Testing: Caveat Emptor

Admixture caveat.

The ancestry testing companies essentially are doing supervised admixture analysis, comparing customers to reference samples. If a reference population is admixed, then customers from the same, or similar, populations will match the reference, and will not necessarily exhibit admixture in the test, even if they are in fact actually admixed. Thus, if admixed population “X” is a reference, representing “100% X” and if a customer of the same (or similar) ethnic background is tested against that reference, then the customer will get a result of approximately 100% X. The customer can in fact be admixed, but if the type and levels of admixture is similar to that of the reference population that they are being compared to, then the admixture (subsumed within the definition of “X”) will obviously not be represented in the results. If the admixture in the reference population is relatively old, and is today considered part of the genome of that population, then it is not going to be recognized as admixture; even if the company has “quality control” algorithms to attempt to “weed out” admixed individuals from the reference populations, that is going to really work only for relatively recent admixture. So, it is entirely possible for individuals with actual admixture to test as “100% pure” if the yardstick against which they are measured is also similarly admixed. On the other hand, customers with little to no actual admixture can be represented as “admixed” if they are not compared against good reference population matches, and, hence, are having their ancestries approximated as mixtures of whatever reference samples exist in the database. 

When customers have their "results" changing +/- 15-30% (or whatever other outrageously high level) every time companies update their databases, it is clear that these tests are completely and laughably inaccurate and imprecise.  Wild fluctuations in results suggest  that the databases lack the proper reference samples and customers are having ancestries assigned to whatever samples are being added to those databases. Combined with high levels of "unassigned ancestry" at high confidence levels, identical wins getting different results, people getting radically different results with different companies (as well as radically different results with the same company after updates, as discussed above) these facts demonstrate that every single of the extant ancestry tests are virtually useless with respect to accurate and reproducible results (at least for individuals deriving from ethnies not well represented in databases).

Thus: Caveat Emptor.

I cannot recommend any of the extant ancestry tests; indeed, I, at the current time, heartily discourage their use - they may work reasonably well for those of British Isles descent (sometimes) and Ashkenazi Jews, and possibly some other ethnies both well represented in their databases and relatively genetically homogeneous, but fail miserably for others. Fetishist retards who breathlessly discuss results (a behavior I recently mocked) demonstrate their own ignorance and/or mendacity.

If this situation is remedied, then my recommendation may change, but it is important to note that improvements must be stable over time. Results that drastically change with each update are unstable and, hence, imprecise and likely inaccurate (after all, if there are drastic changes, it cannot be true that both the original or the update are correct - and in this case, there's no guarantee that future changes will not occur, putting into question additional data sets).