May 06 2011

Judging by appearances: sometimes it works

Category: Evolution,Evolutionary Psychology,law,policy,PsychologyJames @ 7:37 am

We’ve all heard that you shouldn’t judge a book by its cover. But a recent psychology study indicates that we may be able to make accurate judgments about certain aspects of people’s character “after minimal exposure to [their] physical appearance.” The study is titled “The Accuracy of Inferences About Criminality Based on Facial Appearance”1 and was performed by Jeffrey M. Valla, Stephen J. Ceci, and Wendy M. Williams of Cornell University and published in the Journal of Social, Evolutionary, and Cultural Psychology.

I learned about it from a blog post by Satoshi Kanazawa titled Criminals Look Different From Noncriminals.2 Kanazawa says the following about the study:

[C]ontrary to popular belief, you can assess people’s character and personality by simply looking at them. Nice people look nice, and nasty people look nasty, and it appears that humans have innate psychological mechanisms to tell them apart. Now, . . . a truly groundbreaking study, recently published in the Journal of Social, Evolutionary, and Cultural Psychology . . . show[s] that people can tell criminals and noncriminals apart simply by looking at their still photos. Criminals, it appears, look different from noncriminals.

In their experiments, [they] show pictures of the faces of 32 young Caucasian men in their 20s, without scars, tattoos or excessive facial hair, all in neutral expressions. Sixteen of them are convicted criminals, and the other sixteen are not. . . . [They] ask their experimental participants to indicate how likely they think it is that each man is a certain type of criminal . . . on a 7-point scale . . . .Their results from two experiments consistently show that individuals can tell who is a criminal and who is not, by indicating that they believe the actual criminals have higher probability of being a criminal than actual noncriminals.

However, their results also show that individuals cannot tell what type of criminals they are. . . . [C]riminals do not specialize. Men who commit one type of crimes are more likely to commit other types of crimes. . . . In empirical reality, there are men who commit (all types of) crimes, and there are men who do not. And Valla et al.’s experiments show that individuals can tell them apart because the two types of men look different.

There is one seemingly anomalous finding in their paper. In both experiments, women are unable to spot rapists. Women consistently rate convicted rapists to be less likely to be criminal than not only other types of criminals but noncriminals as well! While this may be initially puzzling, upon further reflection, it makes perfect sense. . . . In order to be a successful rapist, the man has to be able to fool the woman and earn her trust initially. Men who ‘fit the bill’ by looking like a rapist or otherwise criminal and dangerous would not be able to do that. They would not be able to get close enough to the women to rape them. This may be why women, but not men, are unable to spot rapists, even though women are equally good as men at spotting other types of criminals.

The photos and answer key are at the bottom of this post. Test yourself and see how well you do. My average rating for a non-criminal was 3.625, whereas my average rating for the criminals was 5.125. As the averages indicate, I was able to correctly categorize the faces most of the time.

Of course, there are other possible explanations why study subjects were able to differentiate between criminals and non-criminals. The sample size of people rating the photos was small (44 people), and as with many psychology studies, the study subjects who were rating the photos were college students, and were thus not a representative sample of the American public, let alone of humanity. Also, there may be subtle differences between the photos. Since the non-criminal photos were all taken as part of the same photo database; the offender photos appear to be mugshots from criminal offender databases. It could be that people were picking up on subtle cues stemming from the photos’ different origins, such as the quality of the photo, different lighting, different cameras, etc. The authors tried to measure and eliminate these extraneous factors. They asked the study participants if they thought it was obvious that certain photos were mugshots; none of the participants thought so. The authors of the study also tried to equalize the photos by removing the background and editing the photos to “maintain a consistent photo quality, and remove differences in lighting, graininess, photo quality, etc.” They also controlled for the level of attractiveness for each person.

It seems that the study authors tried to do all they could to account for all of the extraneous factors to remove, or control for, possible reasons for bias. I’d like to see the results of more studies like this to find out if they yield consistent results (the authors do cite a few old studies which had similar results). This study, though, at least offers some compelling results which we should consider.

I disagree with Kanazawa’s summary of the study when he makes the absolutist statement that “[n]ice people look nice, and nasty people look nasty, and it appears that humans have innate psychological mechanisms to tell them apart.” I had false positives and false negatives. I rated 6 of the 16 non-criminals as looking like they were criminals (meaning I gave them a rating of 5 or higher, on the 7-point scale); I also rated two of the non-criminals as neutral (a 4 on the 7 point scale), meaning that I thought it was equally likely they could be criminal or non-criminal. I rated two of the criminals as looking like they were non-criminal (rating them  a 3 or lower on the scale), and one criminal as being neutral. I identified all of the rapists as looking criminal. Taking the null hypothesis3 to be that a person is a non-criminal, I had therefore had 6 false positives and 2 false negatives.

Presumably, the evolutionary psychological explanation for our ability to categorize people by appearance is that our capacity to make visual distinctions between the criminals and non-criminals evolved so that we can protect ourselves. People with criminal tendencies pose a greater threat to our safety. We will thus naturally be more wary of someone who we suspect of having criminal tendencies, and take extra steps to protect ourselves. It makes sense, then, that that sort of capacity to differentiate would be biased in favor of false positives — protecting oneself from being harmed by a criminal would confer a greater evolutionary benefit than falsely believing someone to be a criminal since the harm from being victimized by a criminal would be more likely to cause death or serious injury than wrongly believing a non-criminal  to be a criminal. Since death or serious injury at the hands of a criminal would make it difficult or impossible to pass on one’s genes to the next generation, a capacity to tell the difference between criminals and non-criminals would likely evolve in favor of making false positives rather than false negatives.

What are the implications for this research? In day-to-day life, it probably means that you should put greater trust in your visual evaluation of someone’s danger to you. It also means that women shouldn’t trust their ability to tell whether a man poses a risk of sexually assaulting them, and take precautions accordingly.

More importantly, it means that juries in criminal trials are probably subconsciously biased against men who look “criminal.” It is difficult to remove such cognitive biases that are rooted in our biology and evolutionary history. Remember that in my case I incorrectly judged over one-third of the non-criminals to be criminal. Innocent people shouldn’t go to jail just because they look like a criminal. I think that one of the big implications for this research is that we need to study how we can make changes to the criminal justice system to make sure that juries aren’t unfairly prejudiced against defendants just because they look criminal.

And now, here are the pictures for you to rate yourself. Rate each person on a 1 to 7 scale, with  1 being people who look the least criminal and 7 being people who look the most criminal. The answer key is at the end. If you try it yourself, please post in the comments 1) your average rating for non-criminals vs. criminals; 2) how many false positives and false negatives you had; 3) whether you classified the rapists as looking criminal or not.


Answer key:
Non-Criminal: 1, 2, 6, 7, 9, 12, 13, 14, 15, 17, 18, 19, 22, 25, 26, 30;
Criminal: Arson – 5, 10, 16, 20; Assault – 4, 24, 27, 28; Drug Dealing – 8, 11, 21, 29; Rape – 3, 23, 31, 32


The images and answer key come from Jeffrey M. Valla, Stephen J. Ceci, and Wendy M. Williams, “The Accuracy of Inferences About Criminality Based on Facial Appearance,” Journal of Social, Evolutionary, and Cultural Psychology. They are reproduced in conformity with  the fair use exception in U.S. copyright law, 17 U.S.C. § 107.



1 Jeffrey M. Valla, Stephen J. Ceci, and Wendy M. Williams, “The Accuracy of Inferences About Criminality Based on Facial Appearance,” Journal of Social, Evolutionary, and Cultural Psychology