As I was going through the list of the Dodecad populations, I realized that there are 5+ participants in each of the Korean, Japanese, and Chinese groups. So, it seemed like a simple exercise to see whether the relatively high success rate of people's guesses could be corroborated using the DNA data.
Below is the MDS plot; there are 9 Chinese, 5 Japanese, 5 Koreans in the Dodecad Project; I have also added 30 HapMap Chinese (CHB) and Japanese (JPT):
Interestingly, when I did not use the extra HapMap individuals, MCLUST did not split Koreans from Chinese. This goes to show that the absence of apparent structure does not imply absence of structure. The extra Chinese and Japanese individuals helped flesh out the existing structure in these East Asian groups.
Below is the list of the Dodecad populations that are below the 5-individual limit:
If you belong to one of the above groups (all 4 grandparents) and have tested with either 23andMe or Family Finder, you are especially invited to contact me at firstname.lastname@example.org (but do not send data right away!), about possible inclusion in the project.
For example, in the most recent Clusters Galore analysis, there was a generic "Balkan" cluster. Does this imply that Balkan ethnic groups cannot be distinguished from each other, or that sample sizes are simply not yet sufficient to make manifest the existing structure?