December 26, 2003
Racial Classification: How Easy?
I've been looking at Howells' Craniometric data for a while now. A question that always interested me is the following: "how easy is it to get a fairly good classification of a skull into a major continental race?" So, I tried a very simple experiment: I picked the medieval Oslo (Caucasoid) and Shang Dynasty from Anyang (Mongoloid) samples and calculated simple differences of the means between the two samples for each of the 45 cranial variables. I then expressed these in standard deviations for each variable and picked the highest difference. This happened to be variable 31, the naso-dacryal subtense (NDS), (*) which was significantly higher in the Oslo group than in the Anyang one, due to the difference between the strongly profiled Europoid vs. the flat Mongoloid face.
Then, I built a simple classifier which was simply the following:
If NDS>10.07 Caucasoid Else Mongoloid
10.07 is the arithmetic mean of 12.29 for the Oslo group and 7.85 for the Anyang group. (+) Then, I run all Oslo and all Anyang skulls through the classifier. The result: 91% of the Oslo skulls and 95% of the Anyang skulls were correctly classified.
To test whether this simple classifier generalizes over other Caucasoid/Mongoloid skulls, I tested it over the Zalavar (Hungary), and Berg (Austria) samples where the accuracy was 83% and 77% respectively. I also tested it over the Hokkaido, Hainan and Buriat Mongoloid samples. Accuracy was 89%, 100%, 96%.
In other words, the most braindead classifier conceivable is able to distinguish between Caucasoid and Mongoloid skulls with a reasonable degree of accuracy.
(*) Roughly speaking, the distance between where your glasses rest on your nose and where tears flow from, but on the skull.
(+) I used male skulls only.