Can Visualization Elicit Empathy? Our Experiments with “Anthropographics

One very interesting question that people in data visualization have been discussing for a while is whether visual depiction of data can in some way help convey (more) empathy to the reader.

That is: “Is it possible to increase empathy with data visualization?”.

The existing research on the empathic value of numbers and statistics paints a very dire picture. People do not seem to connect to human tragedies behind numbers and, even worse, the bigger the numbers the lesser the effect.

The excellent (and somewhat terrifying) research of Paul Slovic and his colleagues has repeatedly demonstrated that numbers and statistics do not work and that humans are affected by what has been called “statistical numbing”: our incapacity to wrap our head around numbers and the lack of empathy elicited by them.

Data visualization researchers and professionals have been debating for a while whether statistics, once they are depicted with evocative and engaging visual representations, may actually have more of a positive effect.

Some people seem to be super enthusiastic and optimistic about this idea, whereas others have expressed doubts and concerns (listen also the excellent PolycyViz podcast episode on the same topic), however most of the existing debate has been grounded exclusively on conjectures, with little support from scientific research.

In order to move first steps towards this direction our team at NYU (see more details about our team at the end of the article) has designed a number of experiments to check whether visualizing data in some specific formats may elicit more empathy in viewers.

Introducing “Anthropographics”

Studying empathy in visualization is much easier said than done, therefore we first had to figure out exactly which, among the many aspects of a visualization one can vary, we wanted to study. Inspired by the idea of “wee people”, we decided to create and explore in a systematic way the design possibilities and came up with “Anthropographics”: visual strategies to make the connection between data and the humans behind them more direct and, hopefully, more empathic. The design space for anthropographics is defined by the following elements.


Class of Visualization: Unit or Aggregate.

Data can be visualized at different levels of granularity. One basic decision a designer can make is to visualize every single data items with a dot or an icon or to abstract away from the units by calculating an aggregate measure of some sort.

One basic intuition here is that when data represent people, especially people involved in some kind of human tragedy, it is useful (more respectful even) to represent them in single units to emphasize the idea that behind each single dot there is a human life. A similar argument has been espoused by a popular essay by Jake Harris titled “Connecting with the Dots”, in which Jake argued that “from a distance, it’s easy to forget the dots are people” and that it’s very important for visualization to keep this issue in mind when creating data visualizations.

Realism and Expressiveness

A second aspect to take into consideration is how realistic a given icon to represent a human is. The icon can have different levels of realism (a dot vs. a human figure) and expressiveness (an icon showing a migrant with a bag on his head). Again, the basic intuition here is that more expressive and realistic depictions should elicit more empathy in the reader.


Of course the way one labels the icons may also have an effect on how people perceive them. Here one can use more generic or more specific, and thus evocative, labels to more strongly connect the data with the actual human being behind them.

Unit Grouping

Finally, the units can be organized in a rigid grid or in a more organic fashion. Rigid grids are less realistic, whereas more organic structures like the one of the right are more evocative of the kind of gatherings humans would create in reality and as such they may be an additional instrument to connect the dots to their human dimension.


So, going back to the original question: “is there a way to make data visualization more empathic?”. More precisely, do the strategies outlined above actually lead to a stronger empathic response? This is what we set out to discover with some experiments.

(Note: I am going to describe the experiments removing lots of important details. If you want to know more take a look at the paper.)

All of our experiments are based on a set of human rights stories (on poverty displacement, access to water, and education) and data we generated together with our rights experts. All the experiments were designed to find an effect of various conditions on empathic concern, personal distress and prosocial behavior, technical terms that translate respectively to amount of empathy felt, how bad one feels about a given situation or story, and donations allocated to a given condition by the reader. To simplify the exposition in the following I will focus exclusively on empathic concern. All of these measures turned out to vary in the same direction across the conditions we tested.

Our work included a total of 7 experiments.

Exp. 1, 2, 3: Comparing anthropographics to a pie chart condition.

The first three experiments aimed at studying variations over the design space of anthropographics we outlined above to see if elements like granularity, level of aggregation, type of labeling and realism had an effect on the outcome. For instance, we compared a pie chart baseline to a version with unit anthropographics organized in an organic layout with unique shapes and labels. The results however did not show any effect of these design elements. The chart below shows the results of the first experiment. As you can see did not find much of a difference between the conditions we tested and all the other experiments look mostly the same: no major effect found.

After dwelling over this first set of experiments for a while we figured that the stories we used elicited already a lot of emotions and also that most of these emotions are probably already heightened by the accompanying text we used to describe the stories (all stories about human tragedies). In a way these may have led to reaching an “empathy threshold” beyond which it is hard to see an effect of our conditions.

Exp. 4, 5: Changing stories and removing the graphics.

In the fourth experiment we substituted the stories on poverty and displacement with stories on water and education. In the fifth experiment we compared the pie chart condition with just text (using the previous stories). The results however did not show any major change. More precisely, the average overall empathy did not change, compared to previous experiments, and the difference between the conditions did not change either.

Exp. 6, 7: Removing narrative text from the stories.

In our experiments the various conditions have always been presented using a total of five slides in which two out of five contained the graphics we tested and the rest included only text to provide the necessary context for the story.

In this last set of experiments we asked: “What is going to happen if we completely remove the context and use exclusively the graphics? Will we finally be able to observe a difference between the conditions?”. To test this idea we kept only the two slides that contained the graphics and compared the pie chart baseline to two anthropographic combinations.

The results once again showed no difference between the graphics but they did show an overall decrease of empathy (as you can see in this chart), meaning that text did play a major role in the way people perceived a story emotionally.

Summary of results

In summary this is what we found:

  1. Anthropographics (using units and more evocative graphics) did not lead to more empathy than standard graphics.
  2. Removing the graphics and changing stories did not change the level of empathy.
  3. Removing the text reduced the level of empathy, but did not lead to a difference between the type of visualization used.

Reflections and future work

In our studies we found no evidence that presenting data with more evocative graphical formats leads to a stronger empathic reaction in the readers. At least not when the story is tragic.

This may seems at first really discouraging, maybe even disappointing, but let’s consider a few additional factors.

First, even though we did not find any positive effect of graphics on empathy we certainly did not find any negative effect either. Adding graphics in a data narrative may serve many additional important purposes other than eliciting empathy, above all gaining more clarity and trust. Furthermore, fancier graphics such as those proposed in the anthropographics framework may have a strong effect on attracting attention and, as such, they may be instrumental in encouraging readers to actually engage with the proposed content.

Second, despite our experiments explore a wide set of graphical methods and stories, they by no means exhaust all the interesting situations and conditions a designer may face in a project. In particular, our designs do not include some possibly very evocative methods such as the use of photographs and cinematic effects such as animations, transitions and sound. These may very well have a strong impact. I really hope someone may want to build upon our work and test additional conditions.

Finally, our results remind visualization designers how important having good text around a visualization piece is. People are heavily influenced by what they read and crafting good prose seems to be at least as important as crafting great visualizations, if not more. Let’s keep this in mind!

Our Team: This study has been developed as a collaboration between our team at NYU School of Engineering and our friends at NYU Law, who are experts in Human Rights. Most of the credit for this work goes to Jeremy Boy, who ideated the anthropographics framework, Anshul Pandey who helped enormously with the experiments, and John Emerson who provided critical information and data from the Human Rights side of things. The project was led by myself and my colleagues Meg Satterthwaite and Oded Nov.

The paper: This work has been recently presented at the ACM CHI 2017 conference. You can find full details on the experimental conditions and results here:

Showing People Behind Data: Does Anthropomorphizing Visualizations Elicit More Empathy for Human Rights Data?
Jeremy Boy, Margaret Satterthwaite, Anshul Vikram Pandey, Oded Nov, John Emerson, Enrico Bertini
Proc. of ACM CHI Conference on Human Factors in Computing Systems (CHI), 2017.

A single golf clap? Or a long standing ovation?

By clapping more or less, you can signal to us which stories really stand out.