So, I decided to look into it once again, to see if there is anything to the complaints. I also wanted to check whether I had included some value errors that I had discovered by a statistical analysis of the original anthropometric data ("International anthropometric study of facial morphology in various ethnic groups/races", J Craniofac Surg 16:615-46) and that the late Leslie Farkas confirmed with me.
The corrected values sent to me by Dr. Farkas (Apr 13, 2007) are:
Croatian female en-ex mean=28.3mm [38.3mm in the paper]Iranian male en-ex mean= 27.2mm [37.2mm in the paper]
The anthropometric calculator was first released in 2006, so these errata were not included, and hence Croatian female and Iranian male estimates should be checked again.
There was another issue with the calculator, with the White American male sample, where the mean values in the paper are presented in the 4th column, while for all the other populations in the 3rd one. Correspondingly, the extractor I built did not use the mean values for this sample but rather the mean minus 1SD values. So, the White American male estimate should also be re-checked.
Why does "everybody" turn Greek?
Where does the impression come that "everybody" turns out Greek in the calculator? Certainly the few test cases (e.g., German facial averages) that I ran back in 2006 did not turn out Greek, so it was a bit perplexing that so many people complained about it.
Judging from the few people who actually accompanied their complaints with their actual anthropometric data, my guess is that they are measuring wrong. Of course, this begs the question: if people don't know how to measure, then they should get random results, and not an over-representation of "Greek". This is true if people introduce random errors about their actual anthropometric values, but this is not the case: in particular the measurements of nose inclination and bigonial diameter seem to be systematically overstimated, and in both these measurements Greeks have a very high value.
I won't say anything more, except this: "garbage in garbage out", and my several-year experience with this and other calculators tends to make me think that people rarely measure themselves correctly. Two of the main sources of error are the inability to measure vertical and horizontal lengths (deviations from these increase measurements) and the use of the photographic method with close-up pictures (which distorts proportions: take a picture of your face from a very close and very far and you will see that the proportions are not the same).
Validation of anthropometric calculator
Finally, I decided to run the calculator with some known good values. Click on each example to see the output of the calculator: the original measurements can be seen in the URL. I have used the mean values for a few populations, as reported by Farkas, and also a few population averages, which represent what you would get if e.g., you average Italian with German.
We see that White American is chosen as the best match (obviously!) and the closest other population is German, which wins all the other pairwise "matches". Again, this makes sense, as Germans are the only other NW European population included.
Greek is the best match with no other close matches.
Croatian is the best match, with more distant scores for Bulgarians and Greeks.
No close matches other than Italian.
German, with a distant match for White American.
No close matches or clear read for such a "mixed" person, with non-zero estimates for Italian, German, Greeks, and a few populations of Slavs.
Closest to Thai, and next to Singaporean-Chinese and Vietnamese.
Closest to itself, with no other close matches. Note, however, how African Americans and Zulu win all their pairwise matches against non-African populations.
The anthropometric calculator works fine if it is given reasonable measurements. With such measurements it does not favor any particular population. Of course, we should not expect perfect results with 14 anthropometric variables measured in the home as we would with 57 measured by experts, but, as an exploratory tool, it serves its purpose well.