Listen to the CBC interview on "Quirks&Quarks".

1. Where did the idea of looking at the color of faces come from?

I have a long-standing interest in human high-level vision and, in particular, object recognition. My approach was/is motivated by the idea that our mental/neural representations of objects are strongly related to their original viewing conditions (when I started in the field this idea was rather more controversial than at present - back then many theories assumed our representations were much more abstract and exclusively shape-based). As an extension of this approach, I wondered whether surface properties played some role in object recognition (I still wonder about this). In the early 90's I regularly visited Tübingen Germany as a guest of Heinrich Bülthoff. Quite often, my friend and colleague, Dan Kersten (University of Minnesota), would also be in Tübingen. Dan is at the top of people I enjoy having a conversation with - one night over beers (this being Germany), we were talking about the role of surface properties and he commented that he bet that male and female faces were different with respect to color. His reasoning was that men have stubble and stubble creates a green cast to the face. When I returned home and did the analysis on a collection of faces, it turned out he was right and wrong. Yes, there were color differences, but men were on average more red and women were on average more green.

2. What about races other than Caucasians?

Really good question. We make the claim only for Caucasian faces because that was the only large collection of face images to which we had access and that were collected under controlled lighting conditions with no makeup. First of all, if I was to hazard a guess as to whether the sex-color difference is present in all races, I would say yes. Here is why. There is good evidence from evolutionary biology that women's faces are lighter than men's faces. Across ALL races. The idea is that women have to synthesize more Vitamin D and need to absorb more UV-B light, therefore they have less of some skin pigments (this is from work and an extensive review by Jablonksi and Chaplin - Jablonski, N. G., & Chaplin, G. (2000). The evolution of human skin coloration. Journal of Human Evolution, 39(1), 57-106.). So again, we know that, across all races, women have lighter skin than men of the same race. My hunch is that the skin pigment in question also has a reddish cast (the actual reflectance properties of human skin pigments is a complex topic and I haven't found a good reference that spells this out clearly), so having more of it, as in men, makes them more red. My understanding from reading about human skin reflectance is that without the particular pigment in question and other pigments, skin is kind of greenish.

In essence I am answering two questions here. First, we expect that this difference holds across races, but we can't prove it because there needs to be an equivalently large, controlled database of non-Caucasian faces. Indeed, we tried to build such a database (see face-place.org), but were unable, using the local university community, to recruit sufficient numbers of non-Caucasians. We tried. I wish we had been more successful. I realize this is a bit of an ivory tower response in that clearly there are many non-Caucasians in the vicinity, but we were not able to recruit those individuals. All I can say is we made an honest effort for several years, but that research project is coming to a close. Second, the reasons cited above for why I think the effect does hold across races also offers a physiological explanation for the effect. It is pure speculation, but it is based on solid research across many studies.

3. Is this true for kids' faces/When does this difference emerge?

Interesting, but hard, questions to answer. I am not an evolutionary biologist, so I can't really speculate a priori about this. These two questions are definitely answerable, but not without some serious effort in stimulus creation. That is, one would need fairly large sets of images of kids' faces collected under controlled lighting conditions. Let's 100 males/100 females for each age one wished to sample. I can imagine sampling faces for very young kids, around 3-6 months of age, two year olds, 4 year olds, 6 year olds, 8 year olds, 10 year olds, 12 year olds, 14 year olds, etc. Given this stimulus set, one could easily run the analysis to see at what age point the color difference became significant. Since I am not really in the business of collecting face images (although see face-place.org), I am unlikely to ever do this. But if someone did have access to said pictures, I would be happy to help them with the analysis. My best seat of the pant's guess is that kids' faces do not show this color difference (this is why we put girl babies in pink and boy babies in blue) and that the color difference emerges as one aspect of post-puberty sexual dimorphism. So probably around 14 years of age or so, boys are more red and girls are more green. But again, this is just a guess.

4. Isn't your subject population (3 subjects!) pathetically small?

The particular technique we are using in this paper - "superstitious perception" (Gosselin, F., & Schyns, P. G. (2003). Superstitious perceptions reveal properties of internal representations. Psychological Science, 14(5), 505-509.) - tends to rely on many many trials (+10,000) but, almost as a necessity, a small number of subjects. Here are several reasons why I think this small N still generalizes across people:

• We objectively know from analysis of this and other face sets that there is a consistent/reliable color difference between the faces of Caucasian men and women. So the information is there. All this technique does is ask whether observers in some sense "know" this and can exploit it. In this task there is no actual information about sex on each trial - just noise. So observers are free to impose whatever model, biases, and beliefs they have about what male and female faces look like. Our reconstruction of the male and female faces based on how they categorize the noise images just reveals those models, biases, etc. We find that observers are sensitive to the red:green distinction in the same direction as the true difference, that is, we end up with a reconstructed male face that was more red and a reconstructed female face that was more green - this is pretty telling.
• We have actually run other types of experiments that will appear in a different paper. For example, there is a high correlation between how red or green a given face is and the probability of a subject correctly judging that face's sex. That is, a very red male face is correctly classified as male more often than a greenish male face. To ensure that subjects were not using shape information, the faces were dramatically blurred so that color (and luminance) were probably the only available cues. Again, subjects seemed to be able to make use of this information. They must have "known" the fact of color differences between the sexes. We also ran an "It's Pat" study in which we created morphs between male and female faces so that the shape of each morphed face was ambiguous as to sex. We then cross these faces with more reddish and more greenish color tones. We found that the SAME faces was more likely to be classified as male when it was reddish and more likely to be classified as female when it was greenish.
• In essence our argument is simply that there is this objective fact about faces. Why wouldn't people use it?
  

5. Isn't the total number of faces you used in your analysis (200) too few to make any claims about faces in general?

We did try the objective analysis with other face image sets that were collected under controlled conditions (and in color). We obtained similar results, that is, on average, the males were more red and the females were more green. You can see point 2 for some of the issues involved in collecting a large number of face images under controlled conditions. But the real answer here is that the 200 faces gave us a hint that this fact was true across Caucasian faces generally. So we did the experiments reported in our paper and the others mentioned here in the previous point to see if people "knew" this fact. If our subjects do use color to tell male faces from female faces (and they did), then they must have learned this fact from their lifelong experience with faces. In essence this is much more telling than analyzing a much larger number of faces - we took people "off the street" (okay, off the pastoral Brown campus) and tested whether they thought (unconsciously) that men were more red and women were more green. Their latent knowledge of this fact tells us that the objective difference we saw in 200 faces is probably present in faces more generally.

6. My face, my mother's face, my best friend's brother's face, [insert face X here], seems to violate your claim...

None of our claims, both in terms of the color of a face and in terms of how people judge the sex of a face, apply to any given individual. Cognitive science/neuroscience is about identifying general principles about how we think, perceive, and interact with the world. In the same way that almost everyone has the same basic body architecture, we assume that almost everyone has the same basic mental/neural architecture. But people look different from one another and, similarly, they have different levels of mental abilities, strategies, etc. So you might be male and very green, you might never use the color of faces to determine sex (even if you are not colorblind), but that just means you fall somewhere on the normal distribution that exists for almost every biological variation that makes us all human. No big deal one way or the other.

7. Makeup?

There are really two questions here. First, could our effect be explained by makeup on the stimulus faces. The answer is no. The faces we analyzed and used as stimuli did not have any makeup on them. Indeed, rouge is red, so women wearing makeup might actually be redder. Which leads to... Second, why is rouge red and what does your finding tell us about the application/use of makeup. Here I have to veer into wild speculation mode. I have two guesses. It is possible that women apply rouge and other makeup in a manner that surrounds, but does not cover, the greenest regions of their face. In doing so (surrounding a green region with red) they make the green regions appear even greener (this is called color contrast). In particular, if they apply rouge on their top cheeks, where there is no red:green color difference, it might serve to increase contrast over the face and make females more separable from males [It is known that luminance contrast (lips, eyes) increases the attractiveness of female faces (Russell, 2003)]. I have no idea if applying rouge really increases red:green contrast, but it would be fun to test this in a controlled (if complicated) experiment. Alternatively, it is apparently a known fact in the perfume industry that women buy perfume, not men, so perfume is scented with male scents (Channel No. 5 is some sort of male cat hormone???). That is, perfumers make perfume attractive to the buyers - women - and don't worry about whether the perfume actually makes the wearer more attractive to the opposite sex (and there is no strong evidence that humans are sensitive to pheromones). See for example the recent popular book about smell by Herz. So maybe the same thing is happening with makeup - it is reddish because that makes makeup more attractive to most women (for rouge anyway; lipstick enhances the redness of lips, but women might want to emphasize their fuller lips).

8. What about colorblind people?

Right. See point 9 below. Colorblind people are fine. They have one less cue to the sex of an individual, but that isn't a big deal. Usually there are so many cues to sex, it is not an issue. Perhaps when dealing with "It's Pat"...

9. How important is this really? Isn't it a rather trivial finding?

Some media have played this up as a generically "important" finding. I am interested in how the human brain represents and recognizes objects and faces, so it is of interest to me and it certainly extends our understanding of the kinds of cues people can use in visual object recognition. Beyond that I think this particular result is simply kind of interesting and fun. Science should be interesting and fun even when it isn't earthshattering. I suspect it will also prove to be a useful fact for those in the business of building automatic face recognition systems. Moreover, it should be pointed out that there are many many cues to the sex of an individual - body shape, gait, face shape, etc. Our point was simply that in the absence of reliable face shape, color is somewhat diagnostic and that people seem to "know" this fact. Not that color supersedes body shape, face shape, hair style, etc.

10. Interesting factoid.

The ancient Egyptians traditionally represented men with red skin color and women with yellow. Apparently archaeologists always more or less assumed it referred to men working outdoors more than women, but on reflection that doesn’t work brilliantly given that they were mostly talking about elite men of the scribal class, not farmers. [per Laurel Bestock, Brown University, Archaeology]

[Bonus]. How far can the Steelers go in the playoffs with the anemic and inconsistent play calling of Bruce Arians?

oops..., wrong FAQ; but at this point (1 week before the 2009 superbowl), some comment is needed. The offense is playing somewhat better. I figure some of this is due to Parker being 100% and some of this is due to the fact that the D is basically not letting the other team's offense hold the ball. I also think I might have underestimated how much improvement there has been in the Steelers' defensive backs - kudos to Tomlin - I thought they were vulnerable to the long pass (and with Warner-Fitzgerald they might just be), but as of now there is simply not much weakness in this D. It isn't quite the force of nature that we saw in the Steel Curtain, but it is unbelievably consistent and opportunistic. I suspect this is due to Polamalu's and Smith's health. [Update: So they won. Barely. The D was great except for that fourth quarter meltdown. The offense was about what I expected (and the D delivered their customary 7 points), but interestingly as one of the linemen said "We are built for this". All season they played it close and pulled it out with a long drive at the end of the game. It was as if they needed to be forced to focus and run the no huddle. Whatever anyone says, Roethlisberger is a winner. He is becoming more and more Bradshaw-like - not always pretty, but almost always a winner. And Holmes became the big play, clutch receiver I never thought he would be. I think the pot-bust woke him up. Basically, this game was like every other game this season. As a final thought, Warner did fumble on the last play - you can only give the "arm going forward" interpretation so much leeway. And Harrison made the endzone. And the safety call on Hartwig was ticky-tacky and a terrible call by the refs in that game situation - it almost cost them the game - luckily they were bailed out by the comeback.