Conference PaperPDF Available

Showing People Behind Data: Does Anthropomorphizing Visualizations Elicit More Empathy for Human Rights Data?


Abstract and Figures

We investigate the impact of using anthropomorphized data graphics over standard charts on viewers' empathy for, and prosocial behavior toward suffering populations, in the context of human rights narratives. We present a series of experiments conducted on Amazon Mechanical Turk, in which we compare various forms of anthropomorphized data graphics—ranging from a single human figure that 'fills up' to show proportional data, to separated groups of individual human beings—with a standard chart baseline. Each experiment uses two carefully crafted human rights data-driven stories to present the graphics. Contrary to our expectations, we consistently find that anthropomorphized data graphics and standard charts have very similar effects on empathy and prosocial behavior.
Content may be subject to copyright.
Showing People Behind Data: Does Anthropomorphizing
Visualizations Elicit More Empathy for Human Rights Data?
Jeremy Boy
UN Global Pulse, USA
Anshul Vikram Pandey
NYU Tandon, USA
John Emerson
NYU School of Law, USA
Margaret Satterthwaite
NYU School of Law, USA
Oded Nov
NYU Tandon, USA
Enrico Bertini
NYU Tandon, USA
We investigate the impact of using anthropomorphized data
graphics over standard charts on viewers’ empathy for, and
prosocial behavior toward suffering populations, in the context
of human rights narratives. We present a series of experiments
conducted on Amazon Mechanical Turk, in which we compare
various forms of anthropomorphized data graphics—ranging
from a single human figure that ‘fills up’ to show proportional
data, to separated groups of individual human beings—with a
standard chart baseline. Each experiment uses two carefully
crafted human rights data-driven stories to present the graph-
ics. Contrary to our expectations, we consistently find that
anthropomorphized data graphics and standard charts have
very similar effects on empathy and prosocial behavior.
ACM Classification Keywords
H.5.m. Information Interfaces and Presentation (e.g. HCI):
Author Keywords
Information Visualization for the People; Anthropographics;
Empathy; Prosocial Behavior; Human Rights
In this article, we investigate the growing assumption that vi-
sually connecting abstract data with iconic representations of
people can elicit empathy for, and encourage prosocial behav-
ior toward those people [36, 42, 55, 66]. We focus specifically
on data visualizations of human rights/humanitarian (HR) is-
sues in data-driven stories, as these generally describe people’s
plight, and have real potential for eliciting empathy. The work
we present is the result of a collaboration between Information
Visualization (Infovis) and HR researchers.
Permission to make digital or hard copies of all or part of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed
for profit or commercial advantage and that copies bear this notice and the full citation
on the first page. Copyrights for components of this work owned by others than ACM
must be honored. Abstracting with credit is permitted. To copy otherwise, or republish,
to post on servers or to redistribute to lists, requires prior specific permission and/or a
fee. Request permissions from
CHI 2017, May 06–11, 2017, Denver, CO, USA.
© 2017 ACM. ISBN 978-1-4503-4655-9/17/05.. .$15.00.
Many HR practitioners are concerned that describing large
scale atrocities using abstract data might create “compassion
fatigue” [28, 60] by distancing readers from the reality of hu-
man suffering. HR-related data are often very sensitive regard-
ing e.g., the privacy, security, or safety of entire populations,
and can seldom be presented publicly without some amount of
aggregation. This makes it impossible for HR advocates to use
re-identifiable information like portrait photos to humanize
the data—even thought photos are known to elicit affective
responses [30]—and forces them to seek out other strategies
for illustrating the “human dimension” [28] of abstract data.
Visualization designers often use anthropomorphized data
graphics [66] (or anthropographics, see [1]) to this end. The
common rationale is that because such visualizations create
an immediate visual connection between abstract data and
actual people [36], they are better suited than standard charts
for eliciting empathy and prosocial behavior [42, 66]. We
refer to this rationale as the anthropographic assumption [36].
Although some critics are skeptical of the need for empathy in
visualization (e.g., [17]), others believe it can be particularly
useful for leveraging public awareness in the pursuit of social
change [31, 42, 66]. Many have discussed and nourished the
anthropographic assumption [17, 36, 55, 58, 67], but there is
no empirical evidence to support it.
Here, we formalize a design space for anthropographics, and
contribute the results of seven experiments as an initial as-
sessment of the anthropographic assumption in the context of
HR narratives. In contrast to our expectations, we find that
anthropographics and standard charts (e.g., pie charts) have
very similar effects on empathy and prosocial behavior. We
conclude that for HR narratives, anthropographics are neither
truly beneficial, nor detrimental. This complements Bateman
et al.’s call to learn more about the effects of different types of
visual embellishment in charts [6], and opens new perspectives
for exploring the benefits of anthropographics.
In this section, we first develop the arguments in favor of
the anthropographic assumption. We then motivate the gen-
eral design of our experiments with prior work on empathy,
and introduce known mechanisms behind a common form of
prosocial behavior: charitable giving.
Anchoring Graphics in Empathy
In January 2015, Jake Harris posted an article discussing the
importance of establishing a connection between data and
actual people in data-driven stories [36]. He argues that “from
a distance,” i.e., when abstract representations are used to
present human-related data, “it’s easy to forget the dots are
people.” He claims that “connecting with the dots” can elicit
empathy, which is “important for any dataset [used] to report
data about people or that affects people.
Lambert, Rees, and Zer-Aviv all agree with this assump-
tion [42, 55, 66]. While Harris mostly suggests using “wee
people” to “anchor graphics in empathy” [36], Rees encour-
ages visualization designers to think about the “atom” of the
data, e.g., an individual person; to “get into the life” of that
person; to “speak to it”; and to let the “greater piece” grow out
of it [55]. Evidence in other fields (e.g., robotics and animal
rights) provides support for these approaches, as it indicates
the more something appears humanlike, the more people feel
empathy for it [37, 54]. From a semiotics perspective, using
“wee people”—instead of only textual annotations on top of
standard charts—presumably has the advantage of eliciting vi-
sual interpretation mechanisms, which are far more open than
those of verbal texts [49]. Visual information typically builds
on experiential knowledge [52], whereas text relies more on
cultural conventions [32]. More realistic and/or expressive
visual renderings of victims [45] may tap into this experiential
knowledge, and affect viewers at a preverbal level [49].
Lester et al. have found that adding lifelike characters to in-
teractive learning environments has a strong positive effect
on students’ affect, motivation, and perception of their expe-
rience [44]. They call this the persona effect, and argue that
expressive characters are most effective. However, Miksatko et
al. have failed to reproduce this effect, and conclude that such
characters are neither detrimental, nor truly beneficial [48].
Genevsky et al. have found that showing pictures of orphan
children to potential charity-donors has a positive effect on
their affect, which in turn drives them to give more money [30].
They link this to the identifiable victim effect, which broadly
describes the belief that people empathize more, and are will-
ing to expend greater resources to help identified victims than
to help equal, or greater numbers of statistical victims [39].
While their results show that photo portraits are more effective
than abstract silhouettes, the iconic human representations
they use are not very expressive, and one could argue that us-
ing “wee people” instead of standard charts might better help
reproduce the identifiable victim effect, and thus encourage
more prosocial behavior [42, 66]. Establishing this is impor-
tant, as photos are not always available, and are often difficult
to use in the context of HR data-driven advocacy—whether
simply because the data are aggregated beyond the point of
individual distinction, or because they are too sensitive.
Finally, despite the “chart junk” debate [6, 62], previous work
in Infovis and HCI has shown that the use of pictorial infor-
mation like “wee people” in visualizations has a number of
benefits. Borgo et al. have found that visual embellishments
help viewers grasp key concepts more effectively [12], e.g., the
connection between abstract data and actual people; Borkin et
al. argue that pictograms can improve recognition and memo-
rability [14], and that creating redundancy, e.g., between text
and graphics, can help convey messages more effectively [13];
and Haroz et al. have shown that Isotype-like visualizations [2,
16] are generally engaging and can facilitate recall [35].
Although Cairo, Harris, Lambert, Rees, Schwabish, and Zer-
Aviv all speak of eliciting empathy with visualizations [17,
36, 42, 55, 66], it is sometimes unclear whether they share
the same definition of empathy. The term often suffers from
inconsistent use [27], and is commonly interchanged with e.g.,
sympathy. This creates confusion [11], which we attempt to
clarify here for the purpose of our experiments.
Empathy is broadly defined as “an affective response appropri-
ate to someone else’s situation rather than one’s own” [7]. It
is generally considered to have two dimensions: an affective
dimension, and a cognitive dimension [7, 10, 21, 27, 38, 47,
64]. Affective empathy relates to the vicarious experiencing of
another’s emotional state. Cognitive empathy relates to one’s
ability to accurately imagine another’s internal states (e.g., her
thoughts, feelings, or intentions) [38].
Affective empathy has two components which lead to empathic
distress [9, 21, 25, 27, 38]: empathic concern and personal
distress. Empathic concern describes other-oriented feelings
of compassion and sympathy [64], which are not necessarily
congruent with the feelings of the distressed other. It relates
to feeling for someone, rather than to feeling with someone.
Personal distress on the other hand describes self-oriented feel-
ings of anxiety, discomfort, and unease triggered in response
to the distress of others. In this article, any generic use of the
term empathy refers to these two emotional components.
Hoffman identifies five “modes” of empathy-arousal [38].
Mimicry,classical conditioning, and direct association are
primitive, automatic, and preverbal modes. Presumably, re-
alistic and/or expressive visual renderings of victims [45]
should help trigger these. Verbally mediated association and
perspective-taking are higher-level cognitive modes that give
scope to one’s feelings, and allow to empathize with others
who are not present. They are central to narrative empathy,
i.e., “the sharing of feeling and perspective-taking induced by
reading, viewing, hearing, or imagining narratives of another’s
situation or condition” [40], and suggest that empathy can be
aroused simply by conceiving of another’s plight [38], and
that it can be generalized to a type of plight, or to the plight of
an entire population (e.g., the poor or oppressed) [7].
Finally, empathy is often considered a precursor and motivator
for prosocial behavior [47, 64]. Batson’s empathy-altruism
hypothesis [7] typically suggests that observers will seek to
relieve their empathic distress by helping others. However, this
is not always true. Observers may enjoy another’s misfortune
if their prior relationship is bad [10]. Egoistic motives may
interfere [8], and the distress may become so aversive that
observers end up focusing solely on temselves [38]. Empathy
also has a familiarity bias [38]: observers generally tend to
have higher feelings for kin, in-group members, family, and
people who share their personal concerns.
Measuring Empathy and Understanding Donating
Research in developmental and social psychology has studied
empathy either as a mental state that can be triggered [8, 24],
or as a personal disposition that develops during childhood [24,
47]. Here, we mainly focus on the prior type of studies.
Batson uses a procedure which consists in prompting partic-
ipants with a fictional scenario describing a person in need,
and asking them to report on a 7-point scale how strongly
they feel different emotions described in a list of emotion ad-
jectives [7]. This list includes eight adjectives assumed to
reflect personal distress, and six assumed to reflect empathic
concern. Similarly, the picture/story procedure [27] consists
in telling participants brief stories while showing them pic-
tures (e.g., photos or drawings) of protagonists in emotional
situations, and asking them to verbally indicate how they feel.
Our experiments are inspired by these procedures.
Meanwhile, Genevsky et al. have shown how empathy can
shift donation preferences [30]. Lee et al. and Verhaert &
Van den Poel have also found that empathic concern positively
influences donations [43, 64]. However, Einolf argues that
this may not be true when the recipients of the donation are
not immediately present [24]. There are a number of factors
that influence donating behavior, which may complexify the
relation between empathy and donation decisions. Here, we
refer to these factors as donation biases for convenience.
Donating is usually seen as a prosocial behavior [43, 57], for
which moral identity is a strong predictor. Moral identity re-
lates to the importance one attributes to traits like fairness,
justice, or kindness [43]. Although it is usually theorized to
increase charitable giving, Lee et al. have found that it can
have a negative effect if the recipients of the charity are per-
ceived as responsible for their plight [43]. They also found
that empathy and the perception of justice can have mediat-
ing effects. In our experiments, we consider these factors as
possible moral,responsibility, and justice biases. Another
phenomenon that influences moral judgement—and thereby
donating behavior—is proportion dominance [5, 59]. Propor-
tion dominance describes the fact that people are likely to
choose to help higher proportions of statistical victims than
absolute numbers of them. Jenni & Lowenstein claim that this
dominance is responsible for the identifiable victim effect [39].
In our experiments, we consider the possibility of a proportion
dominance bias. Finally, availability has a strong influence
on decision-making [63], as it operates on the notion that if
something can be recalled, it must be important. We pay par-
ticular attention to topic availability in our experiments, as we
expect it may influence the familiarity bias, as well as people’s
perception of how important it is to donate to specific causes.
In this section, we present the general design of our experi-
ments. We detail the HR narratives we created, and describe
our design space for anthropographics.
General Study Design
Each experiment consisted of two HR narratives. These were
composed of a single chart condition, set in a unique data-
driven story. Chart conditions were: an anthropographic
(variable), and a standard chart (baseline)—except in Ex-
periment 4 (Exp. 4), where we compared the standard chart
with text alone. We set these conditions in data-driven stories
for ecologic validity, as most data graphics end up embedded
within additional content like text [35]—especially in HR nar-
ratives. Participants were thus exposed to one chart condition
in a first story, then to the other in a different story. We opted
for this design, rather than for a full-factorial design—in which
participants would have seen both conditions twice (once in
each story)—because we did not want to force them to read the
same story several times. We feared this might create lassitude
or frustration, which could have confounded their feelings. Fi-
nally, we counterbalanced the order of conditions and stories
to minimize the possible effects of anthropomorphic carry-
over. As both stories were about human suffering, we feared
that participants who were exposed to the anthropographic
condition in the first narrative might automatically transfer the
“human dimension” [28] of the data to the second narrative,
thus confounding the effect of chart conditions.
Although we only used two stories in each experiment, we
initially created four. All were related to the Syrian crisis, and
were developed with two HR researchers (co-authors of this
article). We focused on Syria, as we expected participants
would be familiar with the context. Each story described chil-
dren’s plight (similarly to [30]) through the lens of one of the
following topics: access to clean
, access to
, and
internal displacement
(IDP). We felt that focus-
ing on children would mitigate the responsibility and justice
biases, since children presumably cannot be held accountable
for their situation. We also expected this would amplify the
moral bias, since helping children should be perceived as at
least as morally important as helping adults, if not more. All
stories shared the same five-step narrative structure, to avoid
possible narrative-structure-related confounds. Step 1 intro-
duced the context. Step 2 showed a geographic overview of
the situation. Step 3 presented a simple demographic statistic
before the crisis. Step 4 showed the same statistic today. Step 5
concluded with a takeaway consideration on how children are
suffering from the situation.
Visualization template:
we created a custom slideshow visu-
alization template [56] to comply with our five-step narrative
structure. Each slide corresponded to one step. Progression
through the different slides was achieved using the arrow keys
on a keyboard. We kept visual compositions quasi-identical
for each story. Slide 1 showed only text. Slide 2 showed a
choropleth map of Syria and a short paragraph of text. Slide 3
showed a chart and a short paragraph of text. Slide 4 showed
the updated chart and a short paragraph of text. Slide 5 showed
only text. Varying chart conditions were restricted to slides 3
and 4. We also kept other visual design attributes like fonts,
background color, and transition types consistent across all
stories. The only slight variations were dominant color—used
to visually differentiate topics—and placement of the short
paragraphs in slides 3 and 4—depending on the chart con-
dition. We hoped this would prevent any kind of aesthetic-
preference-related confounds. To finish, we added a simple
hover interaction to the charts for showing labels and values.
for our chart conditions (slides 3 and 4), we used de-
mographic data at two different time-steps. These were per-
centages of the Syrian population affected by the story’s topic,
before the crisis began and today. We used such simple statis-
tics for three reasons. First, because we wanted aggregated
human-related data, i.e., non-identifiable data, that could easily
be visualized as groups of individuals (e.g.,
100% =100
viduals). Second, because we believed using only proportional
data—instead of using proportions and absolute numbers—
would help avoid possible confounds linked to the proportion
dominance bias. Third, because demographic data are very
common, and should be easily understood by everyone. We
used data at two different times steps to ensure the blame for
the situations would be put on the Syrian conflict, not on the
affected populations. We expected this would leverage the
justice bias, and that it would ensure some kind of emotional
response, as the differences are truly shocking.
A Design Space for Anthropographics
We created eleven anthropographic designs and a baseline pie
chart for each story using the data described above. These
forty four anthropographics were the basis for our design
space. Although we never intended to test all of them—our
goal was to assess whether anthropographics generally have
an effect on empathy and donating behavior, not to test for
most effective designs—and although we do not claim they
are exhaustive, we wanted to get a sense of the creative pos-
sibilities. Establishing this design space also allowed us to
select the most “representative” designs for our experiments
(as done in [15]). We found inspiration in [2, 16, 45, 46],
and varied our designs according to:
class of visualization
human shape,unit labelling, and unit grouping.
Unit Aggregate
Figure 1. Two classes of visualization: unit and aggregate.
Class of visualization:
we created unit and aggregate visual-
izations (Fig. 1). Unit visualizations show each row of a data
table using a single visual mark, or unit [23], while aggregate
visualizations show statistics. Although most anthropograph-
ics are unit visualizations (see [1]), and while anthropomorphic
aggregation is somewhat of an oxymoron, we found examples
of aggregate designs in humanitarian reports (e.g., [3]).
Human shape:
we designed a series of pictograms, varying in
realism and expressiveness, to apply to both unit and aggregate
visualizations. According to Waytz et al., anthropomorphiza-
tion depends on the attribution of humanlike physical features,
like a face and hands; and of humanlike mental capabilities,
like intentions or emotions [65]. The prior can be mapped to
the realism of the pictogram, while the latter to it’s expressive-
ness. This creates a simple
space (Fig. 2), which can be
segmented for simplicity along the realism axis into groups
Generic Iconic Unique
Figure 2. Depicting a migrant: from abstract to realistic, and from neu-
tral to expressive; grouped into generic,iconic, and unique pictograms.
of generic,iconic, and unique pictograms, according to their
congruence with a unique individual.
Internally displaced. This 10-year old is
internally displaced.
Tariq is internally
Generic Iconic Unique
Figure 3. Three ways of labelling units: generic,iconic, and unique.
Unit labelling:
we also created generic,iconic, and unique
labels (Fig. 3) to make abstract and neutral pictograms more
evocative for first-time viewers [34]. Generic labels provide no
personal information, and no sense of individualness. Iconic
labels provide some demographic information, and a general
sense of individualness. Unique labels provide names, and
true individualness. In our experiments, we used popular Arab
first names as proxies for real names, since we chose to avoid
re-identifiable data for ecologic validity.
Grid Organic
Figure 4. Two ways of grouping units: grid and organic.
Unit grouping:
we created grid and organic groupings for
the unit visualizations (Fig. 4). Grid groupings are visually
clear, but they are somewhat unrealistic regarding the way
people form groups in real life. Organic groupings are less
clear, but more realistic. Ensuring the perception of distinct
groups is important, as Bartels & Burnett have found that
group construal,i.e., the way multiple entities are perceived
as either a number of individuals or a single group, can have
an effect on the proportion dominance bias [5].
We measured participants’ empathic concern, personal distress,
perception of story-protagonists’ responsibility, and percep-
tion of justice of donating for each HR narrative on 7-point
scales. We also collected qualitative feedback through free-
from text inputs. For empathic concern and personal distress,
we used Batson’s list of emotion adjectives [7]. We asked par-
ticipants to report how strongly they felt sympathetic,moved,
compassionate,tender,warm, and softhearted for empathic
concern; and alarmed,grieved,upset,worried,disturbed,per-
turbed,distressed, and troubled for personal distress. We
randomized the order of appearance of these adjectives for
each participant. For responsibility, justice, and donation like-
lihood, we used a series of questions extracted from [43]. We
also asked participants a topic availability question to check
whether they knew about each topic beforehand. We refer to
all these measurements as the empathy and biases scales.
We then evaluated prosocial behavior through donation likeli-
hood (on a 7-point scale), and a fictional donation allocation
procedure. This provided a realistic scenario for the HR narra-
tives, as most organizations conduct fundraising campaigns to
support populations in distress, or to advocate for their rights.
We told participants that “someone” was willing to make a $10
donation in their name, and that they could allocate the money
as they liked. We believed this would amplify the moral bias,
since participants would not have to consider donating their
own money. We initially asked them to determine how they
would split the $10 between stories, but we later changed this
to force a dichotomous choice. Finally, we asked participants
to briefly explain their choice.
General Procedure
We conducted our experiments on Amazon Mechanical Turk
(AMT). Upon accepting the Human Intelligence Task (HIT),
and providing consent through AMT, participants were di-
rected to an external webpage, which welcomed them with a
short introduction. A single button allowed them start the first
narrative. After reading it, they were asked to fill out a first set
of empathy and biases scales, and to indicate their donation
likelihood. They then moved on to the second narrative, after
which they were asked to fill out another set of scales. In the
end, they were given the possibility to read through any of the
two narratives again, and as many times as they liked. They
were also prompted to make the donation allocation. Once the
HIT was completed, they were paid $.30.
In this section, we present a pre-study we conducted to se-
lect the pair of least diverging topics for our two narratives
(similarly to what was done in [50]). These had to be differ-
ent, but at the same time provoke similar affective responses
when presented using the same baseline chart condition, so
as to minimize possible story-topic-related confounds when
comparing different chart conditions in our experiments.
Design Specificities
Our pre-study consisted of six pairwise comparisons of each
of our four stories. We used the general materials and methods
described in the previous section, but we restricted the chart
Pair 05
Pair 04
Pair 03
Pair 02
Pair 01
Pair 06
Figure 5. Mean donation allocations for each pairwise comparison.
conditions to the same standard pie chart (no anthropographic).
We also used the ‘split’ donation allocation procedure, as we
believed it would yield finer-grained results.
For each pairwise comparison, we recruited 50 participants
located in the United States who were required to have a 99%
acceptance rate on AMT. We rejected the data of 16, due
to incomplete or irrelevant responses (e.g., participants who
simply answered e.g., “brilliant” when asked to describe their
impressions about the stories). This resulted in a total of
participants (= 155, = 129).
All the analyses and discussions in this article are based on
estimation, i.e., point estimates with 95% confidence intervals
(95% CI). The 95% CI are based on
percentile boot-
strap replicates of the parameter of interest. This complies
with the recommendations put forward in [4, 20, 22].
We first inspected the differences in donation allocations
(Fig. 5), as this should be the most immediate indicator for
diverging topics. We then compared aggregated responses to
the empathy and biases scales for each story (Fig. 6).
Results of the donation allocations clearly show that
was the most divergent topic. There is good evidence in Fig. 5
that participants were willing to allocate more money to sup-
port access to clean water than to any other cause.
indeed considered a more vital need, as participants typically
explained that “having water is more crucial right now be-
cause it is essential to survive,” “no one can survive without
water,” or “the water crisis is more likely to cause deaths and
is therefore the more immediate need.” This illustrates how
rational participants’ decisions to donate generally were. We
expected this might be different in our experiments, since the
anthropographics should elicit more emotional responses.
That said, empathic concern and personal distress were gen-
erally high for each story (
, see Fig. 6); the perception of
responsibility was low, and the perception of justice very high.
This simply indicates that: 1) HR narratives about suffering
743 5 621
Personal Distress
Empathic Concern
Likelihood Education
0 %
Topic Availability
Figure 6. Aggregated mean responses to the empathy and biases scales.
children do indeed elicit empathy; 2) children are not per-
ceived as responsible for their plight; and 3) people think it is
just to help these children with donations. Donation likelihood
however, was intriguingly consistent, considering how much
more money people ended up allocating to the
This discrepancy leads us to believe that donation likelihood
was probably more affected by the moral bias [43], i.e., by
individual differences in the perceived importance of fairness
or justice [57], than the actual allocations, which were more
objective and rational. This could explain the lower reliability
of the donation likelihood results (shown by the wider error
bars [41]). Meanwhile, topic availability was clearly separated
into two groups:
were well known topics,
while education and water were much less so.
Based on these results, we selected the
ries for our experiments. We excluded the
story because
of its highly divergent effect on donation allocations. We then
decided to leave out the
story because the topic was
much less known than the other two. Although it is possible
that higher topic-awareness may reduce the effect of presenta-
tion (e.g., of chart design) on empathy—as people may already
have other pre-stored (or available) empathy-arousing images
in mind—we preferred to use the two stories that had the most
similar levels of topic availability.
In this section, we describe our three initial experiments, in
which we compared different anthropographic designs with
the pie chart baseline. We used the stories selected in the pre-
study, and successively tested the three designs we considered
most representative of the eleven we created. Our hypotheses
were based on the anthropographic assumption:
the anthropographic should elicit more empathy than
the standard pie chart (regardless of the story it is set in) [36,
42, 55, 66]; and therefore
the anthropographic should encourage more prosocial
behavior [30, 43, 64], which should be more emotionally-
driven than in our pre-study.
While piloting these experiments, we realized that the ‘split’
donation allocation procedure often led to even distributions
($5 for each story). This is consistent with the findings in [64],
which show that empathic concern usually drives people to
split resources between multiple charities, rather than giving
everything to a single charity. While such fine-grain results
were useful in our pre-study, and although they suggested early
on that we might fail to confirm
, we wanted to encourage
more emotionally-driven decisions. Therefore, we changed
the procedure to force a dichotomous choice: participants
would have to allocate all $10 to a single story.
We also noticed that very few participants hovered over the
units to display the labels. This meant they did not see them.
As labels were an important attribute of our anthropographic
designs, we added a Suggested Interactivity cue [15] that
sequentially displayed the labels of random units. That way,
participants would see the labels, even without interaction.
Experiment 1
In our first experiment, we tested a unit anthropographic orga-
nized in organic groupings, in which units had unique shapes
and labels (the
unique individuals design
). We chose to test
it first, as we believed the uniqueness and higher realism of
the units would most effectively tap into participants’ expe-
riential knowledge of postures and attitudes [49, 52], thus
facilitating empathy-arousal through mimicry and direct asso-
ciation [38], and ultimately helping to reproduce the persona
and/or identifiable victim effect (as did the “fully expressive”
character in [44]). We expected this would elicit higher lev-
els of empathy, and therefore would attract more donations
). We used our general materials and methods,
and included the minor adjustments mentioned above. We
recruited 50 participants on AMT who had not participated
in our pre-study, setting the same requirements. We rejected
the data of 2 who performed the HIT twice. This resulted in a
subset of 48 participants (= 23, =25).
This time, we first inspected responses to the empathy scales
between chart conditions (Fig.7). We then checked for possible
interactions between empathy and participant variables. We
separated participants by gender, then by topic availability
(Fig.8). We expected people without prior knowledge of the
topics might be more affected by the mode of presentation.
We then looked at donation likelihood between chart condi-
tions (Fig. 9). Here too, we checked for possible interactions
with participant variables. Although we do not show the re-
sults for responsibility and justice here due to space limitations,
743 5 621
Personal Distress
Empathic Concern
Figure 7. Mean responses to the empathic concern and personal dis-
tress adjectives between chart conditions. For each emotional compo-
nent, mean responses in the anthropomorphized data graphic condition
are shown on top; those for the pie chart baseline are shown below.
743 5 621
Personal Distress
men women
Empathic Concern
743 5 621
Personal Distress
available not available
Empathic Concern
Figure 8. Mean responses separated by gender and topic availability.
these were generally consistent with those of our pre-study. As
we believe it is clear people generally perceive that children
are not responsible for their plight (responsibility bias), and
that it is just to help them (justice bias), we do not mention
these possible biases any further. Finally, we inspected the
results of the dichotomous donation allocations.
, 95%
CI [30.7%, 57.7%] of participants allocated the $10 to the
narratives using the anthropographic.
Our results do not support
. Fig. 7 shows that the
individuals design
and the standard pie chart had very similar
effects on viewer’s empathic concern and personal distress.
The results are highly reliable (short error bars), and the fact
that there is some difference between these emotional compo-
nents (independently of chart condition) is a good indication
that participants did not respond randomly—as the list of adjec-
tives we used to assess both feelings was randomly ordered for
each participant. In addition, while Fig. 8 shows no evidence
of an interaction between gender, chart type, and empathy, it
shows that women were generally more empathic than men.
This is consistent with [26]. Fig. 8 also shows no real evidence
of an interaction between topic availability, chart type, and
empathy. The only slightly perceptible effect of the
individuals design
is on the difference in personal distress
between people who knew the stories beforehand and people
who did not. However, it is possible that this is confounded by
743 5 621
available not available
available not available
total population
men women
Figure 9. Mean donation likelihood.
other mediating factors like gender or story—as there was al-
ready evidence in Fig. 6 of a very slight difference in personal
distress between the IDP and poverty stories.
is not supported either. Fig. 9 shows no evidence of a
difference in donation likelihood between chart conditions,
whatever the participant subgroup. That said, the reliability
of these results is quite low (wide error bars), which comforts
our idea that there are individual differences. In addition, the
unique individuals design
attracted less donations than the
pie chart—although the evidence is weak. Once again, par-
ticipants explained their donation allocations very rationally,
which suggests that neither the anthropographic, nor the forced
dichotomous choice truly encouraged them to act on emotions
induced by the mode of presentation. Only one participant
mentioned that “the first slideshow [which used the
individuals design] was more moving.
Overall, our results suggest that the
unique individuals de-
did not reproduce the expected persona and/or identifiable
victim effect. Upon reflection, we considered this may have
been due to a semantic incongruence between the type of data
and their representation: the anthropographic showed abstract
statistics, i.e., normalized proportions, as groups of unique
individuals, not as “abstract statistics.” Although we had not
anticipated that this might have an effect, since it is relatively
common practice (see [1]), participants may have considered it
deceptive [51], and their feelings may have been confounded.
Experiments 2 and 3
In our second and third experiments, we attempted to reduce
the semantic incongruence of the
unique individuals design
by lowering the uniqueness of each anthropographic. We ex-
pected this might positively impact participants’ empathy and
prosocial behavior. In Exp. 2, we tested a unit anthropographic
organized in grid groupings, in which units had iconic shapes
and generic labels (the
iconic individuals design
). In Exp. 3,
we tested an aggregate anthropographic with an iconic shape
iconic statistic design
). Although the efficiency of this
latter design is doubtful from a visual perception perspective,
it had the advantage of being the least semantically incongru-
ent of all our designs. For both experiments, we used the same
materials, methods, and hypotheses as in Exp. 1. We recruited
50 new participants for each, and retained the work of
Exp. 2 (
= 19,
= 27), and of
in Exp. 3 (
= 19,
= 26).
We conducted exactly the same analysis as in Exp. 1. All re-
sults for the empathy scales and donation likelihood for Exps. 2
and 3 are respectively shown in Figs. 10 and 11.
, 95%
743 5 621
Personal Distress
available not available
total population
men women
Figure 10. Results for Exp. 2: the iconic individuals design.
CI [22.6%, 49.2%] of participants in Exp. 2, and
, 95%
CI [37%, 65%] in Exp. 3 allocated the $10 to the narratives
using the anthropographics.
Although we see a higher variability within participant sub-
groups and chart conditions (Fig. 10), our results generally do
not support
in either experiment. The anthropographics
and the standard pie chart had similar effects on empathic con-
cern and personal distress. In fact, our results even show very
slight negative trends. We believe these are likely the result of
individual differences in empathic disposition [24]—people
may be more or less empathic in general. There is also no real
evidence of interactions between gender or topic availability,
and chart type and empathy. That said, once again women
were generally more empathic than men, and the effect of
availability seems confounded by other mediating factors.
is not supported either. Donation likelihood is generally
consistent between chart conditions in both experiments—
even though the reliability of these results is still low. In
addition, there is some evidence that the
iconic individuals
actually attracted less donations than the pie chart; and
there is no real evidence that the
iconic statistic design
tracted more. Explanations were again very rational, even if
people reason differently, e.g., “I wanted to make sure the chil-
dren could find homes. I figure this could also help children
living in poverty so it’s a win win,” or “I think that giving them
homes is more of a short term measure. Permanently getting
these children out of poverty would be best for the long term.
Initial Experiments’ Discussion
These initial results fail to validate the anthropographic as-
sumption. Empathic concern, personal distress, and donation
likelihood were very similar between chart conditions, and
across experiments. Donation allocations however, were some-
what inconsistent. Even our prime candidate, the
unique indi-
viduals design
, seems to have failed to reproduce the persona
743 5 621
Personal Distress
available not available
total population
men women
Figure 11. Results for Exp. 3: the iconic statistic design.
and/or identifiable victim effect. Although we had speculated
on the possible confounding effect of a semantic incongru-
ence after Exp. 1, Exps. 2 and 3 consistently failed to show
a difference when trying to reduce that incongruence. Upon
further reflection, this failure to reproduce persona and/or
identifiable victim effect may be because (at least) the latter
is assumed to be linked to face recognition mechanisms in
the brain [30] (the human shapes in our designs had no faces).
Another possible confound could be that our choice of unit
labeling in the
unique individuals design
suffered from the
familiarity bias [38]. The unique labels showed Arab names,
so our American participants may have perceived the children
in the stories as out-group members, which could have lim-
ited the effect of the anthropographic. Nevertheless, empathy
levels were high in all three experiments (
), which leads
us to believe the baseline level of empathy elicited by our HR
narratives (independent of chart conditions) may have reached
a threshold. This would suggest that the anthropographics
showed no effect simply because there was no room for that
effect to be visible.
In this section, we present four follow-up experiments, in
which we adjusted different aspects of our general materials
and methods, in an attempt to lower the baseline level of
empathy elicited by our HR narratives. Although this made
the narratives somewhat less ecologically valid, we hoped it
would accentuate the effect (if any) of the anthropographics.
Design Specificities
As discussed at the end of our pre-study, the high awareness
of the
topics may have reduced the effect of
presentation on empathy. In Exp. 4, we replaced the stories
we had previously used with the
The standard pie chart may have elicited unforeseen affective
responses, which may have emphasized its effect on empathy.
In Exp. 5, we compared the pie chart with text alone. The fact
that the varying chart conditions were only a relatively small
part of the data-driven stories (2/5 slides) may have favored
the text as a reminder of the human dimension of the data. In
Exp. 6, we stripped the stories to keep only the slides with
the varying charts conditions (slides 3 and 4). Finally, Borkin
et al. have found that people are generally more attracted to
textual elements in visualizations, and that e.g., titles have
a significant effect on what they take away [13]. The word-
ing of the text, and the focus on children may have been so
emotionally charged that they overshadowed the effects of the
anthropographics. In Exp. 7, we again restricted the stories to
slides 3 and 4, and we removed all references to children, and
re-edited the text to make it shorter and dryer.
Considering the familiarity bias, we created a modified version
of the
unique individuals design
for Exps. 4 and 6: the
familiar individuals design
. We replaced the unique labels
with iconic labels, changing the Arab names to age identifiers.
To comply with our re-editing of the text in Exp. 7, we also
created the
generic individuals design
, using generic labels.
We kept the rest of the materials, methods, and hypotheses the
same as before—with a slight exception for Exp. 5, in which,
for consistency, we hypothesized that the pie chart should
elicit more empathy and encourage more prosocial behavior
than text alone. We recruited 50 new participants for each
experiment, and retained the work of
in Exp. 4 (
= 22,
= 27); of
in Exp. 5 (
= 24,
= 16); of
in Exp. 6
(= 25, = 22); and of 47 in Exp. 7 (= 21, = 26).
We first compared the aggregated levels of empathy for each
follow-up experiment with those of our initial experiments
(Fig. 12) to determine whether our adjustments had an effect
on the overall level of empathy—and thereby on the baseline
level of empathy. We then inspected responses to the empathy
scales and donation likelihood between chart conditions. Due
to space limitations, we do not detail the results for Exps. 4
and 5 here. Fig. 12 shows that the adjustments we made in
these experiments failed to lower the baseline level of empa-
thy, and the different chart conditions once again showed no
evidence of differences
. Results for Exps. 6 and 7 however,
are respectively shown in Figs. 13 and 14, as the adjustments
in these clearly lowered the baseline (Fig. 12). For simplicity,
we do not show the gender differences or topic availability
here. We have found that the prior simply highlights a gen-
eral gender bias in empathy-arousal [26], and that the latter
is likely confounded by other mediating factors. Finally, we
inspected the results of the donation allocations.
, 95%
CI [37.2%, 64.7%] of participants in Exp. 6, and
, 95%
CI [47.4%, 74.2%] in Exp. 7 allocated the $10 to the narratives
using the anthropographics.
Despite our adjustments, Exps. 4 and 5 clearly failed to lower
the baseline level of empathy (
, see Fig. 12). This suggests
that neither the topics, nor the standard pie chart were respon-
sible. In addition, empathic concern, personal distress, and
donation likelihood in these experiments were once again very
Results for Exps. 4 and 5 can be found in the supplemental
material—along with those of all other experiments.
Aggregated Levels
of Empathy
743 5 621
Exp. 3
Exp. 4
Exp. 5
Exp. 6
Exp. 7
Exp. 2
Exp. 1
Figure 12. Aggregated levels of empathy for each experiment.
743 5 621
Personal Distress
Empathic Concern
Figure 13. Results for Exp. 6: stripped stories.
similar between chart conditions; and the donation allocations
showed no real evidence of an effect of chart conditions.
Exps. 6 and 7 however, did lower the baseline (
). This
suggests that the extra contextual information and the word-
ing were likely responsible for the higher levels of empathy.
While we expected lowering the baseline would accentuate
the difference between chart conditions, Figs. 13 and 14 show
it did not. Although there is a very slight trend in favor of the
anthropographic in Fig. 13, the evidence is weak, and it is not
confirmed in Fig. 14—where the aggregated level of empa-
thy is lowest. Likewise, donation allocations show that the
anthropographics attracted slightly more money than the pie
chart in both experiments, but the evidence is not very strong.
In addition, we believe all the results of the donation allo-
cation procedure should be considered with caution, as they
may have been confounded by an anthropomorphic carryover
effect. Participants may have automatically transferred the
human dimension of the data from the narrative with the an-
thropographic to the one without, as donation decisions were
made at the end, after having read through both narratives.
Thus, once more our results generally do not support
. We conclude that a tragic HR narrative simply is a tragic
HR narrative. Removing content and making the text dryer
may affect readers’ empathy, but presenting it with anthropo-
graphics instead of standard charts makes no real difference:
empathic readers will inevitably feel for the suffering protago-
nists. This generally complies with the concept of narrative
empathy [40], and indicates that when the graphics are part of
an emotive narrative, they can afford to be abstract.
All our results invalidate the anthropographic assumption, at
least in the specific narrative context in which we tested it.
Anthropographics and standard charts had very similar effects
743 5 621
Personal Distress
Empathic Concern
Figure 14. Results for Exp. 7: stripped and reworded stories.
on participants’ empathy, as well as on their prosocial behavior
in all our experiments. Although it may be argued that the
emphasis we put on being ecologically valid (in the structure,
content, and wording of the HR narratives) may have initially
confounded our results, Figs. 13 and 14 suggest it did not.
Lowering the impact of other materials did not promote the an-
thropographics, which leads us to believe that when narrative
empathy [40] is high, charts can afford to be abstract.
Participants generally reported a high donation likelihood
5). Interestingly though, as their empathy decreased, so
did this likelihood. This concurs with the idea that donation
likelihood is mediated by empathy [43, 64]. However, when
it came to allocating donations, their decisions were mostly
rational. Although we expected the anthropographics would
encourage more emotionally-driven decisions, we failed to
find this true. We relate this to the results of [30], which show
that abstract silhouettes and names have a lesser effect than
photos on viewers’ tendency to donate to victims.
More generally, our results concur with a number of previous
findings. In addition to the mediating role of empathy on do-
nation likelihood [43, 64], we found that women are generally
more empathic than men [26]; that empathic concern and per-
sonal distress are related [7]; and that there are likely some
individual differences in dispositional empathy [24]. This pro-
vides strong external validity, while the overall consistency of
our results provides good internal validity.
However, although our results do not show a higher positive
effect of anthropographics on empathy and prosocial behavior
compared to standard charts, they do not show a negative effect
either. In other words, although anthropographics were not
truly beneficial in our experiments, they were not detrimental.
We stress this, because in more realistic settings where people
may not be as biased toward information as in a controlled
experiment, anthropographics may be useful for e.g., attracting
readers’ attention, and for drawing them into the narrative [35].
This might ultimately be useful for eliciting empathy. As such,
we consider our results an initial assessment of the empathic
and prosocial power of anthropographics. They complement
Bateman et al.’s call to learn more about the effects of differ-
ent types of visual embellishments in charts [6], and open a
number of exciting avenues for future work on the anthropo-
graphic assumption. Typically, focusing on other subjects than
the highly mediatized Syrian crisis may yield different results.
Expanding our design space for anthropographics to more
detailed drawings with e.g., faces (like in [18, 45]), or even to
photos (when possible) may facilitate reproducing the persona
and/or identifiable victim effect, whereby the more lifelike
renderings should have a stronger positive effect on viewers’
affect [44]—although they may also increase the semantic in-
congruence between graphics and data if the prior do not map
unique human shapes to unique individuals. Animating the hu-
man shapes may help too. Using more subtle cues like visual
metaphors [12] (as done in [53]) or motionscapes [29] may tap
deeper into viewers’ experiential knowledge [49, 52], which
may further facilitate empathy-arousal through mimicry and
direct association [38]. Adding cinematographic and sound
effects, like the dramatic zooming-in-and-out and powerful
voice-over narration used in [33] might also increase affective
responses. In contexts where disaggregated data may be more
readily usable, lower data-granularity may contribute to further
reducing the semantic incongruence, which may positively im-
pact viewers’ perception of the uniqueness of units. Similarly,
increasing the amount of units to show absolute values (instead
of normalized values) may provide more concrete scales [19],
i.e., scales that are easier to relate to, which may also help
viewers better grasp the magnitude of certain HR tragedies (as
done in [61]). Finally, further tweaking our general materials
and methods may lead to finer results, especially regarding
donating behavior. A different, between-subjects procedure
may remove the possible confounds of the donation allocation,
and changing the fictional allocation to a real donation (simi-
larly to what is done in [30]) may incite participants to think
differently about how they allocate the money.
In this article, we have presented a design space for anthropo-
graphics, and a series of experiments, in which we compared
various anthropographic designs set in human rights narratives
with a standard chart, to investigate the growing assumption
that because anthropomorphized data graphics create an im-
mediate visual connection between abstract data and actual
people [36], they are better suited than standard charts for
eliciting empathy and prosocial behavior [42, 66]. Contrary
to this assumption and our expectations, we have found that
both types of visualizations have very similar effects. This
initial assessment suggests that when the graphics are part of
a broader emotive narrative, they can afford to be abstract.
However, while this seems to imply that anthropographics
are not truly beneficial for eliciting empathy and prosocial
behavior, we stress it does not mean they are detrimental. We
strongly believe our results should not discourage further ex-
ploration and use of such graphics. We personally intend to
extend our design space, and we have proposed a number
of possible avenues for future work, which, although we ac-
knowledge are not exhaustive, contribute to the elaboration of
a research agenda for anthropographics, and ultimately to an
agenda for better understanding the effects of different types
of visual embellishments in information visualization.
This research was supported by a John D. And Catherine T.
MacArthur Foundation grant.
1. A Collection of anthropographics.
2. A Hommage to Gerd Arntz.
3. World Humanitarian Data and Trends., 2015.
4. American Psychological Association. The Publication
manual of the American psychological association (6th
ed.). Washington, DC, 2010.
5. D. M. Bartels and R. C. Burnett. A group construal
account of drop-in-the-bucket thinking in policy
preference and moral judgment. Journal of Experimental
Social Psychology, 47(1):50–57, 2011.
6. S. Bateman, R. L. Mandryk, C. Gutwin, A. Genest,
D. McDine, and C. Brooks. Useful Junk?: The Effects of
Visual Embellishment on Comprehension and
Memorability of Charts. In Proceedings of the SIGCHI
Conference on Human Factors in Computing Systems,
CHI ’10, pages 2573–2582, New York, NY, USA, 2010.
7. C. D. Batson. Prosocial Motivation: Is it ever Truly
Altruistic? volume 20 of Advances in Experimental
Social Psychology, pages 65–122. Academic Press, 1987.
8. C. D. Batson. The Altruism Question: Toward a Social
Psychological Answer. L. Erlbaum, 1991.
9. C. D. Batson. Empathy-Induced Altruistic Motivation.
Prosocial motives, emotions, and behavior: The better
angels of our nature, pages 15–34, 2010.
10. D. Bischof-Köhler. The development of empathy in
infants. In Infant Development: Perspectives from
German Speaking Countries, chapter 12, pages 245–273.
Lawrence Erlbaum Associates, Inc, 1991.
11. P. Bloom. Against Empathy. http:
// against-empathy,
12. R. Borgo, A. Abdul-Rahman, F. Mohamed, P. W. Grant,
I. Reppa, L. Floridi, and M. Chen. An Empirical Study on
Using Visual Embellishments in Visualization. IEEE
Transactions on Visualization and Computer Graphics,
18(12):2759–2768, Dec 2012.
M. A. Borkin, Z. Bylinskii, N. W. Kim, C. M. Bainbridge,
C. S. Yeh, D. Borkin, H. Pfister, and A. Oliva. Beyond
Memorability: Visualization Recognition and Recall.
IEEE Transactions on Visualization and Computer
Graphics, 22(1):519–528, Jan 2016.
14. M. A. Borkin, A. A. Vo, Z. Bylinskii, P. Isola,
S. Sunkavalli, A. Oliva, and H. Pfister. What Makes a
Visualization Memorable? IEEE Transactions on
Visualization and Computer Graphics, 19(12):2306–2315,
15. J. Boy, L. Eveillard, F. Detienne, and J. D. Fekete.
Suggested Interactivity: Seeking Perceived Affordances
for Information Visualization. IEEE Transactions on
Visualization and Computer Graphics, 22(1):639–648,
16. C. Burke, E. Kindel, and S. Walker. Isotype: Design and
Contexts 1925–1971. Princeton Architectural Press, 2014.
17. A. Cairo. Empathy and Visualization.
empathy-and- visualization.html, 2016.
18. M. Chalabi.
19. F. Chevalier, R. Vuillemot, and G. Gali. Using Concrete
Scales: A Practical Framework for Effective Visual
Depiction of Complex Measures. IEEE Transactions on
Visualization and Computer Graphics, 19(12):2426–2435,
20. G. Cumming. The New Statistics: Why and How.
Psychological science, 25(1):7–29, 2014.
21. M. H. Davis. Measuring individual differences in
empathy: Evidence for a multidimensional approach.
Journal of Personality and Social Psychology,
44(1):113–126, 1983.
22. P. Dragicevic. Fair Statistical Communication in HCI. In
Modern Statistical Methods for HCI, pages 291–330.
Springer, 2016.
S. M. Drucker and R. Fernandez. A Unifying Framework
for Animated and Interactive Unit Visualizations.
Technical Report MSR-TR-2015-65, 2015.
24. C. J. Einolf. Empathic concern and prosocial behaviors:
A test of experimental results using survey data. Social
Science Research, 37, 2008.
25. N. Eisenberg. Empathy-Related Responding: Links with
Self-Regulation, Moral Judgment, and Moral Behavior.
Prosocial motives, emotions, and behavior: The better
angels of our nature, pages 129–148, 2010.
26. N. Eisenberg, A. Cumberland, I. K. Guthrie, B. C.
Murphy, and S. A. Shepard. Age Changes in Prosocial
Responding and Moral Reasoning in Adolescence and
Early Adulthood. Journal of Research on Adolescence,
15(3):235–260, 2005.
27. N. Eisenberg and P. A. Miller. The Relation of Empathy
to Prosocial and Related Behaviors. Psychological
Bulletin, 101(1):91–119, 1987.
28. J. Emerson. Ten Challenges to the Use of Data
Visualization in Human Rights.,
B. L. G. D. Feng, C. Beyond Data: Abstract Motionscape
as Affective Visualization. Leonardo/ISAST, 2014.
30. A. Genevsky, D. Västfjäll, P. Slovic, and B. Knutson.
Neural Underpinnings of the Identifiable Victim Effect:
Affect Shifts Preferences for Giving. The Journal of
Neuroscience, 33(43):17188–17196, 2013.
C. C. Gould. Globalizing Democracy and Human Rights.
Cambridge University Press, 2006.
32. T. Grodal. Visual Communication as Textsigns and as
Embodied Mental Processes. In Nordic Visual
Communication Conference, Oslo, 1994.
33. N. Halloran. The Fallen of World War II.
34. K. Haramundanis. Why icons cannot stand alone.
SIGDOC Asterisk J. Comput. Doc., 20(2):1–8, May 1996.
35. S. Haroz, R. Kosara, and S. L. Franconeri. Isotype
visualization: Working memory, performance, and
engagement with pictographs. In Proceedings of the 33rd
Annual ACM Conference on Human Factors in
Computing Systems, CHI ’15, pages 1191–1200, New
York, NY, USA, 2015. ACM.
36. J. Harris. Connecting with the Dots. https:
// dots/,
37. M. A. Harrison and A. Hall. Anthropomorphism,
Empathy, and Perceived Communicative Ability Vary
with Phylogenetic Relatedness to Humans. Journal of
Social, Evolutionary, and Cultural Psychology, 4(1),
38. M. Hoffman. Prosocial Behavior and Empathy:
Developmental Processes. In N. J. S. B. Baltes, editor,
International Encyclopedia of the Social & Behavioral
Sciences, pages 12230–12233. Pergamon, Oxford, 2001.
K. Jenni and G. Loewenstein. Explaining the Identifiable
Victim Effect. Journal of Risk and Uncertainty,
14(3):235–257, 1997.
40. S. Keen. Narrative Empathy. http://wikis.sub., 2013.
41. M. Krzywinski and N. Altman. Points of Significance:
Error bars. Nature Methods, 10:921–922, 2013.
42. S. Lambert. And What Do I Do Now? Using Data
Visualization for Social Change.
org/2016/01/data-visualization- for-what/, 2016.
43. S. Lee, K. P. Winterich, and W. T. Ross. I’m moral, but I
won’t help you: The distinct roles of empathy and justice
in donations. Journal of Consumer Research,
41(3):678–696, 2014.
44. J. C. Lester, S. A. Converse, S. E. Kahler, S. T. Barlow,
B. A. Stone, and R. S. Bhogal. The persona effect:
Affective impact of animated pedagogical agents. In
Proceedings of the ACM SIGCHI Conference on Human
Factors in Computing Systems, CHI ’97, pages 359–366,
New York, NY, USA, 1997. ACM.
45. S. McCloud. Understanding Comics. A Kitchen Sink
book. HarperCollins, 1994.
46. S. McCloud. Making Comics. HarperCollins, 2011.
N. M. McDonald and D. S. Messinger. The Development
of Empathy: How, When, and Why. In A. A., L. J. A, and
S. J. J., editors, Free Will, Emotions, and Moral Actions:
Philosophy and Neuroscience in Dialogue, pages
341–368. IF-Press Morolo, 2011.
48. J. Miksatko, K. H. Kipp, and M. Kipp. The persona
zero-effect: Evaluating virtual character benefits on a
learning task with repeated interactions. In International
Conference on Intelligent Virtual Agents, pages 475–481.
Springer, 2010.
49. S. E. Moriarty. Abduction: A Theory of Visual
Interpretation. Communication Theory, 6(2):167–187,
50. A. V. Pandey, A. Manivannan, O. Nov, M. Satterthwaite,
and E. Bertini. The persuasive power of data
visualization. Visualization and Computer Graphics,
IEEE Transactions on, 20(12):2211–2220, 2014.
51. A. V. Pandey, K. Rall, M. L. Satterthwaite, O. Nov, and
E. Bertini. How Deceptive Are Deceptive Visualizations?:
An Empirical Analysis of Common Distortion
Techniques. In Proceedings of the 33rd Annual ACM
Conference on Human Factors in Computing Systems,
CHI ’15, pages 1469–1478, 2015.
52. C. S. Peirce. Peirce on signs.
53. Periscopic. U.S. Gun Deaths.
54. L. D. Riek, T.-C. Rabinowitch, B. Chakrabarti, and
P. Robinson. How Anthropomorphism Affects Empathy
Toward Robots. In Proceedings of the 4th ACM/IEEE
International Conference on Human Robot Interaction,
HRI ’09, pages 245–246, 2009.
55. J. Schwabish. Policyviz Podcast Episode #31: Rees &
Mushon on DataViz Empathy. on-dataviz- empathy/,
56. E. Segel and J. Heer. Narrative Visualization: Telling
Stories with Data. IEEE Transactions on Visualization
and Computer Graphics, 16(6):1139–1148, 2010.
57. J. Shang, A. R. II, and R. Croson. Identity Congruency
Effects on Donations. Journal of Marketing Research,
45(3):351–361, 2008.
58. S. Slobin. What if the Data Visualization is Actually
what-if- data-visualization- actually- people/, 2014.
59. P. Slovic, M. L. Finucane, E. Peters, and D. G.
MacGregor. The Affect Heuristic. European Journal of
Operational Research, 177(3):1333–1352, 2007.
60. P. Slovic and D. Zionts. Can International Law Stop
Genocide when our Moral Intuitions Fail Us? In D. J.
Ryan Goodman and A. K. Woods, editors, Understanding
Social Action, Promoting Human Rights, pages 100–128.
61. The New York Times. Death in Syria.
middleeast/syria-war- deaths.html?action=click&
62. E. R. Tufte. The Visual Display of Quantitative
Information. Graphics Press, Cheshire, CT, USA, 1986.
63. A. Tversky and D. Kahneman. Availability: A heuristic
for judging frequency and probability. Cognitive
Psychology, 5(2):207–232, 1973.
64. G. A. Verhaert and D. V. den Poel. Empathy as added
value in predicting donation behavior. Journal of
Business Research, 64(12):1288–1295, 2011.
A. Waytz, J. Cacioppo, and N. Epley. Who Sees Human?
The Stability and Importance of Individual Differences in
Anthropomorphism. Perspectives on psychological
science : a journal of the Association for Psychological
Science, 5:219–232, 2014.
66. M. Zer-Aviv. DataViz—The UnEmpathic Art. https:
// unempathetic-art/,
67. M. Zer-Aviv. Responsible Data Forum: Visualization.,
... Storyline visualizations contain multiple storylines, thus multiple characters, illustrating their relationships. Other examples include visualizations with properties that make them more identifiable or personable, such as anthropographics [8,49]. These representations were shown to be effective in creating an emotional connection with underlying content [8,43], while other evidence revealed their limitations of eliciting a specific emotion from an audience [46]. ...
... Other examples include visualizations with properties that make them more identifiable or personable, such as anthropographics [8,49]. These representations were shown to be effective in creating an emotional connection with underlying content [8,43], while other evidence revealed their limitations of eliciting a specific emotion from an audience [46]. These works motivate a need for the data storytelling community to translate devices and structures from visual storytelling (i.e., film literature). ...
... A data character should consider properties that delve deeper into the communicative effort and that help the audience bridge the gap between science and the story. Thus, a foundation for the properties of a data character can be derived from works that address visualization design [8,49,61,63,74] and visual metaphors [13,21,22,45,51,54,55,69]. There is limited research that addresses the role of a data character in data storytelling. ...
Full-text available
Fig. 1: Character-oriented visual storytelling consists of three stages: story identification, organization, and presentation. A key task in story organization is to determine what main, supporting, antagonist characters, etc. are and their relations to the plot. Abstract-When telling a data story, an author has an intention they seek to convey to an audience. This intention can be of many forms such as to persuade, to educate, to inform, or even to entertain. In addition to expressing their intention, the story plot must balance being consumable and enjoyable while preserving scientific integrity. In data stories, numerous methods have been identified for constructing and presenting a plot. However, there is an opportunity to expand how we think and create the visual elements that present the story. Stories are brought to life by characters; often they are what make a story captivating, enjoyable, memorable, and facilitate following the plot until the end. Through the analysis of 160 existing data stories, we systematically investigate and identify distinguishable features of characters in data stories, and we illustrate how they feed into the broader concept of "character-oriented design". We identify the roles and visual representations data characters assume as well as the types of relationships these roles have with one another. We identify characteristics of antagonists as well as define conflict in data stories. We find the need for an identifiable central character that the audience latches on to in order to follow the narrative and identify their visual representations. We then illustrate "character-oriented design" by showing how to develop data characters with common data story plots. With this work, we present a framework for data characters derived from our analysis; we then offer our extension to the data storytelling process using character-oriented design. To access our supplemental materials please visit
... For instance, researchers have explored affective data visualization as a promising approach to communicating data to a wide audience [8,65]. In the wild, design projects such as the U.S. Gun Death [64] have sparked heated discussion on the role of visualization in evoking emotion and doing good for society [13]. ...
... Kennedy et al. [39] thought that the binary view of emotion and reason has viewed emotion as irrational and has made emotion intentionally undervalued in data science for a long time. In the wild, many practitioners have created emotion-laden visualizations and defended the necessity of communicating emotion with data [13,47]. However, so far, such arguments and practices remain highly scattered and have not been systematically reviewed by the visualization community. ...
... Emotion also influences people on a deeper level, such as changing attitudes, values, or behaviors. For example, Boy et al. [13] and Morais et al. [58] cited prior work from psychology and sociology to illustrate that emotions such as empathy can help people perceive others' misery and promote prosocial behavior. Lan et al. [47] used a set of serious data stories that convey negative messages about COVID-19 as stimuli and found that appropriate design methods helped evoke deep emotions and strengthened contemplative thoughts and self-reflection. ...
Full-text available
In recent years, more and more researchers have reflected on the undervaluation of emotion in data visualization and highlighted the importance of considering human emotion in visualization design. Meanwhile, an increasing number of studies have been conducted to explore emotion-related factors. However, so far, this research area is still in its early stages and faces a set of challenges, such as the unclear definition of key concepts, the insufficient justification of why emotion is important in visualization design, and the lack of characterization of the design space of affective visualization design. To address these challenges, first, we conducted a literature review and identified three research lines that examined both emotion and data visualization. We clarified the differences between these research lines and kept 109 papers that studied or discussed how data visualization communicates and influences emotion. Then, we coded the 109 papers in terms of how they justified the legitimacy of considering emotion in visualization design (i.e., why emotion is important) and identified five argumentative perspectives. Based on these papers, we also identified 61 projects that practiced affective visualization design. We coded these design projects in three dimensions, including design fields (where), design tasks (what), and design methods (how), to explore the design space of affective visualization design.
... Several papers have discussed broader issues around the visualization of data about people. The belief that data visualizations might be neutral has been disparaged by many: Visualizations are political [16], can promote empathy [7], often do not surface inherent uncertainty [24], and have even been called inhumane [17]. Beyond the visualization, data and its collection and processing are also regularly prone to biases. ...
... A self-perception of the individual in data visualization can also be achieved with anthropographics -data visualizations that use human-shaped graphical representation. Researchers studied such visualizations to elicit more empathy for human rights [7] and promote prosocial behavior [31]. Despite initial results that did not show clear benefits of anthropographics, this stream of research is growing. ...
This paper collects a set of open research questions on how to visualize sociodemographic data. Sociodemographic data is a common part of datasets related to people, including institutional censuses, health data systems, and human-resources fles. This data is sensitive, and its collection, sharing, and analysis require careful consideration. For instance, the European Union, through the General Data Protection Regulation (GDPR), protects the collection and processing of any personal data, including sexual orientation, ethnicity, and religion. Data visualization of sociodemographic data can reinforce stereotypes, marginalize groups, and lead to biased decision-making. It is, therefore, critical that these visualizations are created based on good, equitable design principles. In this paper, we discuss and provide a set of open research questions around the visualization of sociodemographic data. Our work contributes to an ongoing refection on representing data about people and highlights some important future research directions for the VIS community. A version of this paper and its fgures are available online at
... Writing about "empathetic emotions" is a challenge: sympathy, compassion, or empathy are often used as synonyms, which can lead to confusion [11]. So, why exactly is the rhetoric of pathos in data visualizations of importance at all? ...
... Research has shown that using such pictographic representations, i.e. anthropographics or isotypes, improves the understanding of the concepts presented, establishing a "connection between abstract data and actual people" while also improving the memorability of the message [11,76]. The "near and far" technique offers a micro and a macro view of the visualized data. ...
Aristotle has considered the art of communication as a balance of logos, ethos, and pathos. While in science, logos (reason) and, recently also, ethos (morality) are discussed as aspects not to be neglected, pathos (feeling) is seen critically. In this work, we take a historical perspective on pathos and weigh the pros and cons of applying this rhetorical concept to the field of data visualizations. To better understand data, connecting it to the human way of thinking is imperative - appealing to emotions is one building block. The theoretical and empirical basis originates from different scientific fields, like social sciences, economics, and humanities. Tangible techniques to target empathetic emotions in data visualizations are introduced, as well as other rhetorical devices, such as interactivity and contextual framing, are highlighted. Researching these different approaches can provide new insights regarding the creation and influence of empathetic emotions in data visualizations.
... (1) Effects Ajani et al. (2022), Bateman et al. (2010), Borkin et al. (2013), Bower and Clark (1969), Boy et al. (2015), Boy et al. (2017) ...
Conference Paper
Full-text available
With the proliferation of Business Intelligence and Analytics, data storytelling has gained increasing importance to improve communicating analytical insights to business users and support decision-making. While conceptual research on data storytelling suggests that these techniques can help improve decision-making, there is a lack of prescriptive knowledge on how to design data stories in Business Intelligence and Analytics. Moreover, it is not understood how data stories can facilitate effective use and support decision-making of business users. To address this challenge, we conduct a Design Science Research (DSR) project. Drawing on the theory of effective use and data storytelling techniques, we propose three design principles that we instantiate in a prototype. The results of two focus groups indicate that enhancing dashboards with data storytelling techniques increases transparent interaction and representational fidelity. Our DSR project contributes novel design knowledge for data stories that facilitate effective use.
... The vast design space for pictorial visualization allows for a great deal of creativity but also poses challenges. Some studies [6,32] focus on expanding the design space and propose various design dimensions as guidelines for practice. Recently, more efforts have been devoted to developing authoring tools for pictorial visualizations. ...
Full-text available
Pictorial visualization seamlessly integrates data and semantic context into visual representation, conveying complex information in a manner that is both engaging and informative. Extensive studies have been devoted to developing authoring tools to simplify the creation of pictorial visualizations. However, mainstream works mostly follow a retrieving-and-editing pipeline that heavily relies on retrieved visual elements from a dedicated corpus, which often compromise the data integrity. Text-guided generation methods are emerging, but may have limited applicability due to its predefined recognized entities. In this work, we propose ChartSpark, a novel system that embeds semantic context into chart based on text-to-image generative model. ChartSpark generates pictorial visualizations conditioned on both semantic context conveyed in textual inputs and data information embedded in plain charts. The method is generic for both foreground and background pictorial generation, satisfying the design practices identified from an empirical research into existing pictorial visualizations. We further develop an interactive visual interface that integrates a text analyzer, editing module, and evaluation module to enable users to generate, modify, and assess pictorial visualizations. We experimentally demonstrate the usability of our tool, and conclude with a discussion of the potential of using text-to-image generative model combined with interactive interface for visualization design.
Conference Paper
Full-text available
In this paper, we present an empirical analysis of deceptive visualizations. We start with an in-depth analysis of what deception means in the context of data visualization, and categorize deceptive visualizations based on the type of deception they lead to. We identify popular distortion techniques and the type of visualizations those distortions can be applied to, and formalize why deception occurs with those distortions. We create four deceptive visualizations using the selected distortion techniques, and run a crowdsourced user study to identify the deceptiveness of those visualizations. We then present the findings of our study and show how deceptive each of these visual distortion techniques are, and for what kind of questions the misinterpretation occurs. We also analyze individual differences among participants and present the effect of some of those variables on participants' responses. This paper presents a first step in empirically studying deceptive visualizations, and will pave the way for more research in this direction.
Motionscapes-the compositions of visual forms in motion- have often been used for the evocation of affects in recent interactive artifacts and environments. While the motionscape aesthetic can be informed by art theory and history, previous empirical work investigating the affective affordances of motionscapes brings new perspectives to the design language of motionscapes. The authors argue that motionscapes that are commonly employed in artistic contexts can be appropriated for the design space of human-computer interaction (HCI) as a rich modality for affective visualization. The authors propose an initial set of principles and guidelines for evoking affect through motionscapes in interactive and immersive environments.
To facilitate a multidimensional approach to empathy the Interpersonal Reactivity Index (IRI) includes 4 subscales: Perspective-Taking (PT) Fantasy (FS) Empathic Concern (EC) and Personal Distress (PD). The aim of the present study was to establish the convergent and discriminant validity of these 4 subscales. Hypothesized relationships among the IRI subscales between the subscales and measures of other psychological constructs (social functioning self-esteem emotionality and sensitivity to others) and between the subscales and extant empathy measures were examined. Study subjects included 677 male and 667 female students enrolled in undergraduate psychology classes at the University of Texas. The IRI scales not only exhibited the predicted relationships among themselves but also were related in the expected manner to other measures. Higher PT scores were consistently associated with better social functioning and higher self-esteem; in contrast Fantasy scores were unrelated to these 2 characteristics. High EC scores were positively associated with shyness and anxiety but negatively linked to egotism. The most substantial relationships in the study involved the PD scale. PD scores were strongly linked with low self-esteem and poor interpersonal functioning as well as a constellation of vulnerability uncertainty and fearfulness. These findings support a multidimensional approach to empathy by providing evidence that the 4 qualities tapped by the IRI are indeed separate constructs each related in specific ways to other psychological measures.
We are living in a time when the activation of mirror neuron areas in the brains of onlookers can be recorded as they witness another's actions and emotional reactions.1 Contemporary neuroscience has brought us much closer to an understanding of the neural basis for human mindreading and emotion-sharing abilities the mechanisms underlying empathy. The activation of onlookers' mirror neurons by a coach's demonstration of technique or an internal visualization of proper form and by representations in television, film, visual art, and pornography has already been recorded.2 Simply hearing a description of an absent other's actions lights up mirror neuron areas during fMRI imaging of the human brain.3 The possibility that novel reading stimulates mirror neurons' activation can now, as never before, undergo neuroscientific investigation. Neuroscientists have already declared that people scoring high on empathy tests have especially busy mirror neuron systems in their brains.4 Fiction writers are likely to be among these high-empathy individuals. For the first time we might investigate whether human differences in mirror neuron activity can be altered by exposure to art, to teaching, to literature. This newly enabled capacity to study empathy at the cellular level encourages speculation about human empathy's positive consequences. These speculations are not new, as any student of eighteenth-century moral sentimentalism will affirm, but they dovetail with efforts on the part of contemporary virtue ethicists, political philosophers, educators, theologians, librarians, and interested parties such as authors and publishers to connect the experience of empathy, including its literary form, with outcomes of changed attitudes, improved motives, and better care and justice. Thus a very specific, limited version of empathy located in the neural substrate meets in the contemporary moment a more broadly and loosely defined, fuzzier sense of empathy as the feeling precursor to and prerequisite for liberal aspirations to greater humanitarianism. The sense of crisis stirred up by reports of stark declines in reading goes into this mix, catalyzing fears that the evaporation of a reading public leaves behind a population incapable of feeling with others. Yet the apparently threatened set of links between novel reading, experiences of narrative empathy, and altruism has not yet been proven to exist. This chapter undertakes three tasks preliminary to the scrutiny of the empathy-altruism hypothesis5 as it might apply to experiences of narrative empathy, an argument I develop more fully in Empathy and the Novel (2007).6 These tasks include: a discussion of empathy as psychologists understand and study it; a brief introduction to my theory of narrative empathy, including proposals about how narrative empathy works; and a review of the current research on the effects of specific narrative techniques on real readers.