PERCEPTION IS REALITY: MEASURING THE EFFICACY 
OF HUMAN-CENTERED DATA VISUALIZATION 
METHODS WITH COVID-19 CASE DATA 
 
 
 
 
 
 
 
 
by 
STEPHANIE SCHOFIELD 
 
 
 
 
 
 
 
 
 
 
A THESIS 
 
Presented to the Department of Computer Information Science  
and the Robert D. Clark Honors College  
in partial fulfillment of the requirements for the degree of  
Bachelor of Science 
 
June 2021 
 
 
 
 
 
An Abstract of the Thesis of 
Stephanie Schofield for the degree of Bachelor of Arts 
in the Department of Arts and Sciences to be taken June 2021 
 
 
Title:  Perception is Reality: Measuring the Efficacy of Human-Centered Data 
Visualization Methods with COVID-19 Case Data 
 
 
 
                               Approved:  Professor Joseph Sventek, Ph.D.  
  Primary Thesis Advisor 
 
Data visualization is a tool used to represent the vast amount of information 
created each day. Although current literature about data visualization illustrates many 
ways to visualize data, only a few methods are used on a recurring basis. Common 
visualizations like bar charts, pie charts, and scatter plots are seen across the internet in 
news articles, scholarly works, and scientific papers. These methods, although prevalent 
and easy to produce, are not always created with the viewer in mind. They can be too 
complex, over-simplified, or misrepresent the data. This research aims to measure 
whether a human-centered approach to designing data visualizations could portray the 
information more effectively, and allow the user to understand the argument the data 
presents. More specifically, this study will measure whether a human-centered approach 
to visualizing COVID-19 case data will more effectively change behavior intentions 
with regards to social-distancing. Results of this research reflect that the human-
centered approach did not produce higher rates of comprehension, visual appeal, or 
willingness to change behavior, and suggests that a more distinct and interactive method 
is necessary for meaningful differences in understanding of the information.  
ii 
 
 
 
Acknowledgements 
I would like to express my sincere gratitude to my thesis committee who have 
helped me in every step of this process--not only for their time and extreme patience, 
but for their intellectual contributions to my development as a computer scientist. 
Professor Joe Sventek’s guidance, wisdom, and support, from our initial discussions 
about data visualization to now, have inspired me throughout this project. To Professor 
Nicole Dudukovic, I thank you for bringing your unique, balancing perspective as 
someone with rigorous experience in psychology, memory, and attention. To Professor 
Juan Flores, thank you for providing me invaluable guidance throughout the writing 
process.  
I would like to thank my family for being the best support system I could ask 
for. Dad, thank you for inspiring my love of math and understanding how complex 
systems work, and sending me the article that influenced this project. Mom, thank you 
for taking time away from your own dissertation to advise me on mine. Sean, thank you 
for frequently interrupting my workflow by showing me funny Youtube videos. I would 
also like to thank all of my friends who offered words of encouragement, a hug, or a 
latte to support me while I wrote this thesis.  
Completing a thesis as an undergraduate is a difficult endeavor, but I am proud 
to say my experience working on this project was intellectually stimulating, fun, and 
rewarding, thanks to the support of each of you. Thank you for having confidence in me 
and offering your guidance through this process.  
  
  
iii 
 
 
 
 
Table of Contents 
Chapter 1: Introduction 1 
1.1 Motivation for this Research 1 
1.2 Research Questions 3 
1.3 Hypothesis 4 
1.4 Outline of Thesis 4 
Chapter 2: Literature Review 5 
2.1 What is Data Visualization, and Why Do It? 5 
2.2 Common Mistakes in the Default Design Process to Visualize Data 9 
2.3 An Argument for a Human-Centered Approach 10 
2.4 Current Methods in Human-Centered Design 11 
Chapter 3: Methodology 13 
3.1 Data Collection 13 
3.2 Creating Visualizations 17 
3.3 Designing the Survey 19 
3.3.1 Structure and Question Design 22 
3.3.2 Social Distancing and Mask Behavior Questions 23 
3.3.3 Measuring Comprehension Questions 25 
3.3.4 Visual Appeal Questions 28 
3.3.5 Demographic Questions 28 
3.4 Participant Criteria 29 
3.5 Data Analysis 30 
Chapter 4: Results 33 
4.1 Participant Demographic 34 
4.1 Overview of Entire Sample 34 
4.2 Age 37 
4.2.1 Ages 18-24 37 
4.2.2 Ages 35+ 38 
4.3 Gender 39 
4.3.1 Female Participants 39 
4.3.2 Male Participants 40 
4.4 Colorblindness 41 
iv 
 
 
 
4.4.1 Colorblind Participants 42 
4.4.2 Non-Colorblind Participants 43 
Chapter 5: Discussion and Recommendations 45 
5.1 Producing Effective Visuals 45 
5.2 Changing Public Behavior 46 
5.3 Future Research 47 
Conclusion 50 
Appendix 51 
Glossary 51 
Survey Materials 53 
Question Breakdown 53 
Recruitment Statement 66 
Survey Information Statement 67 
Charts 69 
Debriefing Statement 73 
Notice of Amendment Review and Exempt Determination 74 
References 76 
 
 
  
v 
 
 
 
List of Figures  
Figure 1.10 2 
Figure 2.10 6 
Figure 2.11 7 
Figure 3.10 14 
Figure 3.11 15 
Figure 3.12 16 
Figure 3.3.1: Human-Centered Daily Cases Visualization 20 
Figure 3.3.2: Default Daily Cases Visualization 20 
Figure 3.3.3: Human-Centered Cumulative Cases Visualization 21 
Figure 3.3.4: Default Cumulative Cases Visualization 21 
Figure 3.30 24 
Figure 3.31 24 
Figure 3.32 27 
Figure 3.33 27 
Figure 4.0: Question Legend 33 
Figure 4.11 34 
Figure SM1: Q62 53 
Figure SM2: Q63 54 
Figure SM3: Q97 54 
Figure SM4: Q130 55 
Figure SM6: Q98 57 
Figure SM7: Q131 57 
Figure SM8: Q50 / Q165 / Q122 / Q166 58 
Figure SM9: Q135 / Q136 58 
Figure SM10: Q134 / Q137 58 
Figure SM11: Q66 / Q170 / Q124 / Q173 59 
Figure SM12: Q145 / Q171 / Q146 / Q174 59 
Figure SM13: Q139 / Q172 / Q138 / Q175 60 
Figure SM16: Q71 61 
Figure SM17: Q100 62 
Figure SM18: Q142 62 
vi 
 
 
 
Figure SM19: Q52 62 
Figure SM21: Q54 63 
Figure SM23: Q56 64 
Figure SM24: Q57 65 
Figure SM25: Q58 66 
Figure SM26: Random ID 66 
Figure C1: Human-Centered Daily Cases Visualization 71 
Figure C2: Default Daily Cases Visualization 71 
Figure C3: Human-Centered Cumulative Cases Visualization 72 
Figure C4: Default Cumulative Cases Visualization 72 
Figure N1: IRB Exempt Determination 74 
Figure N2: IRB Amendment Review 75 
 
 
vii 
 
 
 
Chapter 1: Introduction 
1.1 Motivation for this Research 
Data is the most abundant resource on earth. Everything we touch, hear, see, and 
breathe is information. In fact, we create about 2.5 exabytes of data online every day: 
500 million tweets, 5 billion internet searches, and 294 billion emails are sent daily 
(Desjardins, 2019). With so much data accessible on the internet, it is critical that we 
analyze how this information is presented, and how it is perceived. Now in the face of a 
global pandemic, articles, images, and videos of the Coronavirus’s development have 
spread across the world. Anxious people are seeking answers, but some of the 
visualizations in these publications misrepresent the data. Or worse, the data lack 
sufficient scientific support to be correct. In order for us to stop the spread of this virus, 
the public needs to understand the urgency and reasoning behind wearing a mask and 
practicing social-distancing. Understandable, unambiguous graphics are critical in order 
for the public to understand these ideas.     
In April 2020, I left my job as a Resident Assistant in Eugene to fly home to 
Honolulu, Hawaii due to a pandemic-induced domestic travel ban placed on military 
dependents. At this time, very few people were wearing masks, socially-distancing, or 
practicing precautions that would signal an understanding of the seriousness of the 
virus. I wore two face masks and a shield to protect myself on the plane, but received 
many negative stares as I walked through the airport. Just before taking off, my dad sent 
me an article from the New York Times titled “This 3-D Simulation Shows Why Social 
 
 
 
Distancing Is So Important” (Parshina-Kottas et al., 2020). This article features a 
visualization of how water droplets spread around an open room after a person sneezes.  
 
Figure 1.10 
Image showing the progression of a person’s sneeze from three feet, six feet, and 
farther (Parshina-Kottas et al., 2020).  
After looking at this graphic, I understood how the virus spreads, and was 
reminded how important it is to wear a mask to protect others. Although I had read 
articles about which protocols I should follow, this visualization made it clear why I 
need to be social distancing and masking. I found it easy to connect with this type of 
interactive visual, and see the science and data in a more comprehensible format. This is 
what inspired me to think about the way these visualizations are produced, and how 
2 
 
 
effective charts could motivate people to behave a certain way. I wondered: what if a 
chart representing COVID-19 data could compel someone to socially-distance? This is 
where I began my research and the development of this project.  
1.2 Research Questions 
In order to measure if data visualizations are truly compelling, I had to ask questions 
surrounding the approach to creating those visuals. I wanted to know how visualizations 
are created, what current literature proposes about these methods, and how to improve 
upon that methodology to make more effective data visualizations. The first step in the 
research process was to conduct a literature review. The review was motivated by a few 
important questions: Why use a human-centered design approach to creating 
visualizations? Can visualizations be clear, and convincing enough to change people’s 
behavior after looking at them? What is the best way to present visualizations of data 
for maximum human-understanding? Addressing these questions in my literature 
review narrowed my focus for what I wanted answered in my study.  
I determined that there were questions to be answered about the effectiveness of 
data visualization methods, and areas of public health and data science where I could 
contribute this research. My study addressed: Which method of visualizing COVID-19 
case data will result in the highest number of responses showing a willingness to 
continue social-distancing and mask-wearing behavior, or change their behavior if not 
doing so? Once I completed secondary research and formulated my research questions, 
I had enough context to make a prediction about the results of this study.  
3 
 
 
1.3 Hypothesis 
My hypothesis is that participants who view COVID-19 case data visualizations 
designed with a human-centered approach will show the highest rate of changing their 
current social distancing and mask-wearing behavior. Additionally, I predict that 
human-centered visualizations will be preferred among participants, and be understood 
more clearly than my default charts.   
1.4 Outline of Thesis 
I start by giving context to current methods in data visualization, human-
centered design, and how data visualization affects public health. My literature review 
contains the default practices for designing visualizations, and an argument for why 
human-centered design is critical to creating effective visualizations. I will then provide 
my research methodology in finding the data sets, generating the visualizations, 
designing the survey, recruiting participants, and analyzing the data. Then I will share 
the findings of this study based on several key participant demographics, followed by a 
discussion of the results and future work in this research.  
   
4 
 
 
Chapter 2: Literature Review 
I start this chapter by introducing significant applications of data visualization, the 
motivation behind portraying data as comprehensible graphics, and then I explore 
current methods for producing visualizations. I will make an argument for why data 
visualizations need to be created with human-comprehension at the forefront of design. 
Furthermore, I will explain why compelling visualizations are so critical to public 
health.  
2.1 What is Data Visualization, and Why Do It? 
Tamara Munzner, an expert in information visualization and author of Visualization 
Analysis & Design, defines computer-based visualizations as models that provide visual 
representations of datasets designed to help people carry out tasks more effectively 
(Munzner, 2015). To be more concise, data visualization is modeling large amounts of 
data in a visual representation. Munzner argues that visualization is most necessary 
when one desires to enhance human capabilities rather than replace them with 
computation (Munzner, 2015). This means that visualizations should help people 
understand complex ideas without having to understand the computational analysis 
behind it.   
There are many areas of research that benefit from human analysis of data 
visualizations. Economists, physicians, policy makers, geologists, mathematicians, 
historians, musicians, and many other industry professionals and creatives alike utilize 
data modeling. The website visualcapitalist.com features many applications of data 
visualization across technology, money, energy, healthcare, and more. One recent post 
shows a unique visualization of the richest families in America, as seen in figure 2.10. 
5 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Figure 2.10  
“Visualized: The Richest Families in America “, represented with blocks, by Avery 
Koop (2021). 
6 
 
 
Another notable visual, shown in figure 2.11, portrays the most popular subscription-
based streaming services with slices of a pie chart. These visualizations are noteworthy  
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Figure 2.11 
“Which Streaming Service Has the Most Subscriptions” by Omri Wallach (2021). 
because they represent a large amount of information in one static image, while 
displaying them with unique components like a spiraling pie chart and “brick” pieces. 
7 
 
 
They draw the user’s attention and represent a large amount of information in one 
graphic. Visualizations like the ones shown in these figures are effective because they 
follow a set of important guidelines.     
The goal of effective data visualization is to present comprehensive depictions 
of data that help a user understand the information without having to view a raw data 
file or read long strings of text. Edward Tufte, one of the creators of human-centered 
visualization methods, writes in The Visual Display of Quantitative Information that 
“Often the most effective way to describe, explore, and summarize a set of numbers—
even a very large set—is to look at pictures of those numbers. . . of all methods for 
analyzing and communicating statistical information, well-designed data graphics are 
usually the simplest and at the same time the most powerful” (Tufte, 1985). Tufte 
makes the argument that it is often most effective to describe a data set with a visual 
representation; however, it is critical that these models meet his set of requirements to 
be high quality visualizations.  
Tufte argues that visualizations satisfying graphical excellence will induce the 
viewer to think about the data rather than the design, avoid distorting what the data 
says, present many numbers in a small space, make large data sets coherent, encourage 
the eye to compare information, serve a purpose, and be closely integrated with the 
statistical description of the data set (1985). Many statistics, psychology, and design 
articles refer to these guidelines to support the quality of their visualizations (Venables 
& Ripley, 2001; Kress & Theo Van Leeuwen, 1996; Mayer, 2002). Unfortunately, it is 
far too common that designers and data scientists do not follow these guidelines, and 
8 
 
 
the resulting visualizations are difficult to understand. This goes against the motivation 
for producing visualizations, and works against the user trying to understand them.  
Producing visualizations is not an easy task. There is a long process of finding 
the information, compiling and cleaning the data sets, using the right software and tools 
to create the visuals, and designing a graphic that will portray the information 
accurately, succinctly, and in a way that the viewers will understand it. Therefore, it is 
wasteful to produce imperceptive visualizations that do not appeal to the audience. 
Based on the default graphics offered in Excel, I assumed that these types of 
visualizations would fall into a default design category.  
2.2 Common Mistakes in the Default Design Process to Visualize Data 
In order to describe the default methods of generating data visualizations, I must 
contrast the guidelines that Tufte recommends in section 2.1, and illustrate common 
faults in the design process that produce ineffective visuals. 
Visualizations created without a methodological approach, and that therefore do 
not meet Tufte’s guidelines, are how I defined a default data visualization. I defined the 
default approach this way because it is the inverse of a human-centered design. 
Visualizations that follow Tufte’s principles are inherently human-centered because 
they incorporate the science of human perception and comprehension. If the designer 
wants to meet the goal of producing an effective graphic, then they need to think about 
the user viewing their design. Often the default approach does not include intentional 
thought about human perception, which will result in an ineffective visualization.  
There are some famous examples of confusing and unnecessary visualizations 
that highlight some of these faults. For instance, in Figure 2.20 below, the viewer can 
9 
 
 
see causes of untimely death in regards to their annual change from 2005 to 2010. 
Although Bill Gates voted this the best graph of the year in 2013 (Wonkborg), it has 
been criticized by data visualists and designers alike. There are a few reasons this chart 
is problematic. While it is eye-catching, the use of a three-dimensional block is 
unnecessary and can lead to confusion. This goes against Tufte’s outlines for graphics 
of making large data sets coherent, and doesn’t encourage the eye to compare different 
pieces of data.  
 
 
 
 
 
 
 
 
 
 
 
Figure 2.20 
Bill Gates’s graph of the year in 2013 (Wonkborg, 2013). 
2.3 An Argument for a Human-Centered Approach 
A common saying in the U.S. Air Force is “Perception is reality.” My dad, an Air 
Force fighter pilot, would say this phrase to me often as a child. Although this 
10 
 
 
expression reflects the focus necessary for American servicemen to work towards a 
common mission, it accurately represents how we can measure the effectiveness of data 
visualizations. If the person viewing the chart misunderstands the data being displayed, 
then the chart is ineffective. The viewer’s perception of the chart determines their 
comprehension of the data being portrayed. This is why a human-centered design is so 
critical to ensuring understanding of the important information being illustrated in these 
visualizations.  
2.4 Current Methods in Human-Centered Design 
Two increasingly influential leaders in the human-centered design of data visualizations 
are Edward Tufte and Tamara Munzner. Each of them provides a set of guidelines that 
have motivated my design of the human-centered visualizations, and dispute the errors 
of the default visualization.  
 Edward Tufte is a Professor Emeritus of Statistics and Computer Science at Yale 
University, and has been said to be “The Leonardo da Vinci of data” by the New York 
Times (Tufte, 1985). He proposed a set of guidelines for graphical excellence. He 
defined excellence in graphics to include displays that show the data, induce the viewer 
to think about the substance rather than design of the visualization, avoid distorting 
what the data says, present many numbers in a small amount of space, make large data 
sets coherent, encourage the eye to compare different pieces of data, highlights the data 
at several levels of detail, serve a clear purpose, and are closely integrated with the 
descriptions of the data set (Tufte, 1985). While it is difficult to maintain all of these 
rules in one visualization, I used them to guide my creation of the human-centered 
graphics.  
11 
 
 
 Tamara Munzner is a Professor of Computer Science at the University of British 
Columbia, and is an expert in information visualization. Munzner outlines the “Rules of 
Thumb” that guide the generation of effective visualizations. She asserts that these rules 
of thumb are meant to characterize which idioms are appropriate for tasks in data 
visualization design (Munzner, 2015). These guidelines are as follows: No Unjustified 
3D; No Unjustified 2D; Eyes Beat Memory; Resolution over Immersion; Overview 
First, Zoom and Filter, Detail on Demand; Responsiveness is Required; Get It Right in 
Black and White; and Function First, Form Next (Munzner, 2015). I will highlight an 
important characteristic of these guidelines that have influenced my human-centered 
design methodology.  
 Munzner argues in her section about No Unjustified 3D that designers should 
carefully avoid tilted text due to its illegibility to most readers. She writes, “text fonts 
have been very carefully designed for maximum legibility when rendered on the grid of 
pixels that makes up a 2D display. . . As soon as a text label is tilted in any way off of 
the image plane, it typically becomes blocky and jaggy” (Munzner, 2015). This is an 
interesting argument because, as I explain further in chapter three, Excel’s default 
design for many charts includes slanted text on the x-axis. When designing my human-
centered visualization, I positioned all text to be flat to avoid this misstep. Her rules of 
thumb were central to my design of the human-centered visualizations.  
12 
 
 
Chapter 3: Methodology 
In this section I describe the data sets used in this study and how I gathered them, how I 
designed the visualizations and the survey, how I selected participants and conducted 
the experiment, and how I analyzed the results. This methodology provides an overview 
of my research process from April 2020 until time of writing in April 2021.   
3.1 Data Collection 
After developing the questions that would drive this research, the next step of the 
process was to find relevant data sets to use in creation of the charts. I wanted to ask 
participants questions about their social-distancing and mask-wearing behavior, so the 
data needed to be evocative of their decision to mask and stay six feet apart from others. 
I looked for data sets related to case counts, masking, social-distancing, or any 
information connected with the pandemic.  
There were several challenges in finding a relevant and usable set. The sets 
needed to be publicly available so I would not need a license to visualize the data. They 
also needed to be in raw form so I could prepare the information for the visualization. 
Additionally, at the time of data collection in the early months of the pandemic, it was 
difficult to find publications related to COVID-19. However, I eventually found and 
investigated the provenance of a few data sets.  
In August, I found the New York Times’ “covid-19-data” repository on Github. 
This repository includes a “mask-use” dataset that contains estimates of mask usage by 
county in the United States based on results of a survey conducted by global data and 
survey firm Dynata (New York Times, 2020). Over 250,000 participants were asked, 
“How often do you wear a mask in public when you expect to be within six feet of 
13 
 
 
another person?” with the available responses as Never, Rarely, Sometimes, Frequently, 
and Always. These responses were obtained from July 2 to July 14, 2020.  
 
Figure 3.10 
“A Detailed Map of Who is Wearing Masks in the U.S.” by Josh Katz, Margot Sanger-
Katz, and Kevin Quely (2020). 
These data were freely available for public research use but showed surprising 
results. The data reflected high mask usage in places where cases were rising rapidly. 
For instance, on July 14 during the time of conducting the survey, Florida was entering 
its highest peak since the pandemic began with 9,100 new reported cases (New York 
Times, 2021), but the self-reported mask usage data reflected that the average 
probability one would encounter someone wearing a mask in Florida was 79.8% (New 
York Times, 2020). Public health experts would argue that with a large majority of 
people wearing masks, this would help reduce the spread rather than amplify it.   
14 
 
 
I computed the average probability of encountering someone wearing a mask in 
Florida by assigning 0%, 25%, 50%, 75%, and 100% to Never, Rarely, Sometimes, 
Frequently, and Always, respectively, and adding the product of the frequency of each 
response with these percentages. The raw data and the visualization resulting from 
computing the probability of encountering someone wearing a mask in each county of 
Florida can be viewed in Figure 3.11 and 3.12 below. 
 
 
Figure 3.11 
Raw data from the mask usage file with each county FIPs code, response, and 
probability computation, sourced from the New York Times mask-usage data set 
(2020), and combined with the U.S. Census Bureau’s 2018 FIPs Codes to align each 
row with the corresponding state (2020). 
 
 
15 
 
 
 
 
 
 
Figure 3.12 
The probability of encountering someone wearing a mask in Florida in July 2020. Data 
sourced from the New York Times “covid-19-data” repository (New York Times, 
2020), and visualization created using Google Charts. 
These results led me to believe that respondents were self-reporting higher mask 
usage than their true mask-wearing behavior. The visualizations I produced using these 
data portrayed mask-wearing to have little or no effect on the rising cases of COVID-19 
across the nation, so I decided to find another set that would reflect a more compelling 
argument to practice social-distancing and mask-wearing protocols.  
I decided to use the Centers for Disease Control and Prevention’s (CDC) 
“United States COVID-19 Cases and Deaths by State Over Time” case surveillance data 
(CDC Case Task Force, 2021). This set includes the state, total cases, total deaths, new 
deaths, and the day the counts were recorded for each county in the U.S. This data 
provided a solution to each of my research components: it is available for public 
16 
 
 
research use, in raw form, and addressed the effects of social-distancing and mask-
wearing.  
3.2 Creating Visualizations 
The next step after finding this set was to clean and organize the information 
into a format ready for visualization. The data file contains over 27,800,000 rows and 
15 columns of information about case rate and death count. I separated the data into two 
sets: a daily overview of cases, and a cumulative overview of cases. I then separated the 
sets by state, and generated the following two visualizations: for the daily chart, I used 
information regarding submission date and new cases, whereas for the cumulative chart 
I used submission date and total cases. For both data sets, I narrowed in on Florida as 
the state to represent. I selected Florida because it had some of the highest reported 
cases in the country, and many of the outbreaks were happening on college campuses. I 
hoped a visualization of the rapid spread would make the most compelling argument to 
social distance.  
Once I organized the data, I was prepared to generate the visualizations. I 
initially started designing the visualizations using Google Charts, as seen in Figure 3.11 
above, but ultimately used Microsoft Excel for its diversity of design elements, the 
quality of PNG exports for charts, and the default colors, fonts, and line sizes that 
provide a common representation of data visualizations. I designed four total 
visualizations to test my hypothesis: human-centered daily and cumulative 
visualizations, and default daily and cumulative visualizations. I wanted to provide two 
charts for each design method to provide additional validity for each design’s efficacy.  
17 
 
 
I designed the default charts for the daily and cumulative set first, and then used 
that design as a starting point to build the equivalent human-centered chart. For the 
default visualization, my approach was to use the default settings for most of the chart’s 
design. I used Georgia for font since it is one of the most widely used fonts on the web 
(Rawsthorn, 2006), set the line width at 1.5 point for the daily chart and 3 point for the 
cumulative chart, and left the line color as the default light green. The numbers 
representing case counts were automatically placed to the left of the y-axis, so I left 
them there as well. Excel automatically places dates at a slanted angle, so I left them in 
this position.  
My approach to designing the human-centered charts was to improve or remove 
the aspects of the default chart that have been shown to negatively impact the 
comprehension and aesthetic appeal of the visualization. Based on Tufte’s guidelines 
for graphical excellence and Munzner’s rules of thumb, I made several changes to the 
font, color, lines, text, and positioning in the visuals. I widened the lines from 1.5 to 3 
point in the daily visual, and from 3 point to 6 point in the cumulative visual. I made the 
line a deep blue so that the graph would look the same if it were converted to black and 
white. This is so that any user, including those who are colorblind, could read the lines 
on the graph, and is a strategy suggested by Munzner in “Get it Right in Black and 
White” (2015). Furthermore, I flattened the dates on the x-axis, and put the timeline 
descriptions in black boxes to separate the marking lines from the text. I eliminated the 
tilted dates on the x-axis based on one of Munzner’s Rules of Thumb that argues “Tilted 
Text Isn’t Legible” (2015).  I also placed the digits along the y-axis to put them closer 
to the most recent dates on the x-axis. Lastly, I changed the font to Arial Bold because 
18 
 
 
of a study supporting its readability on computer screens in a user survey with serif and 
sans serif typefaces (Wilson, 2001). All of these design choices were supported by 
Munzner and Tufte’s guidelines about generating effective visuals.  
3.3 Designing the Survey 
My goal for designing the survey was to ask questions that would measure the 
effectiveness of the visualization’s ability to represent the data, determine which 
visualizations were most pleasing to view, and gauge the respondent’s willingness to 
socially-distance and wear a mask before and after viewing the visualizations. I 
structured the questions around the four visuals (shown below), and organized the flow 
of the survey to present half of the participants one design of a visual with two total 
graphs, and the other half would view the alternative design with two total graphs. This 
meant that the participants would have a 50% chance of viewing the human-centered 
visuals or the default visuals. Before viewing the visualizations, they were asked 
questions about their social distancing behavior, during viewing they were asked about 
components of the data to measure their understanding, and after viewing they were 
asked about future social distancing behavior and demographic questions (see 
Appendix).  
19 
 
 
 
Figure 3.3.1: Human-Centered Daily Cases Visualization 
Daily cases of COVID-19 across the state of Florida over March to December of 2020, 
created with a Human-Centered Design methodology.  
 
Figure 3.3.2: Default Daily Cases Visualization 
Daily cases of COVID-19 across the state of Florida over March to December of 2020, 
created using a default approach. 
20 
 
 
 
Figure 3.3.3: Human-Centered Cumulative Cases Visualization 
Cumulative cases of COVID-19 across the state of Florida from March to December of 
2020, created with a Human-Centered Design methodology.  
 
Figure 3.3.4: Default Cumulative Cases Visualization 
Cumulative cases of COVID-19 across the state of Florida from March to December of 
2020, created using a default approach.  
 
21 
 
 
3.3.1 Structure and Question Design 
I decided to use Qualtrics for the survey platform for its professional design, 
data analysis and report capabilities, and familiarity with University of Oregon 
affiliates. I organized the questions in the following structure: survey information and 
consent statement, questions about social-distancing and mask behavior, identical 
questions to measure comprehension of the cumulative data for both the human-
centered and default visualizations, identical questions to measure comprehension of the 
daily data for both the human-centered and default visualizations, follow-up questions 
gauging the respondent’s future masking and social-distancing behavior, demographic 
questions, and then a debriefing form.  
I made a few key choices in constructing the response options in each question. 
For the appeal and behavior questions, I decided to offer participants a Likert scale with 
only four choices so that they must make a polar decision. This is to avoid respondents 
choosing to “sit on the fence” and select a null answer such as “Neutral” or “Unsure”. 
This allowed me to compare the responses between questions and charts, and compute 
significance of comprehension, appeal, and predictions about future behavior.  
Duration of the survey was an important consideration for how I posed each 
question. I kept all questions as multiple choice to keep the participant engaged and 
moving through the survey as fast as possible. The survey duration was limited to spans 
of 15-minute intervals due to the 0.25, 0.5, 0.75, and 1.0 credits offered to student 
researchers in the UO human subjects pool, so I needed to expedite all processes in the 
questionnaire with the way I framed the response choices. With all questions as multiple 
choice, the Qualtrics survey rating system, ExpertReview, predicted that the 
22 
 
 
questionnaire would take approximately 10.3 minutes to complete. I extended the 
duration to 15 minutes so that participants would have ample time to finish the survey.  
All questions and survey materials can be found in the appendix for reference 
while reading this section.   
3.3.2 Social Distancing and Mask Behavior Questions 
My intentions for the mask usage and social-distancing questions were for the 
participant to actively reflect on their own behavior as they moved through the survey, 
and to measure their willingness to change their social-distancing behavior. For these 
questions, I asked about which behaviors the respondent has participated in over the last 
24 hours, for which of those activities they decided to wear a mask, and their 
comfortability participating in a variety of indoor and outdoor activities. I duplicated 
each of these questions so the participant would first reflect on their current behavior 
and later make a decision about their future behavior. To measure their willingness to 
change their behavior, I positioned these questions before and after the participant 
viewed the visualizations. An example of one set of these parallel questions regarding 
social-distancing behavior can be seen in figures 3.30 and 3.31 below.  
23 
 
 
 
Figure 3.30 
Question seven in the survey asks participants about how often they are physically 
interacting with others. After viewing the visualizations, participants are asked the same 
question again regarding their behavior over the next three days in question twenty-two. 
 
Figure 3.31 
Question twenty-two asks participants about their social distancing behavior over the 
next three days to mirror question seven from before viewing the visualizations.  
 
24 
 
 
For the participant to fully reflect on their behavior, I asked about several 
aspects of their daily life regarding COVID-19 safety protocols. The survey asked the 
participant for their estimation of how many people they had direct contact with outside 
of their own household, which included a definition of direct contact: “’Direct contact’ 
means a conversation lasting more than 5 minutes with a person who is closer than 6 
feet from you without either person wearing a mask.” I also asked for an estimation of 
how many people around the participant decided to wear a mask in public spaces when 
social distancing was not possible over the past three days, with a definition of social 
distancing: “’Social distancing’ means staying 6 feet away from those around you.’ 
After answering these questions, the participant would view the randomly selected 
design of visualization and answer the associated comprehension questions, and then 
view the same social-distancing questions regarding their future behavior afterwards.   
Mirroring the precursory and subsequent behavior questions allowed for the results to 
simulate the participant’s choice to social distance and mask in the future.  
3.3.3 Measuring Comprehension Questions 
To measure the effectiveness of the visualizations, I asked questions that would 
gauge the respondent’s comprehension of the data. I structured the questions to ask for 
identical information in both charts so I could compare participant comprehension rates. 
For instance, after viewing the human-centered or default visualization for daily case 
rates, the survey participant would be asked: “Which state is being represented?”, 
“What is the single highest recorded number of daily cases in Florida?”, and “On 
September 25, approximately how many daily cases of COVID-19 were recorded in 
Florida?”  
25 
 
 
For the human-centered and default visualization for cumulative case rates, I 
asked for the same information but referred to a different date in the third question. 
These questions can be seen below in figures 3.32 and 3.33. 
26 
 
 
Figure 3.32 
Participants would first view the cumulative data set visualization, and then answer 
questions about which state is represented, the single highest number of recorded cases, 
and how many cases were recorded on July 7th. 
Figure 3.33 
After seeing the cumulative cases visualization, the participant would view the daily 
cases visualization and answer questions about which state is represented, the single 
highest number of recorded cases, and how many cases were recorded on September 
25. 
27 
 
 
Using the results of these questions, I could determine whether the respondent was truly 
participating in the survey, and whether they understood the information being 
displayed in the visual. 
3.3.4 Visual Appeal Questions 
After being asked about specific pieces of data in the visuals, I asked the 
participants about their personal preference and experience viewing the visualizations. I 
asked, “Complete the sentence: ‘Reading this chart was…’” with the Likert scale 
response of Very easy, Easy, Difficult, and Very difficult. I also asked, “Respond to the 
statement: ‘I enjoyed looking at this chart.’”, and “Respond to the statement: ‘I 
understand the information being portrayed in this graph,” with the responses Strongly 
agree, Somewhat agree, Somewhat disagree, and Strongly disagree. With these results, 
I was able to determine which visualizations were preferred and most pleasing to view.  
3.3.5 Demographic Questions 
In order to measure the effectiveness, appeal, and persuasiveness of my 
visualization across demographics, I asked questions referring to many areas of one’s 
identity. I asked each participant for their current ZIP code, if they have any known 
color blindness, their gender and age, if they are of Hispanic, Latino, or Spanish origin, 
their race, and the highest degree or level of school completed. The results of these 
questions allowed me to compare comprehension rates among younger versus older 
participants, female and male participants, and those with colorblindness and those 
without. Furthermore, knowing which demographic responded best to each 
visualization allowed me to predict how each graphic could be improved for specific 
age groups, genders, colorblindness, and other identity traits.  
28 
 
 
3.4 Participant Criteria 
The survey was distributed through three main channels: the UO Human 
Subjects pool to students taking Psychology courses, posting the survey on my personal 
LinkedIn, Facebook, and GroupMe accounts, and by use of Amazon’s Mechanical 
Turk. Figure 3.41 shows the recruitment statement used to distribute the survey to UO 
Computer Science students. I used a sample size calculator (Creative research systems, 
2012) to determine the number of people I needed to participate in the survey. With a 
confidence level of 95% and a confidence interval of 5, I needed to reach 384 
participants. After the recruitment period through all three channels, I was able to gather 
427 responses.  
Credit allocation affected the number of people I was able to recruit to take the 
survey. The UO human subjects pool (UOHSP) participants taking a psychology course 
must participate in 5 credit hours of research. Since I described the survey to have a 
duration of 15 minutes, I offered 0.25 credits to each student. While I predicted the 
short duration would increase the number of participants, the UOHSP only contributed 
one participant to my overall sample. Many of the surveys and experiments posted on 
the UOHSP site offer more credits to students, which may have negatively impacted my 
ability to recruit respondents. I had greater success recruiting participants using my 
personal social media and offering payment to workers on Mechanical Turk.  
Using my personal social media as a distribution method was a successful tactic 
in recruiting participants. I posted a statement asking viewers to take the survey on my 
Facebook, LinkedIn, and sent a private message to the Oregon Consulting Group, of 
which I am a member. After splitting the sample from the Mechanical Turk workers and 
29 
 
 
UOHSP, the survey results show that I was able to recruit 111 participants over the span 
of a month using this method. This is an interesting finding because I offered no reward 
for taking the survey, which leads me to believe that familiarity with the person 
distributing the survey has a positive effect on participant size. I continued this positive 
recruitment trajectory when I distributed the survey on Mechanical Turk.   
I was able to recruit the largest number of participants from Amazon’s 
Mechanical Turk platform. Amazon’s Mechanical Turk offers researchers access to a 
pool of workers that receive small payments for each Human Intelligence Task (“HIT”) 
they complete. These HITs are often transcribing receipts, reading documents, and 
taking surveys. Thanks to funding from my advisor, I was able to offer participants 
$3.75 for completing the single HIT of my 15-minute survey—an equivalent of $15 an 
hour. I decided to offer this amount after reading an article from the Atlantic (Semuels, 
2018) about how little Mechanical Turk workers earn. The worker interviewed in the 
article shared that she made an average of $4 to $5 an hour from her participation in 
these tasks (Semuels, 2018). With this payment, I was able to receive 275 responses 
within 9 hours.  
3.5 Data Analysis  
My null hypothesis is that the design method used in generating the visualizations 
has no effect on comprehension of data, aesthetic appeal of the chart, or willingness to 
social distance. My alternative hypothesis is that participants who view charts generated 
using a human-centered approach will show higher rates of data comprehension, 
aesthetic appeal, and willingness to social distance than those who viewed the default 
charts. I used these hypotheses to determine whether my human-centered design (HCD) 
30 
 
 
was more effective in each demographic of respondents. For all of my analyses, I 
computed a confidence interval of the difference between proportions. This means that I 
compared the difference between each response coming from pre- and post-
visualization questions, and determined if that difference were larger than the critical 
value associated with p-value of .05. If it was larger than .05, then the result was 
statistically significant, and I could reject the null hypothesis and accept the alternative 
hypothesis for that question.   
In order to determine how many participants of each demographic were viewing the 
HCD visualizations, I filtered the results based on whether every default visualization 
question was empty. The set of visualization questions associated with the default 
visualizations would only be empty if the participant had not been presented with them. 
This leaves the set of participants who were presented the opposite visualizations—in 
this example, the human-centered visualizations. After splitting the respondents based 
on this filter, I had a new set of respondents and needed to determine the new 
confidence interval for the size of the set using an interval calculator (Creative research 
systems, 2012) for a confidence level of 95%.  
In order to measure the effectiveness of each visualization, I analyzed the rates of 
correct responses, the rates of appeal based on the Likert scale responses, and the 
difference in social-distancing behavior and masking after viewing the visualizations. I 
started by computing the confidence interval for each demographic sample size, and 
compared the responses between the two designs. For instance, the entire sample 
contained 427 participants which corresponds to a confidence interval of +/- 4.74% with 
a confidence level of 95%, whereas the sample size of the colorblind participants was 
31 
 
 
56 which corresponds to a confidence interval of +/- 13.1% with the same confidence 
level (Creative research systems, 2012).  
For Likert scale responses, I added the rates of “Very comfortable” and 
“Comfortable”; “Very uncomfortable” and “Uncomfortable”; “Strongly agree” and 
“Agree”; “Strongly Disagree” and “Disagree”; “Very easy” and “Easy”; and “Very 
difficult” and “Difficult” together to compare the results using the confidence interval 
for each set. Each of the categories of responses sum to 100%; e.g., the response rates of 
“Very comfortable”, “Comfortable”, “Very uncomfortable”, and “Uncomfortable” 
totaled to 100%.  
32 
 
 
Chapter 4: Results 
This research was motivated by the questions posed in Chapter One: Why use a 
human-centered design approach to creating visualizations? Can visualizations be clear, 
and convincing enough to change people’s behavior after looking at them? What is the 
best way to present visualizations of data for maximum human-understanding? 
Ultimately, my study answered the question: which method of visualizing COVID-19 
case data will result in the highest number of responses showing a willingness to 
continue social-distancing and mask-wearing behavior, or change their behavior if not 
doing so? In this chapter I will provide a breakdown of the survey results, including an 
analysis of each sample of participants. I will go through each section of the survey to 
address my research questions, and determine if my hypothesis was supported. The 
legend showing the categories of each of the questions is shown below in figure 4.10. 
 
Figure 4.0: Question Legend 
Legend shows each category associated with the questions. The numbers associated 
with each question correspond to the order of creation during survey design.  
 
33 
 
 
4.1 Participant Demographic 
The demographic breakdown of the 392 survey participants is based on a few 
categories: ZIP code, color blindness, age, gender, ethnic background, and education. 
Analysis of the ZIP code entries show that participants, at the time of taking the survey, 
are residing in 38 states across the U.S., including Alaska and Hawaii.  
 
 
Figure 4.11 
Participants were asked to enter their current ZIP code, which resulted in codes across 
38 states.  
4.1 Overview of Entire Sample 
There were 392 responses for the entire sample of survey participants. The 
participants can be split into two groups based on the visualizations they viewed: those 
who were presented the HCD designs, and those who were presented the default 
34 
 
 
designs. The sample was evenly split with 196 HCD participants and 196 default 
participants.  
The social-distancing and masking questions, Q62, Q63, Q97, and Q130, were 
posed in order for the user to reflect on their own masking and social-distancing 
behavior, and prime them for future questions about COVID-19. The results of these 
questions show that respondents are going to a grocery store, a pharmacy, work, or 
school, spending time with people the respondent does not live with, or exercising 
outdoors most frequently. Of those activities, they are wearing a mask most often when 
they go to a market or pharmacy--28.15% of the time. Somewhat surprisingly, 
respondents are wearing a mask most infrequently on public transportation--5.53% of 
the time. On average, participants reported having direct contact with 8.58 people at 
work or school, 8.9 people at a grocery store or pharmacy, 8.94 people at social 
gatherings, and 7.93 people during other activities. Across many demographics 
including this set of all participants, results did not show increases in comfortability for 
exercising, whether indoors or outdoors. All these activities regarding comfortability are 
related to how the respondent feels participating in these activities in general, with no 
assumption of wearing or not wearing a mask, to measure whether there was an impact 
of viewing the visualizations on their willingness to participate in these activities. 
Respondents reported that the people around them chose to wear masks “All of the 
time” at a rate of 34.78%, and “Most of the time” at a rate of 37.92%. 
The pre-visualization behavior questions, Q93, Q98, and Q131, all ask the 
respondent about their current mask-wearing and social-distancing behavior. The post-
visualization behavior questions, Q71, Q100, and Q142 ask the same questions but refer 
35 
 
 
to future masking and social-distancing behavior. These questions are used to measure 
the participant’s willingness to change their behavior after viewing the visualizations. 
Results from the questions prior to viewing the visualizations show that 63.63% of 
respondents reported that they felt comfortable going to work or school in the past two 
weeks. After viewing the visualizations, participants reported that they feel 71.51% 
comfortable in the same setting when every person is wearing a mask—an increase of 
7.88%.  
Results from the remaining four pre- and post- visualization questions also show 
significant increases in comfortability across activities. Participants said they felt 
comfortable going to a bar or restaurant in the past two weeks 47.38% of the time, but 
reported they feel comfortable 56.01% of the time when every person is wearing a 
mask—an increase of 8.63%. Similarly, respondents showed an increase in their 
comfortability rating in every activity aside from exercising outdoors.  
For the set of comprehension questions asked while viewing the visualizations, 
the results show that for Q161 and Q162, there is a difference between the accuracy of 
responses for the human-centered design and default charts, with the daily default chart 
receiving 81.38% correct responses, and the daily HCD receiving 72.82% correct 
responses—a difference of 8.56%. The remaining three questions in this block have 
participants rate readability and preference of the visualization they viewed. On Q66 
and Q124, the cumulative default chart was rated as easiest to read, with a response rate 
of 89.8% for the cumulative default visual, and 81.63% for the cumulative HCD visual. 
For questions 145 and 146, 83.16% of participants who viewed the cumulative default 
36 
 
 
visual said they enjoyed looking at the chart, whereas only 75.39% of those who viewed 
the cumulative HCD chart said they enjoyed it.  
4.2 Age 
For this demographic, I split the age groups of participants on those who are 18 to 34 
years old, and those who are 35 years or older.  
4.2.1 Ages 18-24  
There were 221 responses for the set of 18-to-34-year old’s. Recall that the pre-
visualization behavior questions, Q93, Q98, and Q131, all ask the participant about their 
current mask-wearing and social-distancing behavior. The post-visualization behavior 
questions, Q71, Q100, and Q142 ask identical questions but refer to future masking and 
social-distancing behavior. Results from Q93 and Q71 reflect an increase of 7.54% in 
overall comfortability going to work or school after viewing the visualizations. 
Similarly, respondents showed an increase in comfortability when going to a market or 
pharmacy; a bar, restaurant, or café; attending an event with more than ten people; using 
public transit; and exercising in a fitness facility. For question 98 and 100, which ask 
about how often the participant will wear a mask, the respondents reported they were 
“always” wearing one with a frequency of 53.39% before the visualization, and a 
frequency of 63.80% afterwards—an increase of 10.41%.  
For the set of comprehension questions asked while viewing the visualizations, 
results show that for Q135, Q161, Q136, and Q162, there is a statistically significant 
level of difference between the accuracy of responses for the human-centered design 
and default charts, with the daily default chart receiving 82.88% correct responses over 
the daily HCD chart that received 73.64% correct responses. Inversely, the cumulative 
37 
 
 
HCD chart received 83.64% correct responses, a jump over the cumulative default chart 
that received 68.47% correct responses.  
For the questions regarding readability and visual appeal, results reflect that 
participants enjoyed looking at the cumulative default chart most frequently. When 
asked “Respond to the statement: ‘I enjoyed looking at this chart.’”, 85.59% of 
participants who viewed the cumulative default chart responded “agree”, whereas only 
78.18% of participants responded “agree” for the cumulative HCD chart.   
4.2.2 Ages 35+  
For the group of participants 35 years or older, there were 148 responses. The pre-
visualization behavior questions reflect an increase in level of comfortability for 
participants when going to work or school; going to a market or pharmacy; going to a 
bar, restaurant, or café; spending time with someone they are not staying with; attending 
an event with more than ten people; and using public transit. There was no difference 
between comfortability levels in exercising in a fitness facility or exercising outdoors 
before and after viewing the visualizations. For Q98 and Q100 about mask usage, 
respondents reported that they would wear a mask “frequently” 15.54% of the time 
prior to viewing the visualization, which increased to a rate of 25.00% after viewing the 
visualizations. 
For the set of comprehension questions asked while viewing the visualizations, 
the results show that respondents correctly answered the daily HCD chart most 
frequently, but rated the cumulative default chart as easiest to read. In Q164, the daily 
HCD visualization received 65.38% correct responses, whereas the daily default 
visualization only received 55.71% correct responses. In regards to readability, 
38 
 
 
participants rated the cumulative default chart as “easy” with 90% frequency, whereas 
the cumulative HCD chart only received “easy” for 75.64% of the responses.  
4.3 Gender 
The survey results reflect a gender split of 38% female, 61% male, and 1% other 
participants. In this section, I will report the results of the female and male 
demographics.   
4.3.1 Female Participants 
There were 140 responses from female participants. Results from the pre-visualization 
behavior questions reflect that female participants were less persuaded by the 
visualizations regarding comfortability in comparison to the demographics analyzed 
above. Participants showed an increase in comfortability before and after viewing the 
visualizations in regards to going to a market or pharmacy; spending time with someone 
they are not staying with; attending an event with more than ten people; and using 
public transit. There was no meaningful difference reported in levels of comfortability 
regarding going to work or school; going to a bar, restaurant, or café; exercising in a 
fitness facility; or exercising outdoors. For questions 98, 100, 131, and 142, there was 
no meaningful difference between willingness to wear a mask or intentionally avoid 
contact with others before and after viewing the visualizations.  
The results of the comprehension questions for female participants show a mix 
of both the default and HCD visualizations in terms of comprehension and readability. 
For Q161 and 162, there is a level of difference between the accuracy of responses for 
the human-centered design and default charts, with the daily default chart receiving 
81.54% correct responses, whereas the daily HCD chart received 70.67% correct 
39 
 
 
responses. Furthermore, the cumulative HCD visualization received 77.33% correct 
responses in Q136, where the cumulative default chart received 66.15% correct answers 
in Q135. The other comprehension questions in this block did not show a meaningful 
difference in accuracy of responses. In terms of readability, participants rated the 
cumulative default chart as “easy” 90.77% of the time, whereas the cumulative HCD 
visualization only received 74.67% “easy” responses. The other questions regarding 
enjoyment of looking at the chart and understanding the information being portrayed 
did not reflect meaningful differences between the HCD and default charts.  
4.3.2 Male Participants 
There were 228 responses from male participants. The pre-visualization behavior 
questions reflect that male respondents reported increased levels of comfortability in 
more activities after viewing the visualizations in comparison to the female participants. 
The results of Q93 and Q71 reflect that male participants showed increases in comfort 
levels when asked about going to work or school; going to a market or pharmacy; going 
to a bar, restaurant, or café; attending an event with more than ten people; and using 
public transit. Results reflected no meaningful differences between the level of 
comfortability spending time with someone they are not staying with; exercising in a 
fitness facility; or exercising outdoors.  
For question 98 and 100, which ask about how often the participant will wear a 
mask, the respondents reported they were “always” wearing one 47.37% of the time 
before viewing the visualization, but reported being willing to wear one “always” 
53.95% of the time after viewing the visualizations—an increase of 6.58%. Results 
40 
 
 
reflected no meaningful differences between the extent to which participants avoid 
contact with others before and after viewing the visualizations.  
The results of the comprehension questions for the male participants show that 
the default charts had the highest rates of comprehension, readability, and visual appeal. 
For Q161 and Q162, there is a level of difference between the accuracy of responses for 
the human-centered design and default charts, with the daily default chart receiving 
82.76% correct responses, where the HCD received 75.89% correct responses. On Q163 
and Q137, the cumulative default chart received 68.10% correct responses, whereas the 
cumulative HCD chart received 60.71% correct responses—a difference of 7.39%.  On 
Q134 and Q137, participants answered the questions associated with cumulative default 
chart correctly at a rate of 68.10%, whereas the participants who viewed the cumulative 
HCD chart responded correctly at a rate of 60.71%. Respondents who viewed the 
cumulative default chart responded “agree” at a rate of 85.35% to the question asking 
about enjoyment of looking at the chart, whereas the respondents who viewed the 
cumulative HCD selected “agree” at a rate of 74.77%. Similarly, participants agreed 
that the daily default visualization was enjoyable to look at with a rate of 79.31%, 
where the daily HCD received “agree” for only 71.98% of the responses. There was no 
meaningful difference between comprehension of the state being represented, or ratings 
of understanding the information being portrayed or readability of the visualization. 
4.4 Colorblindness  
For this demographic of respondents, I separated the group into two sections: those with 
any form of color-blindness, and those with no colorblindness. I will provide an 
analysis of both samples below.  
41 
 
 
4.4.1 Colorblind Participants 
There were 56 responses from individuals who responded that they were Red-Green, 
Blue-Yellow, or completely colorblind.  
 The results of the pre-visualization behavior questions reflect little differences 
between those who viewed the HCD visualization and the default visualization. For 
Q93 and Q71 asking about activities the respondent is comfortable participating in, 
there were no meaningful differences between levels of comfortability before and after 
viewing the visualizations. In Q98, which asks about how often the individual wears a 
mask before viewing the visualization, respondents answered “sometimes” at a rate of 
37.50%, whereas after viewing the visualizations, they reported “sometimes” at a rate of 
21.43%--a decrease of 16.07%. In Q131 and Q142, participants selected that they 
intentionally avoid contact with others “most” of the time at a rate of 37.5% before 
viewing the visualizations, and at a rate of 55.36% after viewing the visualizations—an 
increase of 17.86%.  
The results of the comprehension questions for the colorblind (RG, BY, and 
complete) participants show that on Q135 and Q136, the cumulative HCD chart 
received 60% correct responses and the cumulative default chart received 30.77% 
correct responses—a difference of 29.23%. Furthermore, on question 163 and 164, the 
daily default chart received 38.46% correct responses, whereas the daily HCD chart 
received 53.33% correct responses.  
For the questions about readability, visual appeal, and understanding 
information being portrayed, the default visualization was rated highest overall. The 
daily default chart was rated “easy” to read at a rate of 100% in comparison to the daily 
42 
 
 
HCD chart that received “easy” at a rate of 86.67%. Regarding visual appeal, 
participants reported that they enjoyed looking at the daily default visualization at a rate 
of 96.15%, whereas the daily HCD visualization received 79.31% of responses 
corresponding to enjoyment. In response to the questions 139 and 138 that ask about 
understanding the information being portrayed, participants who viewed the cumulative 
default visualization selected “agree” 100% of the time, whereas participants who 
viewed the cumulative HCD visualization selected agree 86.66% of the time. Although 
the comprehension rates for the HCD designed visualizations are significantly higher 
than the default visualizations, the colorblind participants rated the default designs as 
more appealing and easier to comprehend.  
4.4.2 Non-Colorblind Participants 
There were 313 participants who responded that they have no form of colorblindness. 
The results of the pre-visualization behavior questions reveal that participants reported 
increased levels of comfortability in nearly every activity before and after viewing the 
visualizations. Respondents showed an increase in levels of comfort when going to 
work or school; going to a market or pharmacy; going to a bar, restaurant, or café; 
spending time with someone they are not staying with; attending an event with more 
than ten people, and using public transit. Like many other demographics in this chapter, 
there was no meaningful difference between levels of comfortability when exercising in 
a fitness facility or outdoors before and after viewing the visualizations.  
Results from the questions about social-distancing and mask behavior reflect an 
increased willingness to wear a mask before and after viewing the visualizations. Before 
looking at the visual, respondents reported that they “always” wore a mask at a rate of 
43 
 
 
56.23%, but responded that they would “always” wear a mask in the next three days at a 
rate of 62.30%. Furthermore, participants also responded that they are intentionally 
avoiding contact with other people “all” of the time at a rate of 23.32% before viewing 
the visualization, but responded they would avoid contact with others “all” of the time 
in the next three days at a rate of 29.71%. 
The results of the comprehension questions reflect that the default visualizations 
received the highest rate of correct responses overall. In question 161 and 162, 
participants viewing the daily default visualization answered correctly at a rate of 
85.81%, while those who viewed the daily HCD visualization answered correctly at a 
rate of 77.85%. The results of the remaining comprehension questions reflect that there 
is no meaningful difference between the comprehension rates of participants who 
viewed the HCD visuals and those who viewed the default visuals.  
Participants reported that the cumulative default visualization was easiest to read 
and most enjoyable to look at overall. In question 66, participants rated the cumulative 
default chart as “easy” to read at a rate of 89.03%, whereas the cumulative HCD 
visualization was rated “easy” to read at a rate of 80.38%. Furthermore, participants 
reported that they enjoyed looking at the cumulative default chart at a rate of 81.94%, 
where participants responded that they enjoyed viewing the cumulative HCD 
visualization at a rate of 73.89%.  
44 
 
 
Chapter 5: Discussion and Recommendations 
I will now discuss the above results, provide recommendations for producing effective 
visualizations, and offer ways to improve this research in future work.  
5.1 Producing Effective Visuals 
The results of the survey do not provide sufficient evidence to reject the null hypothesis, 
and suggest that there is room for improvement in the human-centered design process 
for producing visualizations. Across participant demographics, the human-centered 
designs were not shown to be significantly more effective or compelling than the 
default designs. In the cases of female, male, 18-34, 35+, and non-colorblind 
participants, the default designs had higher rates of comprehension, readability, and 
visual appeal.   
There are a few potential factors as to why participants did not respond to the 
HCD visualizations with a higher rate of comprehension than the default visualizations. 
One factor is the tool I used to generate the visualizations. I decided to produce the 
visualizations using Microsoft Excel for its diversity of design elements, the quality of 
PNG exports for charts, and the default colors, fonts, and line sizes that provide a 
common representation of data visualizations. While this did provide an adequate 
foundation for producing the default visualizations, it was not effective in creating a 
human-centered design distinct from the default visualization. Excel uses slanted labels 
on the x-axis as the default design for their charts, despite Munzner’s specific guidance 
against this positioning, yet this did not seem to prevent users who were presented these 
labels from successfully reading the graphs. The elements of the visualizations were not 
altered enough for the participants to show a meaningful change in their understanding 
45 
 
 
of the data. If I had chosen to generate the visualizations using more extensive 
technology, like D3.js, R, ArcGIS, or Python’s datascience library, I could have 
produced interactive visualizations supported by Tufte and Munzer’s recommendations 
that may have commanded the attention of my participants more effectively. 
Another factor that contributed to these results is the testing environment of this 
survey. Due to this being a COVID-19 constrained virtual study, I could not enforce the 
device that they used to view the visualizations. Although my consent statement asked 
participants to use a laptop, desktop computer, or tablet, many of them could have been 
taking the survey on their phone. This variation of device sizes contributes to variation 
in the participant’s ability to comprehend the visualizations. If this were an in-person 
study, all participants would take the survey on the same device so I could ensure that 
they viewed identical visualizations.  
5.2 Changing Public Behavior 
One component that affected willingness to change behavior is the timeliness of 
the datasets. My visualizations were produced in late December 2020, and the survey 
was not approved by the IRB until February 18, 2021. In December, the US was 
reaching some of its highest counts of new reported cases of the entire pandemic (The 
New York Times, 2021). By the time participants were taking the survey in February 
2021, COVID cases had dropped and the risk of infection may have been perceived as 
diminishing. The rate of decline in cases could have impacted the way participants view 
their own behavior, and their decision to wear a mask and socially-distance in the 
future.  
46 
 
 
5.3 Future Research 
It is important to note some ways that this research could be improved for future study. 
There are several components of my research process that I would have changed if I had 
the opportunity to conduct this study again: generating interactive visualizations, 
dividing participants based on different data sets, designing the survey for use with 
Mechanical Turk, and completing more comprehensive statistical analysis.    
 In future work, I suggest researchers use more effective technology to produce 
interactive visualizations. As I mentioned in Chapter One, the visualization that inspired 
this research was an interactive article from the New York Times about the spread of 
water droplets from a person’s sneeze. My lack of expertise in dynamic visualization 
generation and decision to use Excel limited my ability to produce these types of 
visuals. It would be interesting to see how D3.js, Python libraries, and R could be used 
to generate interactive visualizations that meet more of Tufte and Munzer’s guidelines. 
This would result in visualizations that are significantly different than the default 
designs I produced with Excel, and may lead to higher rates of comprehension and 
willingness to change behavior.  
 It would be valuable to modify this survey to divide participants based on the 
ordering of the HCD and default designs, rather than the daily and cumulative data sets. 
I would show the participant the default images first, ask them questions to measure 
their understanding of the data, and then show them the human-centered design with the 
same questions from the first block. The results from this modified survey would show 
a difference in accuracy that would measure the readability of each design. If I included 
47 
 
 
the preference questions from the original survey, then it would also be a more accurate 
measure of which visuals were more pleasing to view.  
 Researchers should take advantage of Amazon’s Mechanical Turk to distribute 
their online surveys, but be very intentional about the way they design the question set 
to capture and verify completion. Mechanical Turk recommends that those posting their 
surveys with Qualtrics should track their responses by either generating a unique code 
for the worker to upload to Amazon, or by entering their WorkerID directly into the 
survey itself. I chose to have the survey generate a random ID number to each 
participant. Although I followed the appropriate steps suggested by MTurk, several 
workers entered false ID numbers into Amazon and did not truly complete the survey. 
Because I didn’t ask for any information from the workers, verifying that each worker 
had completed the survey involved comparing every random ID from Qualtrics with the 
random ID from Amazon. Furthermore, the survey was not closed for distribution in 
other sources, so there were several participants presented with a random ID that were 
not associated with Amazon. The verification process took a significant amount of time 
and prevented me from filtering the results on those from Mechanical Turk and those 
who were not. If I had asked the participants to enter their WorkerID, then I could have 
verified their response and successfully filtered results based on this value. My 
recommendation for survey designers is to ask participants for their WorkerID to avoid 
a lengthy verification process.  
Future work in this area should include more comprehensive analysis of the 
survey results using an ANOVA test, Pearson’s chi-squared test, and a two-sample t-
test. My lack of familiarity with high level statistical analysis and time prevented me 
48 
 
 
from completing these analyses, but would highlight variation within the data that I did 
not present. The ANOVA test would determine whether the survey results are 
significant, especially among the questions that involved Likert scale responses. 
Pearson’s chi-squared test provides an evaluation of categorical data with likelihood 
that any observed difference between the sets arose by chance. This would provide a 
more in-depth analysis of the comprehension questions to see which visualizations are 
more effective. The t-test determines if there is a significant difference between the 
means of two groups, and would be most effective in analyzing the questions about 
comfortability in participating in different activities. 
With these statistical tests, I propose that researchers continuing this work also 
take advantage of the many demographic questions asked in this survey, and separate 
the participant pool using other filters. If I had time, I would conduct analyses of 
college and non-college educated, white participants and participants of color, 
Mechanical Turk workers and those not working for Mechanical Turk, and participants 
by region of the US based on ZIP code. 
49 
 
 
Conclusion 
In summary, the results of this survey do not reject the null hypothesis. This 
means that there is no difference between the effectiveness of my human-centered 
design approach and the default design approach. However, one could draw from the 
results of this study that presenting participants with visualizations related to public 
health have the power to compel them to reconsider their future behavior. There are 
many ways to expand this research to determine the best way to visualize data for it to 
be more persuasive, effective, and pleasing to view, but this study provides a foundation 
for measuring the effectiveness of human-centered data visualization methods.  
 
 
50 
 
 
Appendix 
Glossary 
Amazon Mechanical Turk: Amazon Mechanical Turk is a crowdsourcing website for 
businesses to hire remotely located "crowdworkers" to perform discrete on-demand 
tasks that computers are currently unable to do. It is operated under Amazon Web 
Services, and is owned by Amazon. (Wikipedia Contributors, 2019) 
 
Comprehension: the capacity for understanding fully (Definition of 
COMPREHENSION, n.d.) 
 
Data visualization (data vis): Models that provide visual representations of datasets 
designed to help people carry out tasks more effectively (Munzner, 2015).  
 
Direct contact: A conversation lasting more than five minutes with a person who is 
closer than 6 feet from you without either person wearing a mask. 
 
Graphical excellence: Tufte’s guidelines for producing effective visualizations as 
written in The Visual Display of Quantitative Information. See chapter 2.4 about current 
methods in human-centered design.  
 
Human-Centered Design (HCD): An approach to problem solving, commonly used in 
design and management frameworks that develops solutions to problems by involving 
51 
 
 
the human perspective in all steps of the problem-solving process (Wikipedia 
Contributors, 2019). 
 
Human-subjects research: Research involving "a living individual about whom an 
investigator (whether professional or student) conducting research: bbtains information 
or biospecimens through intervention or interaction with the individual, and uses, 
studies, or analyzes the information or biospecimens; or obtains, uses, studies, analyzes, 
or generates identifiable private information or identifiable biospecimens (Definition of 
Human Subjects Research | grants.nih.gov, n.d.). 
 
Institutional Review Board (IRB): Under FDA regulations, an Institutional Review 
Board is group that has been formally designated to review and monitor biomedical 
research involving human subjects. In accordance with FDA regulations, an IRB has the 
authority to approve, require modifications in (to secure approval), or disapprove 
research. This group review serves an important role in the protection of the rights and 
welfare of human research subjects (Center for Drug Evaluation and Research, 2019). 
 
Repository: This is a folder where files are stored inside a project on Github, a version 
control platform.  
 
Social distancing: The practice of staying 6 feet away from those around you. 
52 
 
 
Survey Materials 
Question Breakdown 
 
Figure SM1: Q62 
53 
 
 
 
Figure SM2: Q63 
 
Figure SM3: Q97 
54 
 
 
 
Figure SM4: Q130 
55 
 
 
 
Figure SM5: Q93 
56 
 
 
 
Figure SM6: Q98 
 
Figure SM7: Q131 
57 
 
 
 
Figure SM8: Q50 / Q165 / Q122 / Q166 
 
Figure SM9: Q135 / Q136 
 
Figure SM10: Q134 / Q137 
58 
 
 
 
Figure SM11: Q66 / Q170 / Q124 / Q173 
 
Figure SM12: Q145 / Q171 / Q146 / Q174 
 
59 
 
 
Figure SM13: Q139 / Q172 / Q138 / Q175 
 
 
Figure SM14: Q161 / Q162 
 
Figure SM15: Q163 / Q164 
60 
 
 
 
Figure SM16: Q71  
61 
 
 
 
Figure SM17: Q100 
 
Figure SM18: Q142 
 
Figure SM19: Q52 
62 
 
 
 
Figure SM20: Q88 
 
Figure SM21: Q54 
63 
 
 
 
Figure SM22: Q55 
 
Figure SM23: Q56 
64 
 
 
 
Figure SM24: Q57 
65 
 
 
 
Figure SM25: Q58 
 
Figure SM26: Random ID 
Recruitment Statement 
The following statement was posted during the recruitment stage of my survey 
distribution. I sent the statement with a link to the survey on LinkedIn, Facebook, and 
through the UO computer science undergraduate and graduate student listserv. It was 
66 
 
 
not attached to the distribution post on Mechanical Turk because Amazon’s system put 
the link in their catalog independently.  
Hello! My name is Stephanie Schofield, and I’m a senior Computer 
Science student in the Clark Honors at the University of Oregon. I am 
conducting research about the effectiveness of data visualization 
methods using COVID-19 case data. The research is conducted in a 
survey that should take about 15 minutes. This study has been reviewed 
by an independent institutional review board, and is not expected to pose 
a risk to you. You are eligible to participate in the survey if you are at 
least 18 years old, and have the ability to take the survey on a laptop, 
desktop computer, or tablet. If you choose to participate, your responses 
will remain strictly anonymous. Any data which is presented will be 
reported in aggregate.  
 
Link to survey: 
https://oregon.qualtrics.com/jfe/form/SV_dmbWguXGatb5Ekl  
 
Your input is very important, and will help us understand how to create 
visualizations that are clear, impressionable, and have lasting effects on a 
person’s behavior. If you have any questions, please feel free to contact 
me: sschofie@cs.uoregon.edu. Thank you for your time! 
Survey Information Statement 
The following statements were attached at the beginning of the survey and introduced 
the survey’s purpose and scope. There were two statements prepared: one for 
Mechanical Turk participants, and the other for UO students and those not associated 
with Amazon. For live or in-person human-subject research, a consent statement is 
required in order to maintain ethical research practices according to the Declaration of 
Helsinki (World Medical Association, 2014). However, remote survey human-subject 
research requires a survey information statement. Writing and including this statement 
is a key factor in human-subject research, and was necessary in order to receive 
67 
 
 
approval from the UO Institutional Review Board (IRB). The statement for UO students 
and non-MTurk Workers reads:  
This voluntary survey is part of a research study led by Stephanie 
Schofield at the Robert D. Clark Honors College and University of 
Oregon Computer Science department. Your responses will help us 
understand how to produce effective visualizations for public health 
data.  
  
This survey is intended to take roughly 15 minutes to complete, 
and must be taken on a laptop, desktop computer, or tablet. The 
survey questions do not request any unique or personally identifiable 
information about you and your answers to all questions will remain 
confidential. 
  
Your responses may be shared with other researchers studying data 
visualization and human-computer interaction. We may also publish 
aggregate tables of results for public research use. Results published will 
be in aggregate and will not identify individual participants of their 
responses. There are no foreseeable risks in participating and no 
compensation is offered. 
 
If you are taking this survey as a worker on Amazon’s Mechanical Turk, 
you will be paid $3.75 for your participation. The survey should not take 
longer than 15 minutes.  
 
If you are taking this survey as a SONA participant, you will be awarded 
0.25 credits for your participation. If you choose to discontinue 
participation in this survey at any point after clicking through this 
consent page, you will receive 1/4 credit for each 15 minutes of 
participation, rounded up to the next 15 minutes. For example, if you 
complete 1-15 minutes you will receive 1/4 credit, if you complete 16-30 
minutes you will receive 1/2 credit, and so on. If you discontinue 
participating in the middle of the study, contact the listed researcher to 
receive partial credit. This research was reviewed by the University of 
Oregon Institutional Review Board. 
 
Your participation is voluntary. Your decision whether or not to 
participate will not affect your relationship with the UO Psychology 
Department or the UO Linguistics Department. If you decide to 
participate, you are free to withdraw your consent and discontinue 
68 
 
 
participating at any time without penalty. The Psychology and 
Linguistics Departments have established alternative assignments for 
students who do not wish to participate as research subjects. Please see 
your instructor if you would rather complete an alternative assignment. 
  
If you have any questions, please contact Stephanie Schofield at 
sschofie@cs.uoregon.edu. 
 
The statement for MTurk Workers reads: 
 
This voluntary survey is part of a research study led by Stephanie 
Schofield at the Robert D. Clark Honors College and University of 
Oregon Computer Science department. Your responses will help us 
understand how to produce effective visualizations for public health 
data.  
 
This survey is intended to take roughly 15 minutes to complete, and must 
be taken on a laptop, desktop computer, or tablet. The survey questions 
do not request any unique or personally identifiable information about 
you and your answers to all questions will remain confidential. 
 
If you are taking this survey as a worker on Amazon’s Mechanical Turk, 
you will be paid $3.75 for your participation. If you decide to participate, 
you are free to withdraw your consent and discontinue participating at 
any time without penalty. 
 
Your responses may be shared with other researchers studying data 
visualization and human-computer interaction. We may also publish 
aggregate tables of results for public research use. Results published will 
be in aggregate and will not identify individual participants of their 
responses. There are no foreseeable risks in participating. This research 
was reviewed by the University of Oregon Institutional Review Board. 
 
If you have any questions, please contact Stephanie Schofield at 
sschofie@cs.uoregon.edu 
Charts 
There were four visualizations that participants could potentially view while taking the 
survey, which were created using two datasets: daily cases and cumulative cases of 
69 
 
 
COVID-19 across the state of Florida from March 1 to December 19, 2020. The two 
datasets were represented in the form of two separate line charts. Each of the two charts 
for the respective datasets were created using a Human-Centered Design (HCD) 
methodology, and a default methodology. The four different charts can be seen below.  
70 
 
 
 
Figure C1: Human-Centered Daily Cases Visualization 
Daily cases of COVID-19 across the state of Florida over March to December of 2020, 
created with a Human-Centered Design methodology.  
 
Figure C2: Default Daily Cases Visualization 
Daily cases of COVID-19 across the state of Florida over March to December of 2020, 
created using a default approach. 
Participants had a 50% chance of viewing either the HCD or the default visualizations, 
but were asked the same questions regardless of which chart they viewed.  
71 
 
 
 
Figure C3: Human-Centered Cumulative Cases Visualization 
Cumulative cases of COVID-19 across the state of Florida from March to December of 
2020, created with a Human-Centered Design methodology.  
 
Figure C4: Default Cumulative Cases Visualization 
Cumulative cases of COVID-19 across the state of Florida from March to December of 
2020, created using a default approach.  
 
 
72 
 
 
Debriefing Statement 
The following statement was attached to the last block of the survey to inform the 
participant of what the survey is measuring. This debriefing is another important aspect 
of conducting ethical human-subjects research.  
Thank you for your participation! 
  
Background: Today’s study examined the correlation between 
comprehension of data visualizations and willingness to wear a mask and 
social distance. While a lot of research is done into the efficacy of 
charting public health data, there isn’t as much research about how these 
charts affect people’s behavior after viewing them.  
  
Purpose of this study: We hope to determine which kinds of charts are 
most impressionable, clear to understand, and have lasting effects on a 
person’s behavior. These results will help data visualization researchers 
create more effective graphics for people to understand large, important 
data sets. Furthermore, we hope this survey showed participants the risks 
posed from not wearing a mask, and leads to a higher rate of social-
distancing and mask-wearing behavior in young adults most prone to 
spreading Coronavirus.  
   
Your part: The part you play in this research is very important! Just 
giving us an idea of which visualizations were difficult or easy to read, 
some of your social distancing behavior, and your willingness to 
continue or change those habits will provide great insight. Thank you for 
contributing your valuable time to this survey today.  
  
Feedback and further information: If you have additional questions 
about this study, please feel free to ask the experimenter, Stephanie 
Schofield at sschofie@cs.uoregon.edu, 541-704-8191, or her advisors, 
Nicole Dudukovic, ndudukov@uoregon.edu, Department of Psychology; 
Joe Sventek, jsventek@uoregon.edu, Department of Computer Science. 
If you have any questions concerning your rights as a research 
participant, please contact Research Compliance Services, 5237 
University of Oregon, Eugene, OR 97403, 541-346-2510, or email 
ResearchCompliance (at) uoregon.edu. You can also email the Human 
Subjects Coordinator for psychology and linguistics research at hscoord 
(at) uoregon.edu. 
73 
 
 
Notice of Amendment Review and Exempt Determination 
 
Figure N1: IRB Exempt Determination 
Initial IRB Exemption for the study was received on February 18, 2021, and allowed 
me to move forward with distributing the survey to participants.  
74 
 
 
 
Figure N2: IRB Amendment Review 
The IRB approved my amendment application to use Amazon’s Mechanical Turk on 
April 14, 2021, and allowed me to distribute the survey to Mechanical Turk workers
75 
 
 
 
References 
CDC Case Task Force. (2021, April 29). United States COVID-19 Cases and Deaths by 
State over Time | Data | Centers for Disease Control and Prevention. 
Data.cdc.gov; Centers for Disease Control and Prevention. 
https://data.cdc.gov/Case-Surveillance/United-States-COVID-19-Cases-and-
Deaths-by-State-o/9mfq-cb36 
Center for Drug Evaluation and Research. (2019). (IRBs) and Protection of Human 
Subjects. U.S. Food and Drug Administration. https://www.fda.gov/about-
fda/center-drug-evaluation-and-research-cder/institutional-review-boards-irbs-
and-protection-human-subjects-clinical-trials 
Clark, J. (2011). Let’s reflect: what is the point? British Journal of General Practice, 
61(593), 747–747. https://doi.org/10.3399/bjgp11x613232 
Creative research systems. (2012). Sample Size Calculator - Confidence Level, 
Confidence Interval, Sample Size, Population Size, Relevant Population - 
Creative Research Systems. Surveysystem.com. 
https://www.surveysystem.com/sscalc.htm 
CSSLOhioStateU. (2014). Your Survey Closed, Now What? Quantitative Analysis 
Basics. In YouTube. https://www.youtube.com/watch?v=4q5UZwwidRI 
Definition of COMPREHENSION. (n.d.). Www.merriam-Webster.com. 
https://www.merriam-webster.com/dictionary/comprehension 
Definition of Human Subjects Research | grants.nih.gov. (n.d.). Grants.nih.gov. 
https://grants.nih.gov/policy/humansubjects/research.htm 
Desjardins, J. (2019, April 17). How much data is generated each day? World 
Economic Forum. https://www.weforum.org/agenda/2019/04/how-much-data-
is-generated-each-day-cf4bddf29f/ 
etanrioven. (2015, March 8). Bill Gates Graph of the Year – Causes of Untimely Death 
– Class comments. Nehirege. https://nehirege.wordpress.com/2015/03/08/bill-
gates-graph-of-the-year-causes-of-untimely-death-class-comments/ 
Katz, J., Sanger-Katz, M., & Quealy, K. (2020, July 17). A Detailed Map of Who Is 
Wearing Masks in the U.S. The New York Times. 
https://www.nytimes.com/interactive/2020/07/17/upshot/coronavirus-face-mask-
map.html 
Koop, A. (2021, March 27). Visualized: The Richest Families in America. Visual 
Capitalist. https://www.visualcapitalist.com/visualized-the-richest-families-in-
america/ 
76 
 
 
Kress, G. R., & Theo Van Leeuwen. (1996). Reading images: The Grammar of Visual 
design. Routledge. 
Moraes Bueno Rodrigues, A., Diniz Junqueira Barbosa, G., Côrtes Vieira Lopes, H., & 
Diniz Junqueira Barbosa, S. (2021). What questions reveal about novices’ 
attempts to make sense of data visualizations: Patterns and misconceptions. 
Computers & Graphics, 94(0097-8493), 32–42. ScienceDirect. 
https://doi.org/10.1016/j.cag.2020.09.015 
Munzner, T. (2015). Visualization Analysis & Design. Crc Press, Taylor & Francis 
Group. 
New York Times. (2020, July 28). nytimes/covid-19-data. GitHub. 
https://github.com/nytimes/covid-19-data/tree/master/mask-use 
New York Times. (2021, April 15). Florida Coronavirus Map and Case Count. The New 
York Times. https://www.nytimes.com/interactive/2021/us/florida-covid-
cases.html 
Parshina-Kottas, Y., Saget, B., Patanjali, K., Fleisher, O., & Gianordoli, G. (2020, April 
14). This 3-D Simulation Shows Why Social Distancing Is So Important. The 
New York Times. 
https://www.nytimes.com/interactive/2020/04/14/science/coronavirus-
transmission-cough-6-feet-ar-ul.html  
Rawsthorn, A. (2006, July 9). Quirky serifs aside, Georgia fonts win on Web - Style - 
International Herald Tribune. The New York Times. 
https://www.nytimes.com/2006/07/09/style/09iht-dlede10.2150992.html 
Semuels, A. (2018, January 23). The Online Hell of Amazon’s Mechanical Turk. The 
Atlantic; The Atlantic. 
https://www.theatlantic.com/business/archive/2018/01/amazon-mechanical-
turk/551192/ 
The New York Times. (2021, May 16). Coronavirus in the U.S.: Latest Map and Case 
Count. The New York Times. 
https://www.nytimes.com/interactive/2021/us/covid-
cases.html?pageType=LegacyCollection&collectionName=Maps+and+Trackers
&label=Maps+and+Trackers&module=hub_Band®ion=inline&template=storyli
ne_band_recirc 
Tufte, E. R. (1985). The Visual Display of Quantitative Information. Graphics Press. 
US Census Bureau. (2020, May 20). 2018 FIPS Codes. The United States Census 
Bureau. https://www.census.gov/geographies/reference-
files/2018/demo/popest/2018-fips.html 
77 
 
 
Venables, W. N., & Ripley, B. D. (2001). Modern applied statistics with S- PLUS. 
Springer, Cop. 
Wallach, O. (2021, March 3). Which Streaming Service Has the Most Subscriptions? 
Visual Capitalist. https://www.visualcapitalist.com/which-streaming-service-
has-the-most-subscriptions/ 
Wikipedia Contributors. (2019, January 31). Human-centered design. Wikipedia; 
Wikimedia Foundation. https://en.wikipedia.org/wiki/Human-centered_design 
Wikipedia Contributors. (2019, March 8). Amazon Mechanical Turk. Wikipedia; 
Wikimedia Foundation. 
https://en.wikipedia.org/wiki/Amazon_Mechanical_Turk 
Wilson, Dr. R. F. (2001, March 1). HTML E-Mail: Text Font Readability Study. 
Practical Ecommerce. https://www.practicalecommerce.com/html-email-fonts 
Wonkborg. (2013, December 23). Bill Gates’s graph of the year. Washington Post. 
https://www.washingtonpost.com/news/wonk/wp/2013/12/27/bill-gatess-graph-
of-the-year/ 
World Medical Association. (2014). WMA - The World Medical Association-WMA 
Declaration of Helsinki – Ethical Principles for Medical Research Involving 
Human Subjects. Wma.net; WMA - The World Medical Association-WMA 
Declaration of Helsinki – Ethical Principles for Medical Research Involving 
Human Subjects. https://www.wma.net/policies-post/wma-declaration-of-
helsinki-ethical-principles-for-medical-research-involving-human-subjects/ 
 
78