(Chernoff)
Faces of
European Union
An experiment

Written by Matheus Freitag

Let's start with the end! In this article, we will explore how humans are amazing at recognizing faces, specifically facial expressions, and how we can use this ability to quickly interpret multiple domains of data by turning them into a facial expression. Based on country-level data from the 27 EU countries, we will use this technique to algorithmically produce a facial expression for each nation. Further explanations are provided in the following sections. The end result of this experiment can be seen in the map below.
Faces on the map are clickable.

Context

Data. One of the most trending words in modern times. If you ever dealt with data you probably realized that looking at tables full of number is not the most effective way to make sense of it. Humans, creative as we are, created a powerful concept to make data interpretation easier: Data Visualization.

Far from being a niche concept, data visualization examples like bar or line charts are massively broadcasted daily to a broad audience. Our capacity to abstract complex information into simple geometric elements like bars and lines is remarkable.

We have good results when interpreting simple geometric forms, but there is something humans are naturally better at recognizing: Faces.

1. Introduction

Humans are amazing at recognizing faces. In fact, we are so good at it that we see faces even when there aren't any. "Pareidolia" is the name given to the tendency we have to see patterns in ambiguous locations.

Faces, and more in-depth, facial expressions, were a life-saving ability during our evolution, so it is understandable that we can, so intuitively and quickly, recognize those who are a threat, a friend, someone who needs help, all with a glimpse at their expressions. These insights were brought by Herman Chernoff, a statistician who proposed in 1973 the concept of Chernoff Faces. The correct way to explain Chernoff Faces isa tool to display multivariate data in the shape of a human facewhich can be interpreted by us as "a technique in which multiple data entries are used to build a human face." The intuition is to explore our ability to recognize facial expressions and use it as a way to quickly interpret multiple domains of data. Here is where I fell into the rabbit hole and started to think about how I can apply this "Chernoff Faces" technique to something meaningful for me and others.

2. Motivation

Geopolitics is a personal topic of interest, and like many others, I like to keep up on how different nations deal with challenges and realities. I do not recall any integration happening at the moment that is deeper than the one performed by the nations within the European Union, so it is natural to use this as the scenario for my experimentations with Chernoff Faces. Here, I gathered multiple data sources on the national level and created a script to procedurally create Chernoff Faces for each EU country. My expectation is to quickly identify how each nation is performing on multiple fronts by the facial expressions that represent such nations.

3. Constraints

We have from 20 to 43 muscles in the human face (depending on the source and how you count those muscles). It is challenging to add 20+ attributes to a digital face, so in this context, a limited set of attributes was chosen: Eyebrows, nose, mouth, neck, and skin color. Eyebrows here can assume 3 different configurations, as well as mouth and skin color. Nose and neck can assume only 2 different configurations.

This small set of features and variations already allows for 108 different facial outputs, and from examples online you can easily find projects with other attributes being used, like face shape, eye variations, and many others. And all this assumes the face is symmetrical. For asymmetrical faces, you could potentially double the possible outcomes.

Personally, I think this is too much because, in the end, you still need to keep a mental map of what each attribute symbolizes. Too many attributes will naturally lead to confusion and a loss of this technique's power: transmitting information easily and instantaneously. It's important to notice that some features are more representative than others; for example, eyebrows and smiles yield more information and are first perceived than slight tweaks in the shape of the nose.

For this experiment, the number of features and variations were explicitly limited to refrain from an explosion of possible outcomes, avoiding the overload of our mental map.

4. Methodology

Multiple data domains were gathered in order to form the Chernoff Faces representing each EU country. The sources for those datasets will be linked in the sections below. For all cases, the quartile technique was used, separating the data points into 4 chunks with more-or-less equal sizes, where each chunk (or chunks) of data was (were) then mapped to a specific configuration of the face's attributes, as follows:

4.1. Perceived Corruption as Eyebrows

The "Perceived Corruption Index" is a metric measured by the Transparency International NGO widely used to measure the perceived corruption by the population. The index ranges from 0 to 100, where 0 means more corrupt, and 100 means not corrupt. For our Chernoff Faces, data was divided into its four quartiles.

Countries with perceived corruption index values below the 1st quartile receive a Relaxed Eyebrow, representing the least corrupted countries (as perceived by the population). Between the 1st and 3rd quartile, a Neutral Eyebrow is given. Finally, above the 3rd quartile, an Angry Eyebrow is assigned, indicating a high corruption index.

The 3 different variations possible for eyebrows, indicating how corrupt countries are perceived by their citizens.

4.2. Cocaine-related Death Rate as Nose

The European Union has public data on"Cocaine death rates" per 100,000 inhabitants. In this experiment, data was split into two quartiles, with the first quartile being mapped to a smaller nose, while the second quartile was mapped to a bigger nose.

The 2 different variations possible for the nose, indicating death rate caused by cocaine abuse.

4.3. Happiness Index as Mouth

The World Happiness Report is a well-known source of cross-country data and research on self-reported life satisfaction. The concept behind this measurement is called the Cantril Ladderindex to evaluate how happy citizens consider themselves to be. If you imagine a ladder with steps numbered from 0 at the bottom to 10 at the top, you can see it as 10 being you in your best life possible, and 0 being your worst outcome.

Based on each country's self-report, a smile was introduced to the face. The smile is happy for countries with a high happiness index. A neutral smile is given to countries with a medium happiness index, and a sad mouth is given to countries with a low happiness index.

The 3 different variations possible for the smile, indicating how happy countries see themselves.

4.4. Obesity Index as Neck

The World Health Organization provides data on worldwide obesity index among the population, measured as a percentage of the overall population. For this project, data on EU countries were gathered and split into two quartiles, with the first quartile being mapped to an absent neck, while the second quartile was mapped to a prominent neck.

The 2 different variations possible for the neck, indicating obesity index in the country.

4.5. UV Exposure as Skin

The World Health Organization also provides data on the average UV Radiation daily exposure in each country, measured in J/m². For this project, data on EU countries were gathered and split into three quartiles. Countries with an exposure lower than the value for the first quartile received a lighter skin tone. Countries with exposure between the 1st and 3rd quartiles received an in-between skin tone. Countries above the 3rd quartile, which means they receive more UV exposure on average, received a darker skin color.

The 3 different variations possible for the skin tone, indicating the exposure to UV Radiation daily.

5. Conclusion

It is fascinating how fast we can perceive a facial expression, and generating faces to communicate data is a powerful tool that I plan to explore more. During this project, some key constraints evolved:

• It is all about the mental map. Each extra attribute, and each variation of this attribute, creates stress on the amount of information a user needs to keep in their minds. Be mindful of which attributes you want to choose.
• Try to map meaningful data to equivalent facial attributes. UV radiation is well known to tan human skin, so it is logical to use this data to lighten or darken the skin tone. The same can be said about obesity and neck prominence, or happiness and smiles.
• From a UX perspective, if you display a map with Chernoff Faces, even with intuitive attributes, it is interesting to offer the option to check what each attribute means.

The result, with each EU country holding its own Chernoff Face, can be seen in the Result section at the top of this page.