Shiu Pei Luu, an art director at Facebook, and I were discussing character design — something she knows a lot about from her work on Facebook Stickers and developing games. Since it’s such a vital component of film/comics/cartoons/etc, a lot has been written about how to design characters. But we struggled a bit trying to define the space. Certainly, we could notice some similarities across characters, e.g. the introverted nerd type or the goofy sidekick type.
We wondered, is there a way to quantify characters so that we can compare or group them? Are there patterns in character personalities we can learn from successful shows? We decided to start by seeing if we could measure characters’ personalities.
Character Personality Surveys
We wanted to pin down the “space” of cartoon characters. Lots of dichotomies came to mind: Are they adventurous or mischievous? Helpful or adversarial? Outgoing or timid? It’s easy to list a bunch of traits but harder to put together a cohesive set of dimensions that capture the range of possible personalities. Fortunately people have worked on this exact problem for a long time, trying to measure and describe human personalities.
There are two main contenders in the personality measurement game: Myers-Briggs and the Big Five (or Five Factor Model). Each presents a respondent with a list of Likert-scale questions and then aggregates them into either a 4- or 5-dimensional personality profile. We chose to go with the Big Five because it was openly available and is more popular in the academic literature.
Crowdsourcing Personality Measurement
Obviously we can’t get these characters to take these personality tests for themselves, but it’s not difficult for people to answer questions on their behalf. You could even answer these questions about characters you haven’t even watched before, based on your first impression of their appearance!
We did a survey for Angelica (above) as a pilot. It worked!
We got results similar to what we’d expect: low agreeableness and high extraversion definitely fit what we’d expect. Almost every person who took the survey was familiar with Angelica, and indeed almost every character we did the assessment for. Turkers sure remember their cartoon characters!
So it cost 25 cents per response, and 9 responses makes a complete survey — so about $2.25 to get one character’s personality assessed. All told, we measured personalities for 27 characters. It’ll be easy to do more if we want.
Mapping the Characters into the Big Five
What’s neat about having measured personalities for all these characters is that we can flip the script and use them to describe the personality scale! Shiu Pei designed this Big Five visualization:
Clusters of Characters
We ran a hierarchical clustering algorithm over responses to the 44 questions to see if there were natural groupings.
In the dendrogram above, we see some neat pairings of similar characters: the introverts Helga and Daria, mischievous Homer and Bart, wild and unpredictable Dale and SpongeBob, silly sidekicks Patrick and Heffer, and the disagreeable Butthead and Roger Klotz.
We can also project the 44-question answers into a 2 dimensional space using t-SNE, a technique for visualizing high-dimensional data sets. The key idea behind this technique is that distance on the plot corresponds to distance in the survey answers about the characters — closer characters have more similar personalities.
We loved how intuitive this layout ended up being! And we’re looking forward to adding more characters to this personality space.
You can see two big clusters: the bottom left which contains mostly protagonists and the top right which has a mix of characters who are somewhat adversarial. There’s the goofy sidekick cluster (Stimpy, Patrick, Heffer) and we see that nerd cluster (Chuckie, Doug, Rocko, and Milhouse) we speculated about at the start of the project.
Your Most Similar Characters
If you’re interested, I can send you the cartoon characters with the most similar personality to yours! Take this Big Five survey and include your email address and I’ll send you the most similar and dissimilar characters to you! For example, here’s my report:
The space of cartoons. We didn’t accomplish our goal of understanding what groups of characters make successful shows. We can build on this measurement technique by treating shows as groups of character personalities and then trying to visualize and describe the space of shows.
Visual personality design guides. We can code character appearances (e.g. glasses, large eyes, small mouth) and see how they are correlated with our perceptions of their personalities. This is a little bit tougher to do with established characters because it’s difficult for us to separate their appearance from their behaviors on the show. But we could design new characters with different features and test who people perceive their personalities.