> I used no one I thought was under 20 and no one I thought was over 50. I used any internet source I could to get about 15-20 pictures each composite. The hardest thing is finding that many pictures of good frontal shots. Most people take pictures at an angle.
> It is also not always easy to find pictures of every nationality. At times I would use Google Image and type in the 10 most common last names for a nationality. For Latin America, where names are shared, I would use the most common last names for that country and put the country name in the search. I would try to verify that the person is at least in that country or from it (not easy). That was especially tricky with the UK (English, Welsh, etc) and Irish.
> Strangely for some nationalities it was easier to find men than women or vice versa. For example, very easy to find pictures of Italian men, but not women. It is very easy to find pictures of French women, but not men, etc.
> Anyway, for every composite I think I viewed at least a couple of thousand photos. :-(
The fact that some very individually distinctive features are still present on some pictures (why do some of these faces smile, and some don't?), I would guess that the number of pictures that was averaged was ridiculously low, 10 maybe.
The result is therefore all about the original ~10 pictures that were chosen... and it is certainly not an "average face" (which would require a statistically representative sample, north of 10k for most countries). Chose 10 other people from the same ethnicity and you get a different face.
This is incorrect, a sample of 10 is statistically representative but has a high margin of error. If you can live with a high MOE then you can use very small samples, often small studies of 10 are used in medicine to see whether further studies should occur.
Upon doing 5 minutes of research the data appears to come from FaceResearch which has many thousands of images on file, seems like any issues would be related to sample biases and not sample size.
10 is ok for measuring a single variable (although even then, you probably want something like 30, even in the social sciences).
But for faces, it is possible that there are large numbers of relevant variables (various features of the eyes, nose, etc. can all vary separately). That increases the sample size you need to get decent results.
You're allowed to reuse the same sample for multiple variables, pollsters do it all the time, they'll phone one person and ask them 10 different questions.
The fact that you're measuring faces as opposed to political leanings does not affect the underlying statistical methods. Confounding variables only come into play when you're doing regressions, and/or trying to prove causation. Since they aren't attempting to explain why the faces look like this there aren't any confounding variables.
That pollsters do something does not make it valid! ;) Yes, they do ask dozens of questions of a small number of people. But this is also mathematically dubious.
If you do multiple random samples, the chance at least one is way off increases. If you have a 95% chance of being close to correct on your small sample, and you do 20 samples, it is quite likely you are wrong on at least 1 of them.
My claim is that (1) there is a very large number of relevant facial features for the goal of the survey (find the "average" face), and (2) that being wrong on even a small number of them is significant. My basis for both is a combination of intuition and that we know humans are extremely good at perceiving faces and subtle facial differences. There are parts of the brain that are apparently adapted specifically to facial recognition, and disorders where facial recognition is impaired but nothing else, for example.
There's a good rebuttal to the critique in the comments of that story:
--- 8< ---
There is no ethnic cleansing. Some site on the net stole these from me, by not sourcing them to me. I created all of these and more (ones for men) right here: [pmsol3.wordpress.com]
My account on FaceResearch.org is Rasfarengi. I have e-mailed the editor about this already. The "South Africa" composite on here is labeled wrong, because who ever took it from my site did not label it correctly. It is actually the Argentina composite. The South African composite I did only uses Khoi-San and South Bantu populations found in South Africa, not whites or colourds.
I am African American, so there is no racism here. I didn't make specific composites of many African nationalities because it is just too difficult. My process requires front facing photos with limited smiles. I have to look through nearly 2,000 photos online to find good pictures of Italians. Let alone a Fin. In Africa it is just too hard outside of Nigerians and South Africa, because there are less pictures on the net, because those people don't have good internet access or access to digital cameras the way East Asians and Europeans do.
It is even hard to find pictures of Latin Americans of specific nationalities outside the larger nations.
I did one of "West Africans" specifically to see if I can tell how much Euro and Native American admixture changed the African American population, whose ancestors mostly originated from the areas between Senegal and Angola (with a very very small % from Mozambique) I tried to use the same % that came to America from specific regions, like the Blight of Biafa, during the slave trade. I think you can see a difference. I did do a separate South African and Ethiopian one. Doing specific tribes is too difficult.
- the average Golden Age male actor looks similar to his modern-day counterpart, maybe a bit older going by the receding hairline
- the Golden Age actress looks somewhat different from the modern - rounder eyes, somehow? Was that the preferred look for actresses of the era? (they doesn't look like averages, though...)
- the average porn actress is blonder than the celeb
Relatedly, it seems quite natural for average faces to be attractive - all the blemishes and asymmetries and extremes get evened out, leaving a very neutral look that most people don't have. Extremes are usually not attractive, after all.
TL;DR of the critique: These images are a blend of many "real" images, and these real images may have been sampled in a way that doesn't represent the country's real distribution of women.
I prefer to have no input from that site whatsoever. Particularly after I read them and their commenters celebrating an actual assault on someone who made a tasteless comment. They only care about victim blaming in specific circumstances, it seems.
They are not for equality but are instead for gossip-mongering and creating false controversy; just like the entire rest of the Gawker network.
The obvious take away is that good looking people are actually average looking (as opposed to median looking). The first comment on that blog is from the creator of the images and he explains some of the issues with the collection.
Slightly different meaning. The colloquial idiom of an "average" face is one that is unremarkable in a crowd, one of median attractiveness. Taking the average of all of the faces in that crowd results in a face that has median features, but measurably higher attractiveness. Reality can be counter-intuitive.
Right. The average of a number of samples can exhibit characteristics not present in each sample, and so it does here.
One characteristic often considered attractive in a woman's face is a rounded soft jawline. The average of a large number of samples will tend towards rounded soft features, just by the nature of averaging which blurs together any large outliers.
A starker example: People who weigh 80 pounds and people who weigh 240 pounds are both considered unattractive. But average a bunch of each and you get a perfectly attractive mean or median of 160.
But supposedly what's defined as average is someone who's features look like most other people. So by definition, the good looking should be a plurality in the population right? That's probably a good thing, otherwise most of the population would end up with people they think are ugly.
JERRY: Elaine, what percentage of people would you say are good looking?
ELAINE: Twenty-five percent.
JERRY: Twenty-five percent, you say? No way! It’s like 4 to 6 percent. It’s a twenty to one shot.
ELAINE: You’re way off.
JERRY: Way off? Have you been to the motor vehicle bureau? It’s like a leper colony down there.
ELAINE: So what you are saying is that 90 to 95 percent of the population is undateable?
JERRY: UNDATEABLE!
ELAINE: Then how are all these people getting together?
JERRY: Alcohol.
Also, if you believe OKCupid Trends, "2/3 of male messages go to the top 1/3 of women" and "women rate an incredible 80% of guys as worse-looking than medium": http://blog.okcupid.com/index.php/your-looks-and-online-dati... (Says nothing about what they think about the people they eventually end up with, of course.)
I always wondered how much of that effect was due to the smoothing out of irregularities in the skin that averaging produces. Reading http://www.uni-regensburg.de/Fakultaeten/phil_Fak_II/Psychol..., I get the impression that that is a main cause of the effect. I base that on the findings that
"the more attractive the original faces, the more attractive the resulting compound face (r=0.75 for female and r=0.68 for male faces). Thus, not simply the number, but also the attractiveness of the original faces influences the average attractiveness rating of compound faces. This result is in contrast to the 'attractiveness is averageness' hypothesis."
and
"Rather, it seems that not the proportions but the surface characteristics (e.g. skin texture) of a face decide wether it is regarded attractive or not. In terms of facial proportions, this means that the 'attractiveness is averageness' hypothesis is clearly falsified."
Edit: of course, further research is needed. For example, the subjects doing the ratings all did their shoppings in Regensburg. That's way better than the typical selection of around-20-years-old psychology students (at least 80% female) of late 20th century Western European upbringing, but it will have had an effect on what constitutes beauty.
I was contacted through some posts on Edward Tufte's forum by the admin of an anthropology forum about filtering for hate speech. Suffice to say, my big take-away after taking a look is that anthropology on the internet attracts a tremendous number of xenophobes, racists and bigots. That these folks are not known for their intellect, let alone intellectual integrity, is not surprising. My only regret is not knowing how to solve that problem other than massive education overhaul at the K-12 level.
Remember that this is not what the average female will look like in these countries. It's actually the physical average of the faces of a sampled group.
Yes, which can be misleading. In Japan for instance, there are three main body types and a narrow genetic variation (theorized because Japan may have been originally inhabited by a small group of shipwrecked sailors).
Averaging these three body types results in a picture of no-one in Japan.
Funny that China is the only country where they specify the ethnic group, even though many other countries in that list have sizable ethnic minorities.
91.5% of the people in China are Han, and the next biggest minorities (Zhuang and Hui) are very Han-like in appearance (the Hui are descended from Han who converted to Islam with a bit of Persian/Arab mixing). Consider that the visually distinctive Uighurs and Tibetans, which are like #5 and #9, barely add up to 1.25% of the total population.
China has "sizable" ethnic minorities when considering raw population, but as a percentage they aren't so sizable anymore; i.e. China is relatively homogeneous. Of course, the minorities tend to concentrate in areas where they are sizable, especially in the sparsely populated west. The problem is that the Han are so overwhelmingly large in number that all the other minorities are worried about encroachment and complete assimilation (even in the west).
Good point. I also don't know enough about India to comment on that. What I was pointing out in my post was a subtle bias towards overemphasizing China's multi-ethnic nature. China is not especially multi-ethnic, and many of the countries in that list have ethnic minorities of similar size to China. E.g. Afghanistan, Israel, Poland, Russia and South Africa (Khoisan vs Bantu).
The comments downthread that point out that this submission appears not to be from the original source (contrary to HN guidelines) are helpful, as are the comments describing the original source methodology and critiques of that methodology. I'm not at all sure that there is reliable data here.
That page needs to add an explanation. I've seen similar images before, but nearly every commenter on that site seems to think that they selected a bunch of really hot women to represent what the average woman should look like and posted the pictures to make people feel bad.
You can see that even picking to very lopsided in the same way face, like Right eye higher than left, and mouth off center, results in a face that looks good when averaged.
I found it interesting that you can see faint traces of where some of the sample images from Burmese/Myanmarese women were wearing thanaka. It seems like in a lot of other countries the sample images are likely to include that particular culture's makeup styles.
Funny not to capitalize country names, and why does India get a double entry? There are other places where geography correlates with different appearing faces too.
http://pmsol3.wordpress.com/2011/04/07/world-of-averages-eur...
> I used this site: http://www.faceresearch.org/demos/average
> I used no one I thought was under 20 and no one I thought was over 50. I used any internet source I could to get about 15-20 pictures each composite. The hardest thing is finding that many pictures of good frontal shots. Most people take pictures at an angle.
> It is also not always easy to find pictures of every nationality. At times I would use Google Image and type in the 10 most common last names for a nationality. For Latin America, where names are shared, I would use the most common last names for that country and put the country name in the search. I would try to verify that the person is at least in that country or from it (not easy). That was especially tricky with the UK (English, Welsh, etc) and Irish.
> Strangely for some nationalities it was easier to find men than women or vice versa. For example, very easy to find pictures of Italian men, but not women. It is very easy to find pictures of French women, but not men, etc.
> Anyway, for every composite I think I viewed at least a couple of thousand photos. :-(