Putting aside the issue of companies intentionally altering results for sociopolitical reasons – for which they should be sued by customers – I think these tests are reasonably good at determining majority ancestry at the continental level and also good at determining high (10-15+ %) “admixture” at the continental level as well. 23andme, with all the SNPs they use, will most likely be reasonably accurate for intra-continental majority ancestry – whether someone is of British Isles descent or more generally Northern European or Eastern European or Southern European, etc. Some other information may be more or less accurate as well, but there are limits. As I’ve said before, I would have a healthy skepticism for any minor ancestry under 10%, definitely skeptical under 5%, and once you get below the 1-2% range it’s laughable to believe that’s realistically statistically different from zero. The possibility that these companies screw around with customers is real; it may be more likely though that they make the significance levels so broad that they virtually guarantee some “exotic admixture” for many people, particularly those not people derived from parental populations from which gene frequencies were determined (*).
The tests are more useful than some of the more extreme detractors claim; they do work in the broad sense stated above. But they do not work to the extent that they are advertised by the companies, and they certainly cannot be used to definitively determine ancestry in the Nutzi fashion of obsessing over “purity” – even the most stringent confidence intervals used by these companies is not at the levels commonly accepted in the literature, and these companies have already admitted that genetic distance from parental populations can create artefactual admixture. Tools are useful when used properly and when you understand their limitations, and these tests are tools that can be used or misused.
The one good thing about these tests is the accumulation of raw autosomal DNA data, which could be used in the future for genetic kinship assays or genetic structure analyses. Years ago, Decodeme allowed 23andme users to upload their raw data and gave back a ranked list of ethnic groups most genetically similar to the uploaded profile. That was a form of genetic kinship analysis (crude and qualitative) that was useful, generating findings that refuted both the crazed race-deniers of the Left and the crazed ethnic fetishists of the Right.
I’d like to finish by answering certain fundamentally dishonest population geneticists, interviewed by leftist journalists, who make comments such as “there’s no thing as a White European group” and “there are no single gene variants that distinguish one group from another; we all share such variants.”
As regards the first comment, your own data refute your statement, a continental European group (which can of course be further subdivided as one gets ever more fine-grained in the analysis) clearly falls out of the data, and the fact that the group – depending on how things are studied – may be slightly fuzzy around its edges does in no way suggest it does not exist.
As far as your second comment goes, I do notice the word “single” in there, you misleading, mendacious bastards – who says single gene variants define any population group? Isn’t that why you study hundreds or thousands or hundreds of thousands of SNPs, you dishonest scum?
Fat chance any of these people ever doing genetic kinship assays or DifferInt analyses on large-scale human data. It’s not “paranoia” to say they have political agendas, it is just plain fact. Anyone who doubts that should meet some of these folks at scientific conferences, go up to them at their poster presentations, and delicately bring up the implications of their work with respect to race. You will get a hysterical SJW tirade in response; for these people anti-racist politics are more important than scientific integrity, by at least one order of magnitude. Still doubt it? Read this, particularly the last paragraph.
*Consider the category “unassigned” in these test results, which for customers of certain ethnies can reach very high percentages when the most conservative confidence intervals are used (and even those confidence intervals are too “liberal” by scientific standards). So, what does it mean? That these people are part-Martian? No, what is actually does mean is that these companies have very poor ethnic coverage, and when one approaches reasonable confidence intervals large fractions of these people’s genomes are left undetermined. That also raises the question of how accurate the assigned determinations of the genome for these people are, for minority “admixture,” as noted above.