Other factors to control for include piercings, haircut, facial hair, glasses, makeup, any clothes in the picture (shirt/hat), hygiene, and tattoos.<p>However, it really wouldn't surprise me if some aspects of personality can, on average, be discerned from facial appearances. In hunter-gatherer days, being able to make an accurate snap judgment of a stranger's trustworthiness would be a big evolutionary advantage. And it wouldn't surprise me either if genes that affect one's personality (e.g. those relating to testosterone) affect one's appearance as well.
Preventing emotion from showing on your face is nearly impossible. If you try to hold a smile you don't actually feel, it is not difficult to others to notice intuitively. They may not realize exactly what aspects of your face didn't match up to the smile, but often they'll be able to observe that the eyes didn't look right.<p>So I'm not all that surprised, really. Ability to recognize and read faces, along with ability to recognize vocal stress, are both important skills for a human.
Ah this is truly a business opportunity - a nice supplement to your FICO score forget mechanical turk:<p>Step (1): Get access to frontal photographs of a large random sample of borrowers along with their financial history and records of any defaults.<p>Step (2): Extract a vector of facial features from each photo. Similar to the feature vectors used to judge attractiveness in the work of Cohen-Or et al. <a href="http://www.cs.tau.ac.il/~dcor/" rel="nofollow">http://www.cs.tau.ac.il/~dcor/</a><p>Step (3): Have users apply for a loan by submitting a mugshot. Use support vector regression or some other similar technique to predict the probability of default.<p>Step (4): Profit
Their test subjects came from Mechanical Turk. Is that considered reliable? I'm skeptical. And if they were using Mechanical Turk, why didn't they get 1000 subjects instead of just 25?
But do we react to the underlying physiognomy or the muscle tone (posture and expressions on their faces)? It would be hard to separate these two factors, but you could start by generating 3D faces with different physiognomy but the same neutral expression.
Makes sense:
* If I'm usually hiding something, my muscle memory will keep my face in that shape
* If I'm usually open, vice versa
Of course if I really believe stealing is good, my face will be open. So it's basically showing others what I believe about myself.
Can you really come to a conclusion based on inputs of just 25 MTurkers?
25 is less than 30 (the minimum required numbers to assume normal distribution) and also all being MTurkers is also not a good sign of unbiasedness.