Faces can be judged instantly

Alexander Todorov, assistant professor of psychology and public affairs and one of the two authors of the study, garnered much attention in 2006 when he was able to use his work on facial inferences to accurately predict 68.8 percent of the outcomes in U.S. gubernatorial races without looking at the candidates’ backgrounds or track records. He demonstrated that the competence perceived in the faces of the candidates contributed to biases in voting choices.

Motivated by the importance of facial evaluation in predicting social outcomes, Todorov and Nick Oosterhof, a research specialist at the Center for the Study of Brain, Mind and Behavior, are researching facial evaluation across multiple traits.

“Accurate perceptions of emotional expressions and the dominance of [members of the same species] are critical for survival and successful social interaction,” according to the study, published in an August issue of the Proceedings of the National Academy of Sciences. “In the absence of clear emotional cues broadcasting the intentions of the person, we argue that faces are evaluated in terms of their similarity to expressions of anger and happiness in an attempt to infer the person’s intentions.”

“We started with a set of traits that people used to describe a set of faces,” Todorov said. “Using computer modeling, we were able to develop a 2D model of face evaluation.” Each face was originally represented in models with as many as 12 variables before trustworthiness and dominance were isolated to construct the final two-variable model.

Next, Todorov and Oosterhof generated 300 random faces and asked people to judge the trustworthiness and dominance of each face, which allowed them to determine which features were specific for these traits.

One challenge that researchers faced was obtaining truly consistent results from trial participants.

“People themselves are not highly consistent,” Todorov said. Observers became more consistent when they viewed the same sample set a large number of times and refined their own rating system.

“If you ask a person to rate many faces, they may have different ratings for later observations [of the same face] compared with earlier ones,” he added.

The final step involved using computer modeling to generate new faces that were varied along the dimensions of trustworthiness or dominance.

A very trustworthy face usually has an almost surprised look, while an untrustworthy face has eyebrows slanted inward and an upside-down-U-shaped mouth. Mature-looking faces are interpreted more as dominant, while a babyish face is linked with submissive behavior.

Todorov explained that one of the successes of the research was the finding that people over-generalize based on their preliminary facial evaluations of others and make unsubstantiated conclusions. These early evaluations may be discarded, however, once additional informational about a person is realized.

“If people get to learn behaviors associated with the faces, such as this guy is a thief or this guy exhibits trustworthy behaviors, then the judgment changes very quickly,” Oosterhof said. “I’m kind of glad it is that way because otherwise people could have no control over how they are judged.”

Get the best of the ‘Prince’ delivered straight to your inbox. Subscribe now »

With such results in mind, Todorov said that he takes precautions to avoid biases when evaluating his students. “As a professor, I always try to do blind grading because it’s more fair,” he said. “Everyone thinks that they are objective but it’s not about what you think that matters.”

Oosterhof said that the evidence is very mixed about whether people with certain facial structures actually develop behaviors that match the way they are perceived by strangers.

“Many studies report no association between real behaviors and inferred traits,” he said. “There is no clear evidence for most general traits.”

For specific traits, however, behavior and inferred traits may be related. For example, some research indicates that testosterone can affect both facial structure and aggressiveness of behavior.