concept of data humanism. Lupi and Posavec sent one another weekly data sketches on a postcard for a year, sharing data with one another about their personal lives. Figure 2.7 displays week 42 of this project. The drawings embrace the pathos and make the designer visible through hand-drawn expressive lines and personal data. To interpret the drawings, the legend 28 RHETORIC AND EMOTIONAL APPEALS IN VISUALIZATION

Figure 2.7 Week 42 from Dear Data: “Laughter” (Lupi, Posavec, and Popova 2016). RHETORIC AND EMOTIONAL APPEALS IN VISUALIZATION 29

on the other side of the postcard is necessary, but the beauty of the visual interpretation of data can be admired without immediate clarity.

Techniques Among the discourse about emotions in visualization, a handful of practitioners and academics have also offered techniques which may elicit emotions. Harris (2015) investigated three techniques that data journalists have used to anchor their graphics in empathy and connect readers to the dots in a graphic. The first is the near and far, informed by Klein (2013), which incorporates both micro and macro views of the data. The near view can allow the user to locate themselves in the data.

Harris’ second technique is putting people first, which simply means Figure 2.8 showing the people behind your data in images. This technique can partner Images of children from the with data, but it can also stand alone. Slobin (2014) learned this lesson while Trials story in the Wall Street working on a project about a group of families with children who have a Journal (Marcus, 2014). fatal disease. In the end, she did not create graphics for the story because pictures of the children were enough. “We need to remember that behind the data are stories and inside those stories are people and those people are connected to the statistics in a way that we never will be” (Slobin 2014).

The third technique Harris investigates is the use of wee people, which are silhouettes of the human form. Wee people relate to a greater system of pictographic elements to display information, known as ISOTYPE: the International System of TYpographic Picture Education, first described by Brinton (1914) and later defined by Neurath (2010). Cairo (2016) doubts the effectiveness of wee people to foster emotion, but finds promise in the use of near views.

Rost (2017) presents five techniques that may appeal to emotions: 1) make use of colors, 2) show what you’re talking about, 3) show what the data 30 RHETORIC AND EMOTIONAL APPEALS IN VISUALIZATION

would mean for your experience, 4) zoom into one dot, and 5) show the mass as individuals. The first two techniques relate to style; the second technique, show what you’re talking about, is synonymous with Harris’ proposed technique of using wee people. The remaining three techniques focus on getting to know the data points. The third technique involves showing the audience what it means to be one of the data points in the visualization. This can be done by showing the audience data for their location or their situation.

The fourth technique of zooming into one dot inserts an individual’s story in the visualization. The purpose of incorporating the story of one person is to make the individual visible. Cairo (2016) extends Harris’ near and far approach to apply to this form of a near view. While this technique shows the individuality of one data point, the fifth technique Rost presents is about representing each data point individually in the visualization. It can also be implemented through text to help the audience make sense of large numbers.

Kostelnick (2016) pays particular attention to the historical use of emotional appeals in visualization, identifying their use in the later nineteenth century and their re-emergence in the present day. He categorizes today’s pathos techniques in digital visualizations as 1) sensory stimulation, 2) personal proximity, and 3) expressive displays. Within the bucket of sensory stimulation are the use of color, novel forms, and animation. The second category focuses on principles of personal proximity, where the data relates to the audience through time, their spatial location, or their interests. This aligns with the near and far concept (Harris 2015) as well as Rost’s technique of showing what the data means for the audience’s experience (2017). The third category focuses on expressive displays, which is more of a type of visualization than an individual technique. Kostelnick highlights expressive displays that are referred to as data art (Yau 2013) and the increasing trend of visualizing personal data.

Pathos Research While proposing techniques that are likely to appeal to emotions, Rost (2017) calls for a better understanding on whether the techniques do indeed evoke emotion. In a similar light, Harris (2015) calls for a systematic exploration of when certain techniques work and when they do not. Many techniques in which practitioners recognize as having the ability to stir emotions may also be identified as types of visual embellishments. Nigel Holmes, who encourages the use of visual embellishments, asserts that a chart must be engaging and that “the purpose for making a chart is to clarify or make visible the facts that otherwise would lie buried in a mass of written materials” (1984). A body of research has begun to develop around the impact of visual embellishments in visualization. RHETORIC AND EMOTIONAL APPEALS IN VISUALIZATION 31

The term visual embellishments takes on different meanings, like decorations and other non-essential imagery (Bateman et al. 2010) or the more encompassing term of visual metaphor, where visual embellishments are considered “a form of non-linguistic rhetorical figures” (Borgo et al. 2012). Despite different terminology use, visual embellishments have consistently been a subject of debate (Borgo et al. 2012) as some chart design experts like Tufte advocate strongly against their use, favoring a minimalist style that maximizes the proportion of ink or pixels used to represent the data (Tufte 2001). Therefore, several empirical studies have assessed the impact of visual embellishments, but not in the context of emotions.

Bateman et al. investigated the effects of visual embellishments on Figure 2.9 interpretation accuracy and memorability (2010) and, when reflecting on Two classes of granularity in the results, considered the participant’s emotional response as a potential the anthropographics design hidden factor in the increase in memorability with embellished charts. space (Boy et al. 2017). Borkin et al. (2013) evaluated what specifically makes visualizations memorable, and found that several factors like color, novel forms, and embellishments increased memorability. Their study was later built upon to investigate what visual elements had an effect on visualization recall and understanding (Borkin et al. 2016). Borgo et al. (2012) evaluated the ability of visual embellishments to aid in memorization, visual search, and concept comprehension. Vande Moere et al. (2012) investigated the impact of three visualization styles that differed in terms of visual and interactive embellishment: analytical, magazine, and artistic, where the analytical style was referred to as non-embellished. The impacts analyzed included usability, insight types, and insight depth. Haroz et al. (2015) tested how the use of the ISOTYPE style impacts memory, speed of finding information, and engagement.

UNIT AGGREGATE 32 RHETORIC AND EMOTIONAL APPEALS IN VISUALIZATION

EXPRESSIVE

ABSTRACT REALISTIC

NEUTRAL GENERIC ICONIC UNIQUE

This 10-year-old is Tariq is internally Internally displaced internally displaced displaced

GENERIC ICONIC UNIQUE

GRID GROUPING ORGANIC GROUPING RHETORIC AND EMOTIONAL APPEALS IN VISUALIZATION 33

An empirical study by Boy et al. (2017) also investigates the use of Figure 2.10 pictographic elements, but appears to be the first to answer the call from Representations of the Rost and Harris for a better understanding of how visualization techniques human shape, varying in impact the audience’s emotions. This study tests if “anthropographics” realism and expressiveness elicit more empathy and encourage more prosocial behavior than the (Boy et al. 2017). standard pie chart. Anthropomorphized visualizations were created based on the following design decisions, shown in Figures 2.9-12: granularity, human shape, unit labeling, and unit grouping. Visualizations incorporated different combinations of these design decisions to represent the anthropographic design space.

The tested charts were embedded in human rights stories about Syrian children. The stories were presented in a slideshow form following a five-step narrative structure: 1) context, 2) geographic overview, 3) simple demographic statistic before the crisis, 4) presentation of the same statistic today, and 5) a takeaway consideration on how children are suffering from the situation. Steps three and four varied in the chart used, Figure 2.11 either as a anthropographic or a pie chart. Despite various adjustments Variations of unit labeling, over seven experiments, the results did not find that the anthropographics from generic to unique elicited more empathy than the standard pie chart, or that they encouraged (Boy et al. 2017). more social behavior.

Figure 2.12 Two types of unit grouping in the anthropographics design space (Boy et al. 2017). 34 RHETORIC AND EMOTIONAL APPEALS IN VISUALIZATION RHETORIC AND EMOTIONAL APPEALS IN VISUALIZATION 35

From a distance, it’s easy to forget the dots are people.

Jake Harris

Techniques for Appealing to Emotion: A Taxonomy

I present a working taxonomy of rhetorical techniques that have been Figure 3.1 proposed by various practitioners and researchers to invoke emotions. The YES vs NO data visualizer taxonomy categorizes techniques by their primary intention: to engage, (“Domestic Data Streamers” humanize, or personalize. Given the current interest in the visualization 2013). community to identify techniques, this taxonomy serves as a foundation for the community to build on.

Intention Technique Source(s) engage make use of color (Kostelnick 2016; Rost 2017) create novel forms (Kostelnick 2016) animate through a narrative (Kostelnick 2016) humanize zoom into one dot (Harris 2015; Rost 2017) show the mass as individuals (Rost 2017) show what the data are about (Harris 2015; Rost 2017) put a face to the data (Harris 2015) personalize connect to the relevance of time (Harris 2015; Kostelnick 2016; Rost 2017) connect to the audience’s space (Harris 2015; Kostelnick 2016; Rost 2017) connect to the audience’s interests (Harris 2015; Kostelnick 2016; Rost 2017) 38 TECHNIQUES FOR APPEALING TO EMOTION: A TAXONOMY

ENGAGE

Table 3.1 Techniques that engage audiences do so through sensory stimulation. Previous page: Taxonomy Three techniques that stimulate the senses are the use of color, the creation of pathos techniques in of novel forms, and the use of animation to tell a narrative. Color and novelty visualizations. have been embraced in static visualizations historically, particularly in the nineteenth century (Kostelnick 2016). The birth of digital visualizations enabled the use of animation to engage the viewer and walk them through a narrative.

While color historically has been a constraint in print visualizations, it appears ubiquitously in digital data visualizations. Both Kostelnick (2016) and Rost (2017) identify the use of color to evoke emotion. Color can evoke an emotional response on three levels: psychological, aesthetic, and cultural (Richards and David 2005). Psychologically, we respond immediately to the difference between warm and cool colors; the former causes excitement or anxiety and the latter is often more calming. Aesthetically, colors can please or excite us if the hues complement one another. Poor color selection can also disrupt our aesthetic sensibilities. Culturally, color can trigger conventional interpretations. For instance, red is often related to joy and festivity in Chinese culture, but it is more likely to conjure caution and anxiety in the U.S. Due to the potential power of color, designers need to be constantly vigilant of their color selection.

Today, many visual forms have become established conventions that users have learned to read. These socially-constructed genres stabilize the user experience. Therefore, the exposure to novel forms, or established forms that have been reshaped, can evoke surprise, excitement, and perhaps initial bewilderment (Kostelnick 2016). Lupi (2017) describes this as moving beyond standards, where the design is uniquely constructed for the data problem at hand without limitations or boundaries of standard genres. However, designers must beware of assuming that the audience can easily interpret a novel chart. This assumption cannot be safely made even with standard chart types for a general audience (Goo 2015). Designers implementing novel forms should be cognizant of sacrifices to interpretability and should stray from chasing novelty for novelty-sake.

A powerful way to engage the audience and evoke an emotional response is to tell a narrative with data through animations that change with space or time. This dynamic form of storytelling brings out a sense of immediacy and can elicit varying emotions as the information is animated. A prime example is Gapminder (“Gapminder Tools” 2016), which is arguably one of the most famous uses of animation in data visualizations. The visualization is a scatterplot of countries, telling the story of our collective health and wealth through 215 years. As the chart animates through time, the country circles move along the coordinates that represent life expectancy and income TECHNIQUES FOR APPEALING TO EMOTION: A TAXONOMY 39

per person. Watching the story unfold through the animation encourages Figure 3.2 a variety of emotions, such as curiosity and excitement as the overall Gapminder (“Gapminder situation for humanity improves (Kostelnick 2016). The positive response Tools” 2016). is dampened when the countries noticeably lagging behind are identified. While Kostelnick focuses on animation as a technique that stimulates the 40 TECHNIQUES FOR APPEALING TO EMOTION: A TAXONOMY

senses and narrates with data, animation also has the potential to humanize the data. The U.S. Gun Deaths Visualization (Periscopic 2013), highlighted in the previous chapter, does so by animating life arcs and dropping the once- active dot when the life arc is cut short.

HUMANIZE

Efforts to humanize data help the audience connect the data to what they are about, which is most often people. When the audience is able to see the humanity behind the data, they are more likely to care about what is being visualized. Designers have proposed to humanize data by zooming into one dot, showing the mass as individuals, showing what the data are about, and putting a face to the data.

A humanizing way to tell a story within a data visualization is by zooming into one dot. To zoom into one dot, the designer highlights one data point, which represents one individual, and inserts the story of that individual in the visualization. This adds qualitative information to a quantitative display and makes the individual’s story visible, connecting the human to the data and enabling the viewer to see the individual. This technique is described by Rost (2017) and is applied to near and far approach (Harris 2015) by Cairo (2016), where zooming into one dot represents a close, macro view of the data.

While zooming into one dot dives into the story of one individual, showing the mass as individuals draws attention to every individual in the data. This technique, presented by Rost (2017), aims to visually tackle the issue of desensitization that often accompanies large numbers. By visually representing each individual, the quantity of individuals can be more easily realized. When the viewer is aware that each visual element represents one individual in the data, they are able to draw that connection between data and individual, thus humanizing the data and encouraging an emotional response.

Showing the mass as individuals can be done with abstract forms. A technique which humanizes the forms is to show what the data are about. This refers to the use of wee people: pictorial, anthropomorphized silhouette forms representing data. For instance, if the data is representing people, silhouettes of human forms can be used instead of dots. The intention of representing data as a more humanized form is for the audience to more readily connect to the data to what they represent. Rost (2017) and Harris (2015) both discuss this technique. Boy et al. (2017) investigate variations of the form, ranging in realism and expressiveness. Boy et al. also experiment with organic grouping of the silhouettes as opposed to rigid, grid grouping. Harris discourages the use of individual wee people when the quantity reaches a certain threshold. While Cairo (2016) acknowledges this technique, he questions its reliability for the task at hand. TECHNIQUES FOR APPEALING TO EMOTION: A TAXONOMY 41

The Out of Sight, Out of Mind visualization (Pitch Interactive 2013) Figure 3.3 humanizes data by showing the mass as individuals and as wee people, with Out of Sight, Out of Mind a very deliberate use of animation and color. The visualization calls attention visualization (Pitch to the deaths of innocent civilians from US drones strikes in Pakistan. The Interactive 2013). visualization is oriented on a timeline spanning 2004 to 2015. Animated arcs drop onto the timeline and mimic an explosion with an expanding and disappearing red circle. The total deaths of that drone strike present itself at the same point on the timeline as an inverted bar chart. The bars are made up by ticks for each human life lost, enabling each casualty to be 42 TECHNIQUES FOR APPEALING TO EMOTION: A TAXONOMY

represented individually. The bars are colored to categorize deaths being children, civilian, high profile, or other. The color encoding draws the most attention to children deaths with bright red, and high-profile deaths with white. On the Victims tab, a monthly view shows the counts as horizontal collections of wee people, where each human figure represents a victim.

Figure 3.4 The final humanizing technique shows what the data are about with a Faces of the Dead greater sense of realism; this technique, proposed by Harris (2015), is (Dance et al. 2010). about putting a face to the data. This may be the individual face that accompanies the individual’s story when you zoom into one dot, or the mass is represented by individual faces. This technique is implemented in Faces of the Dead by The New York Times (Dance et al. 2010), an interactive graphic that displays photos of fallen U.S. service members in Iraq or Afghanistan. The Photos tab of the graphic emphasizes one portrait image of a service member. Each pixel that makes up the image represents another individual in the dataset. By selecting a pixel, the image changes to the newly selected service member, requiring the viewer to confront the face of a fallen soldier and selected information about them. TECHNIQUES FOR APPEALING TO EMOTION: A TAXONOMY 43

PERSONALIZE

Philosopher George Campbell asserts that audiences care about, and are more likely to emotionally invest in, people and events near them (1776). Three circumstances he lists that are instrumental in “operating of the passions” are proximity of time, connection of place, and importance (1776). These three circumstances are applied to visualizations as techniques that emphasize a close proximity between data and viewer (Kostelnick 2016). By personalizing the data in this way, the audience can more deeply connect to the data that is most relevant to them. These proximity techniques can be implemented through interactions, allowing the user to personalize the information they receive. Personalization may also mean that the data is already filtered to have a particular connection with the audience. These techniques relate to Harris’ near and far approach and Rost’s technique of showing what the data means for the audience’s experience.

Typically, data visualizations represent past events. If the event is more recent in time, it is more likely to be affecting to an audience than a visualized event that is more distant in time. With today’s technology, we can visualize data in real time. The ability to monitor a visualization populated by ever-present data fosters a connection between user and data. If the user is monitoring data on a topic they have involvement in, like financial stocks, real-time data can evoke fear and anxiety (Kostelnick 2016). This technique is less about customizing the data from person to person, but rather personalizing to meet temporal relevancy with the viewer. Connecting to the relevance of time is a technique that is less accessible as it depends on the data that is available and what the data is about.

Like events that are close in time, individuals are more likely to emotionally invest in people and events geographically near them. This includes events like natural disasters, crime, and spread of diseases. Visualizations showing data relating to these topics may engender empathy if they are distant to the viewer. However, if the threat is near, the visualization has greater potential to elicit emotions like fear and anxiety (Kostelnick 2016). Visualizing data that connects to the audience’s space triggers and heightens an emotional response.

When a user is able to interact with and explore a visualization, their individual interests begin to dominate their visual experience. This invokes Campbell’s circumstance of importance, the type of proximity that connects to the interests of the audience. The Jobless Rate for People Like You visualization by The New York Times shows an unemployment rate timeline from 2007 to 2009 (Cox, Carter, and Quealy 2009). When the user selects a race, gender, age group, and education level, the unemployment rate line that is appropriate for their situation is highlighted. In doing so, the user sees the data that is closest in proximity to them, and therefore most relevant to them. 44 TECHNIQUES FOR APPEALING TO EMOTION: A TAXONOMY

Figure 3.5 The Jobless Rate for People Like You from The New York Times (Cox, Carter, and Quealy 2009). TECHNIQUES FOR APPEALING TO EMOTION: A TAXONOMY 45 46 TECHNIQUES FOR APPEALING TO EMOTION: A TAXONOMY TECHNIQUES FOR APPEALING TO EMOTION: A TAXONOMY 47

Passion is the mover to action, reason is the guide.

George Campbell

Evaluating the Emotional Effect of Proximity Techniques

While diving into the conversation about emotions in the visualization community, a research opportunity became apparent: to empirically evaluate techniques that are said to appeal to emotions. Several techniques have been proposed, but very little research exists in assessing if these visualization techniques do indeed evoke emotion. This evaluation intends to shed light on the emotional power of proximity techniques, which personalize the dataset visualized so that it has increased relevancy and connection to the user. Specifically, I test techniques that connect to the relevance of time, the space of the audience, and the interests of the audience. This chapter outlines the value of testing pathos techniques, what these proximity techniques are, how emotions are measured, the methodology of this study, and discusses the analysis process and results.

THE VALUE

Visualizations today often communicate information about issues central to Figure 4.1 our humanity to the general population. Journalism outlets have popularized One month at an animal the use of visualizations as a tool to convey powerful messages about shelter. My prototype for the topics like health, politics, climate change, education, humanitarian crises, thesis exhibition. and the list goes on. One example is Bloomberg’s visualization of record- breaking years for temperature over 138 years (Randall and Migliozzi 2018).

Advocacy groups are increasingly utilizing data and their visual representation to engage audiences in a cause they care about. For 50 EVALUATING THE EMOTIONAL EFFECT OF PROXIMITY TECHNIQUES

Figure 4.2 instance, Greenpeace developed an interactive research tool called Exxon Earth’s Relentless Warming Secrets to explore how far ExxonMobil’s funding for climate change denial Sets a Brutal New Record in has reached (On and Balkin 2005). A deeper look into the emotional 2017 (Randall and Migliozzi impact of visualizations has value for designers like those in journalism 2018). and advocacy. Practitioners that wish to create engaging, humanizing, or personalized visualizations will be able to utilize techniques that have been empirically evaluated. In doing so, designers and advocacy groups can better enable audiences to construct meaning from data, thus connecting it to the human experience.

As the power of visualizations as a communication tool increases, so does the potential to misuse the power. Therefore, it is important to consider their rhetorical capacity from every angle. As stated in the second chapter, Aristotle is credited with laying some of the foundational groundwork for how we approach rhetoric today. This groundwork identifies three modes of persuasion as key ingredients to a persuasive argument. Logos, the appeal to logic and reason, is highly valued in the visualization field. Ethos, EVALUATING THE EMOTIONAL EFFECT OF PROXIMITY TECHNIQUES 51

the appeal to character, is strived for as designers pursue building credibility. Pathos, however, the appeal to emotions, is a critical aspect of visualization that is often overlooked.

These three modes make up the rhetorical triangle and all serve a purpose in Figure 4.3 a successful argument. Therefore, appeals to emotion must be considered Exxon Secrets (On and Balkin when evaluating the influence of visualizations and particular visualization 2005). techniques. The more aware and informed the visualization community is about emotional appeals, the greater the critical lens to assess one other’s work and thus hold one another to a higher ethical standard.

PROXIMITY

Most methods for appealing to emotion are implemented in the visual layer of the visualization. However, the techniques tested in this study focus instead on personalizing the data so that it more closely connects to the viewer, relying on the data layer and the context surrounding the chart. 52 EVALUATING THE EMOTIONAL EFFECT OF PROXIMITY TECHNIQUES

ETHOS character of the speaker

LOGOS PATHOS logic of the message emotions of the audience

Figure 4.4 These techniques are referred to in this thesis as proximity techniques and The rhetorical triangle, they extend beyond the scope of data visualization. To reiterate Campbell’s reiterated. assertion, a person is more likely to emotionally invest in people and events close to them (1776). In The Philosophy of Rhetoric, Campbell identifies several circumstances that are the most influential in exciting emotions, three of which are importance, proximity of time, and connection of place.

While these circumstances were identified within the context of speaking, the visualization community has liberally employed these techniques since interactive visualizations were possible. When an individual is able to customize the information that a visualization displays, the visual experience that results is one that is dominated by their interests and what is important to them. This technique, referred to in this study as proximity to interests, cites Campbell’s circumstance of importance.

Unsurprisingly so, events are more affecting if they occurred recently in time. Proximity of time relates to both events in the past and events lingering in the future. With the future moving towards us and the past moving away, Campbell argues that when the evidence, importance, and the distance of objects are equal, the future is more affecting than the past (1776). However, the past has stronger evidence, formed from captured data. The future is filled with conjecture and uncertainty, although today data from the past is utilized to make projections into the future. Digital visualizations enable the ability to show data in real time, leaving the viewer in a constant state of anticipation. Depending on the type of information portrayed, real-time data can elicit a range of possible emotions. For instance, data regarding a natural disaster can evoke feelings like anxiety, terror, or relief (Kostelnick 2016).

A person’s emotional reaction to visualized data about a natural disaster is also dependent on where the event occurs in relation to them. Campbell EVALUATING THE EMOTIONAL EFFECT OF PROXIMITY TECHNIQUES 53

asserts that the connection of place, or spatial proximity, has a more powerful effect than temporal proximity, as it is permanent and therefore has a firmer ground of relation (1776). The impact an event has on the emotions depends on its spatial proximity; the natural disaster is more likely to promote empathy if it is geographically far, and risk or fear if it is too close for comfort (Kostelnick 2016). In data visualization, interaction enables a user to zoom into maps or select their location to show the data that is spatially relevant to them.

Campbell’s Circumstances Study Terminology importance proximity to interests proximity of time temporal proximity connection of place spatial proximity

While the relationship between proximity and emotion seems logical, Figure 4.5 the visualization community lacks empirical evidence that visualizations Campbell’s circumstances incorporating proximity techniques evoke any type of emotional response. mapped to the terminology This study begins to address the gap by evaluating this relationship within used in this study. the context of data visualization.

EMOTION

In order to evaluate the emotional impact of proximity techniques in Figure 4.6 visualizations, one must develop a foundational understanding as to how The emotional response triad emotions are measured. Emotion is a term used in everyday language, often (Scherer 2001). interchangeably with words like affect, feeling, and mood. The scientific community distinguishes these terms, although not with consensual definitions (Scherer 2001; Scherer 2005). Scherer defends the component process definition for emotion as “a process of changes in different components rather than a homogenous state” (2001). The three widely accepted components, physiological arousal, motor expression, and subjective feeling, are known as the emotional response triad (Scherer 2001).

physiological arousal

motor expression subjective feeling 54 EVALUATING THE EMOTIONAL EFFECT OF PROXIMITY TECHNIQUES

A gold standard does not exist for measuring emotion, and measurements often depend on which component of emotion the researcher is interested in capturing. In an ideal world, there would exist a comprehensive measurement for emotion that spans across all of the components. However, this measurement does not exist nor is it realistic (Scherer 2005).

This study focuses on the subjective feeling component of emotion. There is no objective method for measuring someone’s subjective emotional experience, so researchers rely on capturing subjective feeling through self-reporting. In self-reporting, researchers have the option to allow open responses or fixed responses, both providing their own advantages and disadvantages. While open responses do not confine participants to fixed answers, analyzing responses across groups can be quite challenging. Researchers have traditionally used two approaches for measuring fixed responses for subjective feeling: the discrete emotions approach and the dimensional approach (Scherer 2005).

The discrete emotions approach has existed since the origin of language as a way to describe how one is feeling. This approach involves a selection of semantic fields for emotion in natural languages. This approach became concrete with Darwin’s set of basic emotions (1899). Measuring discrete emotions has been done through the use of various scales. While the results that arise from using discrete emotions are easily interpretable, there exists issues in conducting statistical analysis of the results and comparing results across studies that use different sets of emotion labels (Scherer 2005).

The dimensional approach, pioneered by Wundt (1969), identifies three dimensions in which subjective feelings can be measured on: valence (positive-negative), arousal (calm-excited), and tension (tense-relaxed). Modern researchers focus on the dimensions of valence and arousal. In this approach, respondents are asked how positive or negative they feel, and how excited or aroused they feel. Therefore, one’s subjective feeling is identified by a location in the valence-arousal space. This approach lends itself well to statistical analysis, but is not straightforward in determining the specific emotions that were felt (Klaus R. Scherer 2005).

STUDY METHODOLOGY

This study is driven by the following research question: “Does the incorporation of proximity techniques increase the emotional impact of data visualizations?” In order to empirically evaluate this question, several small decisions around the topic, chart design, measurements, etc. assembled together to produce the study methodology, which is described in this section. EVALUATING THE EMOTIONAL EFFECT OF PROXIMITY TECHNIQUES 55

Topic This study focuses on the plight of shelter animals. It is also my topic of focus for the exhibition portion of this thesis (see Figure 4.1). The overpopulation of animal shelters can lead to the unnecessary euthanasia of animals. While animal welfare organizations spread awareness about animal adoption, only 34% of dog owners and 44% of cat owners adopt their pet from an animal shelter, humane society, or rescue group (Springer 2017).

Of all potential topics to select, ranging from those that carry little-to-no emotional weight, to those that carry a significant amount, this topic pulls closely towards the latter. By selecting a topic that does carry emotional weight, I can more likely ensure an emotional response and pick out nuanced differences in responses. Because the study compares differences in emotional responses between treatments, the analysis strips out the emotional weight that the topic initially comes with.

I selected this topic for multiple reasons. First, it is estimated that 68% of U.S. households own a pet. Asserting that even a larger percentage of the population enjoys the company of domestic animals makes the topic an interest to the majority of U.S. households. However, the percentage of dog and cat owners adopting their pets indicates there is room to improve the awareness of pet adoption. Second, research has shown that people are more likely to take part in prosocial behavior towards a recipient if that recipient is not responsible for their own plight (Lee et al. 2014). Third, the ability to incorporate comprehensive data in shelter animal advocacy has only been recently possible through national data collection efforts.

Data and Chart Design Advocacy for shelter animals typically takes the form of qualitative information, e.g. images, videos, and individual stories. Available numbers about shelter animals have primarily been estimates until now. Through a recent initiative by Shelter Animals Count (SAC), data is now being collected at the shelter level. This study utilizes data from the SAC national database of intake and outcome counts for 3,459 shelter animals across the U.S. for 2016 (Shelter Animals Count 2017). The data distinguishes live and non-live outcomes, which the designed chart highlights.

The baseline chart for this study takes the form of a stacked bar chart: one stacked bar for intake counts and one stacked bar for outcomes. The outcomes bar is further divided into live and non-live outcomes. Both the intake bar and outcomes bar identify counts for types of intake and outcomes. Labels and counts for these types are revealed by hovering over the respective bar.

To emphasize the count of non-live outcomes, text sits above the stacked bar chart. This text takes the form of a ratio of live to non-live outcomes. 56 EVALUATING THE EMOTIONAL EFFECT OF PROXIMITY TECHNIQUES

Figure 4.7 The presence of this text aids in the comprehension of the large counts, Baseline chart for aligning with the concept of showing a near view of the data in textual form the control treatment. (Harris 2015). Another pathos technique incorporated in this chart is the use of color. The colors selected to distinguish the live and non-live outcomes symbolize the movement of animals, like a traffic light; shades of green identify the animals that continue their life outside the shelter and shades of orange identify the animals whose journey was stopped short.

Figure 4.8 This chart has a simple design for greater interpretability across the Chart variations for each general population. A recent study by Pew Research Center found that proximity treatment. only 63% of the adult participants could accurately read a scatterplot (Goo 2015). This draws attention to the fact that interpretability of charts is a learned skill, a fact that designers must be cognizant of when designing charts for a broad audience.

Treatments The study includes four treatments: a control treatment and one for each of the three proximity techniques being tested: spatial proximity, temporal proximity, and proximity to interests. One chart is designed for each treatment. Because these proximity techniques are implemented in the data layer, the chart design remains the same across the treatment groups. The textual layer in the chart adjusts to support the changes in the data layer, aiding the saliency of the proximity techniques. Therefore, these techniques EVALUATING THE EMOTIONAL EFFECT OF PROXIMITY TECHNIQUES 57

PROXIMITY TO INTERESTS

TEMPORAL PROXIMITY

SPATIAL PROXIMITY 58 EVALUATING THE EMOTIONAL EFFECT OF PROXIMITY TECHNIQUES

are represented in the surrounding context of the graphic, and do not rely on a change in the graphic elements.

Figure 4.8 illustrates how the design varies per treatment. Two proximity techniques, spatial and interest proximity, require user input to determine the data subset of closest proximity to the participant. In order to implement the most salient representation of temporal proximity, the technique is simulated; the treatment for temporal proximity uses 2016 data but is framed as otherwise through the textual layer. The data layer is manipulated to represent a day’s worth of data, as opposed to a year, and is phrased as projected numbers for tomorrow. The intention of this design is 1) to make the data feel more relevant and comprehensible, and 2) to mimic temporal proximity implemented in visualizations that represent real-time or forecasted data.

Measurements Figure 4.9 The primary measurement in this study, Scherer’s Geneva Emotion Wheel Recreation of the (GEW) (2005), captures the participant’s emotional response to the chart. Geneva Emotion Wheel GEW, shown in Figure 4.9, is comprised of 20 emotion families located (Scherer 2005). along the circumference of the valence-control space. This allows for the differentiation of feelings in this two-dimensional space. The inclusion of a wide range of emotion families spanning throughout this space enables a variety of emotions to be captured. The circular presentation of the GEW EVALUATING THE EMOTIONAL EFFECT OF PROXIMITY TECHNIQUES 59

includes a five-point scale for each emotion family, represented as circles sequentially radiating from the center with an increasing radius. However, the implementation of this survey limited the feasibility of visually presenting the emotion labels in the valence-control space. Therefore the 20 emotion scales were listed linearly following their counterclockwise order in the wheel.

Since pathos is a mode of persuasion, this study incorporates a measurement for attitude change. To measure change in attitude, I followed the procedure of Pandey et al. (2014), utilizing a single-item Likert scale. The scale is first included to measure the participant’s level of agreement or disagreement with the following statement: an increase in animal adoptions will help decrease the number of animals euthanized in shelters. This measurement evaluates the participant’s initial attitude toward the statement. The statement and scale are then repeated following exposure to the chart in order to capture any change from the initial recorded attitude.

Accompanying the two measurements for emotion and attitude change are measurements for the participant’s involvement in the topic and their data visualization literacy. To better understand the conditions in which a person is persuaded under, persuasion researchers are attentive towards the concept of involvement (Johnson and Eagly 1989). This study aims to capture varying levels of involvement the participants may have with dogs or cats. Incorporating a question about involvement allows for an analysis of the relationship between involvement and attitude change.

This study also captures one aspect about the participant’s visualization literacy. As mentioned in the section about chart design, it is unwise to assume adequate visualization literacy across the potential participant population. Recent research has developed methods for evaluating visualization literacy (Boy et al. 2014; Börner et al. 2016; Maltese et al. 2015; Lee et al. 2017). Each of these evaluations include a collection of questions for assessing literacy levels. Since the primary purpose of this study is not to evaluate visualization literacy, I utilize test items for stacked bar charts from Lee et al. (2017) to devise one question. Evaluating the literacy of the stacked bar style aligns with the charts implemented in this study to visualize counts of shelter animals.

Procedure This study utilized crowdsourced participation through Amazon Mechanical Turk. Four Human Intelligence Tasks (HITs) were created on the Turk platform, one for each treatment. The tasks and associated survey questions are identical across HITs; the only variation is the chart treatment. Participants were only enabled to complete one of the four HITs one time. Two qualifications were implemented to filter the potential participant pool: participants must live in U.S. and have a HIT acceptance rate of at least 60 EVALUATING THE EMOTIONAL EFFECT OF PROXIMITY TECHNIQUES

95%. Those that qualified and accepted the HIT were directed to an external webpage with the survey.

Figure 4.10 The survey begins with a consent statement which is required to accept Stages of the survey. before continuing through the survey. Once consent is given, participants were taken through six stages of the survey. The first stage consisted of three demographic questions regarding age, gender, and education level. Stage two measured data literacy, where participants were given an image of a stacked bar chart and asked about the information in the chart. Stage three included a topic introduction and two pre-treatment measurements: initial attitude and involvement. In stage four, participants were instructed to visit an external link that has one of the four interactive charts. After exploring the chart, participants enter stage five: a high-level question to check that the participant paid attention to the chart. Lastly, stage six

included the post-treatment questions: emotional response, post-treatment attitude measurement, and a free-response question regarding why the participant’s attitude did or did not change. The survey concluded with a disclosure statement about the full purpose of the survey.

50 participants were recruited for each treatment, resulting in 200 participants in the study. The study expected to take approximately 5-10 minutes to complete, and each participant was compensated $0.50 for their time.

ANALYSIS & RESULTS

The results of those who answered the data literacy and attention check questions incorrectly were omitted because these questions had one EVALUATING THE EMOTIONAL EFFECT OF PROXIMITY TECHNIQUES 61

correct answer. The goal of omitting these results is to filter out participants who may not be able to read the chart being tested or were potentially randomly selecting responses. The resulting number of participants after omissions range from 47 to 49 per treatment, totaling to 191 participants.

Analysis of the distribution of time it took to complete the survey found that the median duration was 4.76 minutes, with a quarter of participants completing the survey in less than 3.46 minutes. These durations include reviewing the consent statement and disclosure statement. While many participants completed the survey faster than expected, especially with the first quarter of participants, none were omitted for duration.

Demographics and Involvement Figure 4.11 shows the distribution of age, gender, and education across the Figure 4.11 four treatments. The majority of participants in each treatment (62-71%) fell Distributions for age, within the age range 25-44. Three treatments had slightly more males than gender, and education.

GENDER

CONTROL male female GENDER INTERESTS other

TEMPORAL

SPATIAL

AGE

CONTROL 18-24 25-34

INTERESTS AGE 35-44

45-54 TEMPORAL 55-64

SPATIAL 65+

EDUCATION

CONTROL some high school

EDUCATION high school INTERESTS some college

associate’s degree TEMPORAL master’s degree

SPATIAL doctorate 62 EVALUATING THE EMOTIONAL EFFECT OF PROXIMITY TECHNIQUES

females. The temporal treatment, however, had twice as many females as males. The top two most frequently occurring highest levels of education received, bachelor’s degree and some college, was consistent across the four treatments. The interest group is the most educated group, with 60% of participants receiving a bachelor’s degree or higher. The control group had the most participants acquiring less than a college degree, resulting in 46% of the group.

Figure 4.12 Potential responses for involvement ranged from enjoying the company of Distribution of involvement dogs and/or cats to being involved with an animal shelter or animal welfare responses by the 191 study organization. Participants could select multiple levels of involvement; they participants. also had the option of selecting ‘none of the above’. Responses indicate that there’s a high level of involvement within the total participant pool: 90% enjoy the company of dogs and/or cats, and 82% have had a dog or cat. More than half (57%) indicated that they have adopted a dog or cat before and 18% have either worked at, volunteered with, or donated to an animal welfare organization before. Interestingly, one participant identified as a frequent

I enjoy the company of dogs and/or cats 90%

I have/had dogs and/or cats 82%

I have adopted a dog or cat before 57%

I have volunteered at, worked at, 18% or donated to an animal shelter or animal welfare organization before

dog foster and another identified as a dog trainer. 5% of the total population did not identify with any of these categories. The distribution of involvement across the 191 participants in the study is found in Figure 4.12. Across the four treatments, the distribution of involvement appears roughly equivalent. The largest difference occurs in the temporal proximity treatment, where 65% of participants have adopted before, which is 9% more than the next closest treatment.

A population of interest are those that identified with the first two levels of involvement (enjoyed and owned a dog or cat) but not the second two (adopted a dog or cat and involved with any animal welfare organization to some degree). 24% of participants fit this description; per treatment, the temporal group had the smallest proportion (16%) fitting this description, and the spatial group had the largest (32%). EVALUATING THE EMOTIONAL EFFECT OF PROXIMITY TECHNIQUES 63

Attitudes and Changes Initial attitudes from the 7-step Likert scale were categorized into three groups: negative, weak, and positive. The majority of responses across the four treatments are positive (agree and strongly agree), from 67% to 83%. Attitudes were measured once more after exposure to the chart. The change in attitude was then calculated. For the control and temporal treatments, the changes in attitude range from -1 to 2. The spatial treatment experienced the same range, but one step back: -2 to 1. The interest treatment experienced the largest range in changes, from -6 to 2. Without one outlier, the range is -3 to 2. Qualitative responses indicated that 20 participants who recorded a change in attitude through the Likert scale (50% of the total recorded) did not think their attitude changed.

Since it is far easier to form a new belief than change an existing belief (Hoeken 2001), statistical analysis of attitude change focused on those with an initial weak attitude. Despite the small sample size, I ran a one-sided t-test for each treatment where the initial attitude was weak to evaluate if the mean change in attitude is greater than zero. Only the results for the temporal treatment were significant with a p-value of .01. Compared to the control treatment, none of the other three treatments had a significantly different mean of attitude change for those with initially weak attitudes.

Emotion Participants rated how strong each emotion was a part of their reaction to the chart on a scale of 1 to 5. If the emotion was not a part of their reaction, participants had the option of selecting ‘N/A’. Therefore, I first looked at what percentage of participants felt each emotion to any extent. 15 of the 20 emotions were felt at some capacity by over half of the total participants. Sadness, Compassion, and Disappointment were the emotions most frequently felt, being rated by at least 87% of participants. For the remaining five emotions, over half of the participants selected ‘N/A’. Amusement was least felt among participants, followed by Pride, Joy, Pleasure, and Admiration. In the valence-control space, all five emotions are positioned as positive valence, and four of the five are positioned as low control. Due to the response rate for these emotions, any significance that came out of their statistical testing was held to higher scrutiny. Consistent rating of emotions across treatments was verified; for each treatment, the average emotion was rated by 59-64% of participants.

The next step focused on testing if emotions were felt more strongly in treatments with proximity techniques than those in the control treatment. This analysis only included results where the emotion was rated on the scale of 1-5. Figure 4.14 illustrates the frequency of each rating by emotion and treatment. In the analysis, the scale was treated as interval. Therefore, the two-sample, one-sided t-test was selected to compare the mean rating for a given emotion from a proximity treatment against that of the control 64 EVALUATING THE EMOTIONAL EFFECT OF PROXIMITY TECHNIQUES

ANGER

HATE

CONTEMPT

DISGUST

FEAR

DISAPPOINTMENT

SHAME

REGRET

GUILT

SADNESS

COMPASSION

RELIEF

ADMIRATION

LOVE

CONTENTMENT

PLEASURE

JOY

PRIDE

AMUSEMENT

INTEREST

Figure 4.13 treatment. For example, when testing the rating results of Anger from the Proportion of participants temporal group against the control group, the null hypothesis is that the who felt each emotion to difference in mean rating between the temporal and control group is zero. any extent. The alternative hypothesis is that the difference in mean rating is greater than zero, meaning that Anger is felt more strongly in the temporal group than the control group. A significance threshold was set at p = .05, therefore any p-value lower than .05 indicated a significant difference in means.

Statistical significance does not necessarily directly connect to the stand- alone emotional effect, since this test gives significance to a greater difference in means, using the control treatment as the baseline. For instance, an emotion with a mean rating of 3 and a difference of 2 from the control group is more significant than another emotion with a mean rating of 4 and a difference of 1. In this way, I evaluate the emotive impact of the proximity technique in question, rather than the impact of the graphic as a whole. 6.Disappointment 11.Compassion 16.Pleasure 1.Anger 7.Shame 12.Relief 2.Hate 17.Joy 13.Admiration 3.Contempt 8.Regret 18.Pride 19.Amusement 4.Disgust 14.Love 9.Guilt

EVALUATING THE EMOTIONAL EFFECT OF PROXIMITY TECHNIQUES 65 15.Contentment 10.Sadness 20.Interest 5.Fear

This test ran for all 20 emotions in the three tested treatments against Figure 4.14 the control group, resulting in 60 t-tests. The t-test is a parametric test, Frequency of emotion ratings meaning that it assumes that a normal, continuous distribution from the across treatments. data. Figure 4.14 indicates that most emotion ratings do not follow a normal distribution. However, research shows that a sufficient sample size increases the robustness of t-tests to depart from normal distributions

(Lumley et al. 2002). As a sanity check, the same hypotheses were frequency freq evaluated with the Wilcoxon rank sum test, which is the non-parametric 5 10 15 counterpart to the t-test. 5 10 15

ANGER HATE CONTEMPT DISGUST FEAR 1.Anger 2.Hate 3.Contempt 4.Disgust 5.Fear

CONTROL 1.Anger 2.Hate 3.Contempt 4.Disgust 5.Fear

INTERESTS 1.Anger 2.Hate 3.Contempt 4.Disgust 5.Fear

TEMPORAL 1.Anger 2.Hate 3.Contempt 4.Disgust 5.Fear

SPATIAL

6.Disappointment 7.Shame 8.Regret 9.Guilt 10.Sadness

DISAPPOINTMENT SHAME REGRET GUILT SADNESS 6.Disappointment 7.Shame 8.Regret 9.Guilt 10.Sadness

CONTROL 6.Disappointment 7.Shame 8.Regret 9.Guilt 10.Sadness

INTERESTS 6.Disappointment 7.Shame 8.Regret 9.Guilt 10.Sadness freq TEMPORAL 15 freq SPATIAL 11.Compassion 12.Relief 13.Admiration 14.Love 15.Contentment 10 freq 15 5 11.Compassion 12.Relief 13.Admiration 14.Love 15.Contentment 10 freq 15 COMPASSION RELIEF ADMIRATION LOVE CONTENTMENT 5 11.Compassion 12.Relief 13.Admiration 14.Love 15.Contentment 10 15

5 CONTROL 11.Compassion 12.Relief 13.Admiration 14.Love 15.Contentment 10

INTERESTS 5

TEMPORAL 16.Pleasure 17.Joy 18.Pride 19.Amusement 20.Interest

SPATIAL 16.Pleasure 17.Joy 18.Pride 19.Amusement 20.Interest

16.Pleasure 17.Joy 18.Pride 19.Amusement 20.Interest

PLEASURE JOY PRIDE AMUSEMENT INTEREST 16.Pleasure 17.Joy 18.Pride 19.Amusement 20.Interest

CONTROL

INTERESTS

TEMPORAL

SPATIAL

1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5

RATING 66 EVALUATING THE EMOTIONAL EFFECT OF PROXIMITY TECHNIQUES

DISGUST p = .004 FEAR p = .014 DISAPPOINTMENT p = .044 INTEREST p = .025

CONTROL

TEMPORAL

INTEREST p = .002

CONTROL

INTERESTS

INTEREST p = .038

CONTROL

SPATIAL

Figure 4.15 Under a significance level of p = .05, the strength of Interest felt among The emotions and participants in all three tested groups were statistically significant compared treatments with a to the control group. This means that, for the Interest emotion, the null significant difference from hypothesis is rejected in favor of the alternative hypothesis: the mean rating the control ratings. for Interest in the three tested groups is greater than the mean rating for Interest in the control group. This was the only emotion that resulted in statistically significant results within the spatial and interest treatments. For the temporal treatment, three additional emotions resulted in statistically significant results: Disgust, Fear, and Disappointment. The Wilcoxon rank sum test confirmed these differences. While Compassion was rated strongly across all three treatments in question, the difference in results against the control group were not statistically significant.

Sub-Populations of Interest The difference in emotion ratings between gender of the total participant population was evaluated by running a two-sample t-test for each emotion. Results of four emotions were statistically significant under p = 0.5, meaning there is a difference in how men and women reported feeling Anger, Disappointment, Guilt, Sadness, and Compassion. When running the same test within treatments, the spatial treatment had significant difference in ratings for Anger and Sadness, while the interest treatment had a significant difference for Guilt and Sadness. While the temporal treatment had the greatest difference in gender proportion, there was no difference was found between men and women on their initial attitudes and change in attitude.

Emotion ratings for those with weak initial attitudes were compared to those with positive initial attitudes in the total population. No significant difference was found when running a t-test; its non-parametric counterpart found EVALUATING THE EMOTIONAL EFFECT OF PROXIMITY TECHNIQUES 67

a significant difference between the two populations in how they rated Compassion. Emotion ratings for the involvement population of interest were not statistically significant compared to those who have adopted before or have been involved with animal welfare organization.

DISCUSSION

This study evaluated the emotional impact of proximity techniques in visualizations that represented data about shelter animals. These proximity treatments relate the data to the user through proximity of space, time, and interests. By comparing the emotion ratings from one of the proximity treatments to those from the control treatment, I isolate the impact of the technique itself. Results found that the temporal proximity technique had the most emotional impact, with four statistically significant emotions: Disgust, Fear, Disappointment, and Interest. The strength of the Interest emotion was also increased by the other proximity techniques, but this was the only emotion found to have significant difference for spatial and interest proximity compared to the control treatment.

These results validate Campbell’s assertion that the future is more affecting than the past. They did not validate, however, his assertion that the connection of place is more affecting than the relevance of time. The source of this difference, and the general lack of significant results with spatial and interest proximity, may lie in the implementation of these techniques within the study.

Treatments for interest and spatial proximity enabled the user to select the subset of the data that would be visualized. Because this resulted in a true subset of the data, participants were not exposed to the same data and therefore not the same ratio for live to non-live outcomes. For example, the control and temporal treatments present a 5:1 ratio. In the interest treatment, 64% of participants looked at the dog subset, which presents a 7:1 ratio. The ratio for cats, on the other hand, is 4:1. In the spatial treatment, data for 22 states were reviewed. Two of those states present an impressive ratio of 24:1. This means that for every 24 dogs and cats that experienced a live outcome, one did not. The lowest ratio a participant saw from a state was 2:1.

Given the variability of subsets that the participants were exposed to in these two treatments, it is highly probable that different messages were interpreted from the data. Without holding the data constant, the capacity for a direct comparison to the control treatment is limited. This is a hurdle with testing these techniques. To ensure a consistent message is delivered, future research should consider synthetically deriving the subsets or selecting a dataset where the message aligns with the message presented by the subsets. 68 EVALUATING THE EMOTIONAL EFFECT OF PROXIMITY TECHNIQUES

Another difference in the implementation of these techniques within the study comes from the limitations of the data source. I implemented interest and spatial proximity within the restrictions of the data in an effort to maintain an ethical design and an authentic experience for the participants. Because the data source at the time did not offer potential to implement temporal proximity, I incorporated this technique without restrictions. Future research should consider consistency across the treatments in regards to following data limitations or disregarding them across the board.

Another future consideration when testing these techniques involves topic and attitude change. This topic proved ill-suited for measuring attitude change since the majority of initial attitudes agreed or strongly agreed that an increase in adoptions will help decrease the number euthanized. While results indicated a positive increase in attitude change among those with initial weak attitudes in the temporal treatment, the sample size for this population is too small to make any conclusions about the greater population. If attitude change is considered in future emotion evaluations, a topic that initially produces more neutral or weak attitudes should be prioritized.

Despite these future considerations for spatial proximity, interest proximity, and attitude change, I believe the work here successfully identified an increased emotional impact with the use of temporal proximity in visualizations. This study lends itself as initial experimentation for future studies on the emotional impact of proximity techniques or other techniques in visualizations. My hope is that this work helps encourage future work on the matter so that we may better understand how visualizing data can appeal to our emotions. EVALUATING THE EMOTIONAL EFFECT OF PROXIMITY TECHNIQUES 69 70 EVALUATING THE EMOTIONAL EFFECT OF PROXIMITY TECHNIQUES EVALUATING THE EMOTIONAL EFFECT OF PROXIMITY TECHNIQUES 71

To make data relevant, it not only needs to be visual, it needs to be affecting.

Steve Lambert

Conclusion

There has been an increased use of emotions and data together, whether it is verbal or written through discussions in the visualization community or visually through efforts to appeal to emotions. In light of this, I present this thesis, which contributes research on the use of pathos in data visualization. I first present a taxonomy of pathos techniques, which serves as a foundation for practitioners build upon and expand. Secondly, I contribute an empirical evaluation of the emotional impact of proximity techniques in visualizations.

The evaluation tested proximity of time, space, and interests in visualized Figure 5.1 data regarding shelter animals. The results of the study indicated that My digital recreation of temporal proximity has the strongest emotive impact of the proximity Lupi’s phone addiction data techniques tested. The emotions Interest, Disgust, Fear, and Disappointment drawing from the Dear Data were felt significantly stronger with the inclusion of temporal proximity in project. the chart. Interest was the emotion most infuenced by the use of proximity techniques, as it was felt significantly stronger in each proximity treatment.

These findings indicate that the framing of data matters, that people feel greater interest towards a topic when the visualized data are more relevant to them and that data representing events closer in time are more affecting. These techniques take steps towards more human-centric designs, where considerations towards the audience are emphasized, and validate that design decisions in the data layer of the visualization can increase emotive impact. It can be argued that framing the data in such a way leaves data 74 CONCLUSION

points out, thus not telling the full story of the dataset. However, when the framing increases relevance to the viewer, thus increasing its impact on the audience, while maintaining a truthful connection to the underlying data, the framing is justified.

As with all techniques meant to influence, these proximity techniques can be subjected to unethical use. For instance, if real-world visualizations falsely frame data as more temporally relevant than they actually are, they are trading honesty for affect, thus deceiving the user. Therefore, as we begin to understand the emotional impact certain design decisions have, we must be equally vigilant in identifying their use for deceiving.

This research takes a step towards improving our understanding on the use and impact of techniques that appeal to emotions in visualizations. With an enhanced understanding and further validation, visualization practitioners that make an effort to connect feeling to data will have validated techniques to do so. The validation of pathos techniques leads to established uses and critical approaches, so that the cloudy space that is emotional appeals in visualization today can be clarified, evaluated, and reflected upon.

In order to reconsider the value of pathos in data design, one does not have to, and should not, de-emphasize the role of logos and ethos. It is the combination of the three persuasive modes together that create powerful, effective messages. Although emotion is commonly seen as the adversary of reason, the two work better together than apart; as Campbell said, “passion is the mover to action, reason is the guide” (1776). If our community gives pathos proper consideration and attention, we have the potential to improve our communicative ability with data, to add the human element behind the logic, to simultaneaously inform and affect. CONCLUSION 75

References

Aristotle. 2004. Rhetoric. Translated by W. Rhys 139–45. https://doi.org/10.2307/355048. Roberts. Dover Thrift Eds edition. Mineola, N.Y: Dover Publications. Borgo, R., A. Abdul-Rahman, F. Mohamed, P. W. Grant, I. Reppa, L. Floridi, and M. Chen. 2012. “An Bassan, Andrea, Marco Bernardi, Paolo Corti, Giorgia Empirical Study on Using Visual Embellishments in Lupi, Giovanni Magni, Valerio Pellegrini, Simone Visualization.” IEEE Transactions on Visualization and Quadri, Gabriele Rossi, and Marco Vettorello. 2017. Computer Graphics 18 (12): 2759–68. https://doi. “Items: Is Fashion Modern?” Accurat. 2017. http:// org/10.1109/TVCG.2012.197. www.accurat.it. Borkin, M. A., Z. Bylinskii, N. W. Kim, C. M. Bainbridge, Bateman, Scott, Regan L. Mandryk, Carl Gutwin, C. S. Yeh, D. Borkin, H. Pfister, and A. Oliva. 2016. Aaron Genest, David McDine, and Christopher “Beyond Memorability: Visualization Recognition Brooks. 2010. “Useful Junk?: The Effects of Visual and Recall.” IEEE Transactions on Visualization and Embellishment on Comprehension and Memorability Computer Graphics 22 (1): 519–28. https://doi. of Charts.” In Proceedings of the SIGCHI Conference org/10.1109/TVCG.2015.2467732. on Human Factors in Computing Systems, 2573– 2582. CHI ’10. New York, NY, USA: ACM. https://doi. Borkin, M. A., A. A. Vo, Z. Bylinskii, P. Isola, S. org/10.1145/1753326.1753716. Sunkavalli, A. Oliva, and H. Pfister. 2013. “What Makes a Visualization Memorable?” IEEE Transactions Bloom, Paul. 2014. “Against Empathy.” Text. Boston on Visualization and Computer Graphics 19 (12): Review. https://bostonreview.net/forum/paul-bloom- 2306–15. https://doi.org/10.1109/TVCG.2013.234. against-empathy. Börner, Katy, Adam Maltese, Russell Nelson Balliet, Booth, Wayne C. 1963. “The Rhetorical Stance.” and Joe Heimlich. 2016. “Investigating Aspects of College Composition and Communication 14 (3): Data Visualization Literacy Using 20 Information 78 REFERENCES

Visualizations and 273 Science Museum Visitors.” The New York Times. 2009. http://www.nytimes. Information Visualization 15 (3): 198–213. https:// com/interactive/2009/11/06/business/economy/ doi.org/10.1177/1473871615594652. unemployment-lines.html.

Boy, J., R. A. Rensink, E. Bertini, and J. D. Fekete. Dance, Gabriel, Aron Pilhofer, Y. Lehren, and Jeff 2014. “A Principled Way of Assessing Visualization Damens. 2010. “Faces of the Dead: Service Members Literacy.” IEEE Transactions on Visualization and Killed in Iraq and Afghanistan.” The New York Computer Graphics 20 (12): 1963–72. https://doi. Times, March 18, 2010. https://www.nytimes.com/ org/10.1109/TVCG.2014.2346984. interactive/us/faces-of-the-dead.html.

Boy, Jeremy, Anshul Pandey, John Emerson, Darwin, Charles. 1899. The Expression of Emotion in Margaret Satterthwaite, Oded Nov, and Enrico Man and Animals. Champaign, Ill: Project Gutenberg. Bertini. 2017. “Showing People Behind Data: Does http://ezproxy.neu.edu/login?url=http://search. Anthropomorphizing Visualizations Elicit More ebscohost.com/login.aspx?direct=true&db=nlebk&AN Empathy for Human Rights Data?” In . https://doi. =1049600&site=ehost-live&scope=site. org/10.1145/3025453.3025512. “Domestic Data Streamers.” 2013. Behance. 2013. Brinton, Willard C. 1914. Graphic Methods for https://www.behance.net/gallery/11195865/YES-vs- Presenting Facts. New York: The Engineering NO-Data-Visualizer. Magazine Company. Dörk, Marian, Patrick Feng, Christopher Collins, and Buchanan, Richard. 1985. “Declaration by Design: Sheelagh Carpendale. 2013. “Critical InfoVis: Exploring Rhetoric, Argument, and Demonstration in Design the Politics of Visualization.” In CHI ’13 Extended Practice.” Design Issues 2 (1): 4–22. https://doi. Abstracts on Human Factors in Computing Systems, org/10.2307/1511524. 2189–2198. CHI EA ’13. New York, NY, USA: ACM. https://doi.org/10.1145/2468356.2468739. Cairo, Alberto. 2013. “Emotional Data Visualization: Periscopic’s ‘U.S. Gun Deaths’ and the Challenge of Epstein, Seymour. 1994. “Integration of the Cognitive Uncertainty | | Peachpit.” April 3, 2013. http://www. and the Psychodynamic Unconscious.” American peachpit.com/articles/article.aspx?p=2036558. Psychologist 49 (8): 709–24. http://dx.doi.org.ezproxy. neu.edu/10.1037/0003-066X.49.8.709. Cairo, Alberto. 2016. “Empathy and Visualization.” 2016. http://www.thefunctionalart.com/2016/02/ “Gapminder Tools.” 2016. Accessed March 9, 2018. empathy-and-visualization.html. https://www.gapminder.org/tools/#_state_time_ delay:100;;&chart-type=bubbles. Campbell, George. 1776. The Philosophy of Rhetoric. London,. http://hdl.handle.net/2027/ Goo, Sara Kehaulani. 2015. “The Art and Science njp.32101061812242. of the Scatterplot.” Pew Research Center (blog). September 16, 2015. http://www.pewresearch.org/ Card, Stuart K., Jock D. Mackinlay, and Ben fact-tank/2015/09/16/the-art-and-science-of-the- Shneiderman. 1999. Readings in Information scatterplot/. Visualization: Using Vision to Think. San Francisco, California: Morgan Kaufmann. Harley, J. B. 1991. “Can There Be a Cartographic Ethics?” Cartographic Perspectives, no. 10 (June): Cox, Amanda, Shan Carter, and Kevin Quealy. 9–16. 2009. “The Jobless Rate for People Like You.” REFERENCES 79

Haroz, Steve, Robert Kosara, and Steven L. Kogut, Tehila, and Ilana Ritov. 2005. “The Singularity Franconeri. 2015. “ISOTYPE Visualization: Working Effect of Identified Victims in Separate and Joint Memory, Performance, and Engagement with Evaluations.” Organizational Behavior and Human Pictographs.” In Proceedings of the 33rd Annual Decision Processes 97 (2): 106–16. https://doi. ACM Conference on Human Factors in Computing org/10.1016/j.obhdp.2005.02.003. Systems, 1191–1200. CHI ’15. New York, NY, USA: ACM. https://doi.org/10.1145/2702123.2702275. Kostelnick, Charles. 2008. “The Visual Rhetoric of Data Displays: The Conundrum of Clarity*.” IEEE Harris, Jacob. 2015. “Connecting with the Dots.” Transactions on Professional Communication 51 (1): 2015. https://source.opennews.org/articles/ 116–30. https://doi.org/10.1109/TPC.2007.914869. connecting-dots/. Kostelnick, Charles. 2016. “The Re-Emergence of Hoeken, Hans. 2001. “Anecdotal, Statistical, and Emotional Appeals in Interactive Data Visualization.” Causal Evidence: Their Perceived and Actual Technical Communication 63 (2): 116–35. Persuasiveness.” Argumentation 15 (4): 425–37. https://doi.org/10.1023/A:1012075630523. Lambert, Steve. 2016. “And What Do I Do Now? Using Data Visualization for Social Change.” Center for Holmes, Nigel. 1984. Designer’s Guide to Creating Artistic Activism (blog). January 23, 2016. https:// Charts and Diagrams. 2nd Edition edition. New York: artisticactivism.org/2016/01/data-visualization-for- Watson-Guptill. what/.

Huff, Darrell. 1954. How to Lie with Statistics. New Lee, S., S. H. Kim, and B. C. Kwon. 2017. “VLAT: York: Norton. Development of a Visualization Literacy Assessment Test.” IEEE Transactions on Visualization and Hullman, J., and N. Diakopoulos. 2011. “Visualization Computer Graphics 23 (1): 551–60. https://doi. Rhetoric: Framing Effects in Narrative Visualization.” org/10.1109/TVCG.2016.2598920. IEEE Transactions on Visualization and Computer Graphics 17 (12): 2231–40. https://doi.org/10.1109/ Lee, Saerom, Karen Page Winterich, and William T. TVCG.2011.255. Ross Jr. 2014. “I’m Moral, but I Won’t Help You: The Distinct Roles of Empathy and Justice in Donations.” Johnson, B. T., and A. H. Eagly. 1989. “Effects of Journal of Consumer Research 41 (3): 678–96. Involvement on Persuasion: A Meta-Analysis.” Psychological Bulletin 106 (Sep 89): 290–314. Lifton, R. J. 1967. Death in Life: Survivors of Hiroshima. New York: Random House. Jones, G. E. 2011. How to Lie with Charts. LaPuerta Books and Media. Lumley, Thomas, Paula Diehr, Scott Emerson, and Kent, Alexander. 2016. “Political Cartography: From Lu Chen. 2002. “The Importance of the Normality Bertin to Brexit.” The Cartographic Journal 53 (3): Assumption in Large Public Health Data Sets.” Annual 199–201. https://doi.org/10.1080/00087041.2016. Review of Public Health 23 (1): 151. 1219059. Lupi, Giorgia. 2017. “Data Humanism, the Revolution Klein, Scott. 2013. Guides: ProPublica’s News App Will Be Visualized.” Giorgia Lupi (blog). February and Data Style Guides. HTML. ProPublica. https:// 1, 2017. https://medium.com/@giorgialupi/ github.com/propublica/guides. data-humanism-the-revolution-will-be-visualized- 31486a30dbfb. 80 REFERENCES

Lupi, Giorgia, Stefanie Posavec, and Maria Popova. On, Josh, and Amy Balkin. 2005. “Exxon Secrets.” 2016. Dear Data. New York: Princeton Architectural 2005. http://www.exxonsecrets.org/. Press. Pandey, A. V., A. Manivannan, O. Nov, M. Satterthwaite, Maltese, Adam V., Dubravka Svetina, and Joseph and E. Bertini. 2014. “The Persuasive Power of Data A. Harsh. 2015. “Data Visualization Literacy: Visualization.” IEEE Transactions on Visualization and Investigating Data Interpretation Along the Novice- Computer Graphics 20 (12): 2211–20. https://doi. Expert Continuum.” Journal of College Science org/10.1109/TVCG.2014.2346419. Teaching 45 (1): 84–90. Pandey, Anshul Vikram, Katharina Rall, Margaret L. Marcus, Amy Dockser. 2014. “Trials.” The Wall Street Satterthwaite, Oded Nov, and Enrico Bertini. 2015. Journal. 2014. http://on.wsj.com/1due624. “How Deceptive Are Deceptive Visualizations?: An Empirical Analysis of Common Distortion Techniques.” Monmonier, Mark S. 1996. How to Lie with Maps. In Proceedings of the 33rd Annual ACM Conference Chicago: University of Chicago Press. on Human Factors in Computing Systems, 1469– 1478. CHI ’15. New York, NY, USA: ACM. https://doi. Muehlenhaus, Ian. 2012. “If Looks Could Kill: The org/10.1145/2702123.2702608. Impact of Different Rhetorical Styles on Persuasive Geocommunication.” The Cartographic Journal 49: Periscopic. 2013. “United States Gun Death Data 361–375. Visualization by Periscopic.” 2013. http://guns. periscopic.com. Muehlenhaus, Ian. 2013. “The Design and Composition of Persuasive Maps.” Cartography and Pitch Interactive. 2013. “Out of Sight, Out of Mind.” Geographic Information Science 40 (5): 401–14. Pitch Interactive. 2013. http://drones.pitchinteractive. https://doi.org/10.1080/15230406.2013.783450. com.

Muehlenhaus, Ian Alexander. 2010. “Lost in Randall, Tom, and Blacki Migliozzi. 2018. “Earth’s Visualization: Using Quantitative Content Relentless Warming Sets a Brutal New Record in Analysis to Identify, Measure, and Categorize 2017.” Bloomberg.Com, January 18, 2018. https:// Political Cartographic Manipulations.” Ph.D., www.bloomberg.com/graphics/hottest-year-on- United States -- Minnesota: University of record/. Minnesota. https://search-proquest-com. ezproxy.neu.edu/docview/305207782/ Rawlins, Jacob D., and Gregory D. Wilson. 2014. abstract/73A8DC04E3AD4FC2PQ/1. “Agency and Interactive Data Displays: Internet Graphics as Co-Created Rhetorical Spaces.” Technical Mushon, Zer-Aviv. 2015. “DataViz—The Communication Quarterly 23 (4): 303–22. https://doi. UnEmpathetic Art.” Responsible Data Forum (blog). org/10.1080/10572252.2014.942468. October 19, 2015. https://responsibledata.io/ dataviz-the-unempathetic-art/. Reese, Kim, Jon Schwabish, and Mushon Zer-Aviv. 2016. “Episode #31: Rees & Mushon on DataViz Neurath, Otto. 2010. From Hieroglyphs to Isotype: A Empathy.” Policy Viz (blog). February 2, 2016. https:// Visual Autobiography. Hyphen Press. policyviz.com/podcast/rees-mushon-on-dataviz- empathy/. Offenhuber, Dietmar. 2010. “Visual Anecdote.” Leonardo 43 (4): 367–74. Richards, Anne R., and Carol David. 2005. “Decorative Color as a Rhetorical Enhancement on REFERENCES 81

the World Wide Web.” Technical Communication Small, Deborah A., George Loewenstein, and Paul Quarterly 14 (1): 31–48. https://doi.org/10.1207/ Slovic. 2007. “Sympathy and Callousness: The Impact s15427625tcq1401_4. of Deliberative Thought on Donations to Identifiable and Statistical Victims.” Organizational Behavior and Robinson, Arthur H., and Barbara Bartz Petchenik. Human Decision Processes 102 (2): 143–53. https:// 1975. “The Map as a Communication System.” The doi.org/10.1016/j.obhdp.2006.01.005. Cartographic Journal 12 (1): 7–15. Springer, Julie. 2017. “The 2017-2018 APPA Rost, Lisa. 2017. A Data Point Walks Into a Bar: National Pet Owners Survey Debut.” http:// Designing Data For Empathy - Lisa Charlotte Rost. americanpetproducts.org/Uploads/MemServices/ https://www.youtube.com/watch?v=8XgF-RmNwUc. GPE2017_NPOS_Seminar.pdf.

Rougeux, Nicholas. 2016. “Off the Staff.” C82. 2016. Tactical Technology Collective. 2014. Visualising https://www.c82.net/offthestaff/. Information for Advocacy. 2nd ed. http:// visualisingadvocacy.org. Scherer, Klaus R. 2001. “Emotions, Psychological Structure Of.” In International Encyclopedia of Tufte, Edward R. 2001. The Visual Display of the Social & Behavioral Sciences, edited by Neil Quantitative Information. 2nd edition edition. Cheshire, J. Smelser and Paul B. Baltes, 4472–77. Oxford: Conn: Graphics Press. Pergamon. https://doi.org/10.1016/B0-08-043076- 7/01711-3. Tyner, Judith A. 1982. “Persuasive Cartography.” Journal of Geography 81: 140–44. Scherer, Klaus R. 2005. “What Are Emotions? And How Can They Be Measured?” Social Vande Moere, A., M. Tomitsch, C. Wimmer, B. Science Information 44 (4): 695–729. https://doi. Christoph, and T. Grechenig. 2012. “Evaluating org/10.1177/0539018405058216. the Effect of Style in Information Visualization.” IEEE Transactions on Visualization and Computer Segel, E., and J. Heer. 2010. “Narrative Visualization: Graphics 18 (12): 2739–48. https://doi.org/10.1109/ Telling Stories with Data.” IEEE Transactions on TVCG.2012.221. Visualization and Computer Graphics 16 (6): 1139– 48. https://doi.org/10.1109/TVCG.2010.179. Västfjäll, Daniel, Paul Slovic, Marcus Mayorga, and Ellen Peters. 2014. “Compassion Fade: Affect and Shelter Animals Count. 2017. “Shelter Animals Charity Are Greatest for a Single Child in Need.” PLoS Count - The National Database Project.” Shelter ONE 9 (6): e100115. Animals Count. Accessed November 20, 2017. https://www.shelteranimalscount.org. Viegas, F. B., and M. Wattenberg. 2007. “Artistic Data Visualization: Beyond Visual Analytics.” LECTURE Slobin, Sarah. 2014. “What If the Data Visualization NOTES IN COMPUTER SCIENCE, no. 4564: 182–91. Is Actually People?” 2014. https://source.opennews. org/articles/what-if-data-visualization-actually- Wundt, Wilhelm Max. 1969. Principles of Physiological people/. Psychology. London: S. Sonnenschein and Co.

Slovic, Paul. 2007. “‘If I Look at the Mass I Will Never Yau, Nathan. 2013. Data Points: Visualization That Act’: Psychic Numbing and Genocide.” Judgment Means Something. Somerset, UNITED STATES: Wiley. and Decision Making 2 (2): 79–95. http://ebookcentral.proquest.com/lib/northeastern- ebooks/detail.action?docID=1158630.

Appendix

Amazon Mechanical Turk Task Instructions Northeastern University, Department of Art + Design Investigators: Sarah Campbell and Dietmar Offenhuber

We invite you to participate in a survey that involves collecting attitudes towards a specific topic. The topic will be revealed in the first page of the survey. It is possible that the topic may stir up emotions for you, based on your experiences and connection to the topic. The survey is administered through Qualtrics. Please open the survey link in a new browser tab or window.

This survey is estimated to take 5-10 minutes to complete the 10 questions. Some of these questions involve examining charts. You will first be presented with a consent statement, and following the questions is a debriefing statement. The estimated time for completion does not include the time it will take to read the consent and debriefing statements. At the end, you will be asked to create a code to paste in the box below. The code is crucial to verify survey completion and assignment approval.

This HIT is one of four HITs in this research study. Do not complete this HIT more than once, and do not complete more than one of the four HITs. Payment will only result from completing one of these HITs once. You will not be compensated for multiple submissions. For your awareness, the names of the four HITs are listed below: 84 APPENDIX

5-10 minute Survey on Attitude Towards Topic - A 5-10 minute Survey on Attitude Towards Topic - B 5-10 minute Survey on Attitude Towards Topic - C 5-10 minute Survey on Attitude Towards Topic - D

If you have any questions about this study or comments about the survey, please feel free to contact researcher Sarah Campbell at campbell.sar@ husky.neu.edu.

Remember to paste your survey code below prior to submitting the assignment for review. Thank you!

Screenshot of task instructions on Amazon Mechanical Turk APPENDIX 85

Consent Statement Northeastern University, Department of Art + Design Investigators: Sarah Campbell and Dietmar Offenhuber

Request to Participate in Research We invite you to participate in this online survey. The survey is part of a research study that involves collecting attitudes about animal adoption. This survey will take approximately 5-10 minutes to complete. You must be at least 18 years old to take this survey.

The decision to participate in this research project is voluntary. You can refuse to answer any question and can stop the survey at any time. Compensation is contingent on the completion of the survey and submission of the HIT assignment. You will be rewarded $0.50 after approval of the assignment.

There are no foreseeable risks or discomforts to you for taking part in this study.

Your part in this study is anonymous to the researchers. While it is possible that respondents could be identified by the IP address or other electronic record associated with the response, no one involved with this survey will be capturing those data. Any reports or publications based on this research will use only group data and will not identify you or any individual as being affiliated with this project.

If you have any questions regarding electronic privacy, please feel free to contact Mark Nardone, NU’s Director of Information Security via phone at 617.373.7901, or via email at [email protected].

If you have any questions regarding your rights as a research participant, please contact Nan C. Regina, Director, Human Subject Research Protection by emailing [email protected] or calling 617.373.4588. You may call anonymously if you wish.

If you have any questions about this study, please feel free to contact researcher Sarah Campbell at [email protected].

This study has been reviewed and approved by the Northeastern University Institutional Review Board (# 18-01-04).

By clicking on the ‘Accept’ button below you are indicating that you consent to participate in this study. Please print out a copy of this consent form for your records.

Thank you for your time. 86 APPENDIX

Survey Screenshot of the 1. What category below includes your age today? consent statement at the beginning of • 18 – 24 the survey. • 25 – 34 • 35 – 44 • 45 – 54 • 55 – 64 • 65 or older

2. Which gender below do you identify with?

• Male • Female • Other (please specify): ______

3. What is the highest level of education you have received?

• Some high school • High school diploma or equivalent • Some college APPENDIX 87

• Associate’s degree • Bachelor’s degree • Master’s degree • Doctorate degree

Chart utilized in study to evaluate data visualization literacy of stacked bar charts.

4. Based on the chart above, which bar has the largest total value? (chart sourced from http://datavizproject.com/)

• A • B • C • D

An estimated 48% of U.S. households own a dog and 38% own a cat. Of those households, 34% of dog owners and 44% of cat owners adopted their pet from an animal shelter, humane society, or rescue group. Approximately 7.6 million animals every year enter animal shelters nationwide. (Sources: APPA National Pet Owners Survey, ASPCA)

5. To what extent do you agree or disagree with the following statement: An increase in animal adoptions will help decrease the number of animals euthanized in shelters.

(Animal euthanasia is the act of intentionally ending an animal’s life, ideally done in a painless, humane way)

• Strongly agree • Agree 88 APPENDIX

• Somewhat agree • Neither agree nor disagree • Somewhat disagree • Disagree • Strongly disagree

6. Select all of the following statements that are representative of you.

• I enjoy the company of dogs and/or cats • I have/had dogs and/or cats • I have adopted a dog or cat before • I have volunteered at, worked at, or donated to an animal shelter or animal welfare organization before • Other (please specify): ______• None of the above

Please open the following link in a new browser tab: [chart link goes here]. This link contains a chart. Please take a minute and explore the contents of the chart.

7. Which of the following statements describes the data in the chart?

• Common sources for obtaining dogs and/or cats as pets, e.g. from a breeder • Counts of shelter intake and outcome categories for dogs and/or cats • Geographic locations of animal shelters and other animal welfare organizations • Counts of exotic animal species that are surrendered to animal shelters

8. What was your reaction to the information displayed in the chart? Rate the intensity of all the emotions you experienced below with respect to what you felt, even if the intensities are very low. For those emotions that were not at all a part of your reaction, please check ‘Not applicable’. If you experienced an emotion that is not listed here, please write it under ‘Other’.

N/A Weak Strong Anger • • • • • • Hate • • • • • • Contempt • • • • • • Disgust • • • • • • Fear • • • • • • Disappointment • • • • • • Shame • • • • • • Regret • • • • • • Guilt • • • • • • Sadness • • • • • • APPENDIX 89

Compassion • • • • • • Relief • • • • • • Admiration • • • • • • Love • • • • • • Contentment • • • • • • Pleasure • • • • • • Joy • • • • • • Pride • • • • • • Amusement • • • • • • Interest • • • • • • Other: ______• • • • • •

9. Consider once more your level of agreement or disagreement with the following statement: An increase in animal adoptions will help decrease the number of animals euthanized in shelters.

• Strongly agree • Agree • Somewhat agree • Neither agree nor disagree • Somewhat disagree • Disagree • Strongly disagree

10. Did your response to the statement above change or did it stay the same? Explain why your opinion changed or did not change after viewing the chart. 90 APPENDIX

Screenshot of survey: demographic questions.

Screenshot of survey: topic introduction and initial measurement of attitude. APPENDIX 91

Screenshot of survey: instructions for viewing chart and attention check question.

Screenshot of survey: measurement of emotional response. 92 APPENDIX

Disclosure Statement Thank you for your participation in this survey. It is greatly appreciated.

This research aims to measure the change in your attitude on animal adoption after exposure to a chart displaying animal shelter data. This research also aims to measure your emotional response to the chart.

Adjacent sentence: only The chart you explored does not actually show projected counts for included in the disclosure tomorrow; it shows counts for the average day in 2016. statement for the temporal treatment. In this survey, you were asked to explore one of four charts used in this study. Three of the four charts cater the data shown to relate to participants through proximity of time, geography, or interests. Results of this survey will be used to analyze how these charts (1) influence one’s attitude about animal adoption and (2) affect one’s emotional state. The results for each chart will be compared to one another to analyze if these proximity techniques influence attitude and emotion differently.

If you have any questions about this study, please feel free to contact researcher Sarah Campbell at [email protected].

If you have any questions regarding electronic privacy, please feel free to contact Mark Nardone, NU’s Director of Information Security via phone at 617.373.7901, or via email at [email protected].

Right: IRB approval form. If you have any questions regarding your rights as a research participant, please contact Nan C. Regina, Director, Human Subject Research Protection by emailing [email protected] or calling 617.373.4588. You may call anonymously if you wish.

Thank you! APPENDIX 93