Can You Believe Autosomal DNA Test Estimates of Ethnicity?

Question

There is an interesting blog article by Judy Russell about the results of autosomal DNA tests.  You can read it at The Legal Genealogist.

Answer 1

I read the article, but I think it placed too much emphasis on reference database as the primary cause of the significant differences. The main reason for the disparity is because we only have a small percent of the total DNA that would be required to have all siblings get the same or similar results.

On average, we get should get about 25 percent of our DNA from each of our four grandparents.  In a perfect distribution of DNA, this means that we are not looking at 75% of each grandparent's DNA.

In the case of 4 siblings and a perfect 25% distribution from each grandparent, there is still a randomness as to which segments come from each grandparent.  The reference database does not have data for every segment or part of a segment.  This adds to the randomness, One sibling may get a high concentration of reference material, and another, get less or even none.

Of course, there is not perfect 25% distribution, which also contributes to the disparities.

Even if we could map every possible segment, we just don't hold enough DNA. :)

Comments

I should have extended the example back in time a bit.  For example, your grandparents are not getting 75% of their grandparents, and so forth.

Another point is that each siblings DNA actually contains DNA from different ethnicities and the results may actually be a representative sample of the DNA was actually inherited.

---

Ken: I totally understand what you are saying, However the salient point is that these ethnicity percentages are estimates only, and probably very inaccurate.

I've tested via 23andme and FTDNA's Family Finder, Result two completely different ethnicity estimates, and 23andme has me as a small percentage Mediterranean and a small percentage Ashkenazi.. neither of which are possible, especially the Mediterranean, apparently they are working on an assumption that a certain haplogroup or group of STR's are Mediterranean.

Point is that we don't take these ethnicity estimates as gospel.

---

No haplogroups or STRs are used. They use autosomal SNPs which have been identified as being Ancestry-Informative Markers (AIMs) https://en.m.wikipedia.org/wiki/Ancestry-informative_marker

The accuracy of the ethnicity percentages depends on how representative the reference samples are of the ancestral populations.  The ethnicity of siblings will differ because they don't always inherit the same autosomal DNA of their distant ancestors.  Related information at http://www.wikitree.com/g2g/231911/-unmixed-ancestry---rarer-than-hen-s-teeth-

Answer 2

Each company uses different population reference markers so your results will be different at each company.  At Ancestry I am 25% Italian.  At 23andMe I am 0% Italian.  Am I Italian?  Yes.  But probably not 25%.

As the Legal Genealogist said: "these estimates take the DNA of living people — us, the test takers — and compare it to the DNA of other living people — people whose parents and grandparents and, sometimes, even great grandparents all come from one geographic area."

And each company uses a different pool of people.

Is it reasonable for siblings to have the varied results as she showed in her article? Absolutely.  Autosomal DNA is like a salad, not a pie.  And that salad gets mixed up over each generation and each sibling gets a different serving.  This is why my brother has DNA matches that I don't have.  He got a radish while I got the tomato.   :-)

Comments

Ah, I see your point. If a pair of grandparents have different ethnic backgrounds (for example, the maternal grandmother was Irish and the maternal grandfather was Italian), the siblings among their grandchildren could end up with noticeably different ethnic mixes, depending on how much of their DNA from Mom came from each grandparent.

---

@ Ellen Smith    Yes.  Exactly.  My brother is more Irish and German than I am.  I am more Italian and Scandinavian.

Answer 3

Those disparate results for four siblings, as reported on the blog, do suggest that the quality of the ancestry data in the database is not good enough to support the level of precision implied by the conclusions that are reported.

23andMe provided very similar ancestries for my sister and me, and they are generally consistent with our understanding of our heritage. We are the two columns of percentages below:

	European	100.0%	100.0%
	Northwestern European	96.0%	97.2%
	British & Irish	43.0%	40.8%
	French & German	22.1%	24.0%
	Scandinavian	1.0%	2.8%
	Broadly Northwestern European	29.9%	29.5%
	Southern European	1.6%	0.9%
	Italian	1.3%	0.0%
	Iberian	0.0%	0.2%
	Broadly Southern European	0.3%	0.7%
	Ashkenazi	0.2%	< 0.1%
	Broadly European	2.1%	1.9%
	East Asian & Native American	< 0.1%	0.0%

 

I get confused when I supply the same data to the ancestry algorithms from Eurogenes, etc., on Gedmatch, as well as the algorithm used on DNA.land. For example, DNA.land says I'm 100% West Eurasian. I think that's a pretty solid conclusion, but their more detailed breakdown diverges pretty significantly from 23andMe and my known genealogy. They subdivided "West Eurasian" as 67% Northwest European (Scottish,British, Icelandic in Iceland, Norwegian, or Orcadian), 17% Northeast European (mostly North Slavic (Belarusian, Estonian, Lithuanian, Mordovian, Russian, or Ukrainian), 10% Mediterranean Islander (Cypriot, Sicilian, or Maltese), and 5.6% Ambiguous. In that analysis, the 67% Northwest European is what I would expect, but the omission of a German or French component and the large components of Slavic and Mediterranean Islander are totally out of line with what I think I know. It appears to me that the 23andMe ancestry is based on the distribution of peoples in Europe in the late middle ages, but the ancestries supplied by DNA.land, Eurogenes, etc., must dig farther back into the origins of the Vandals, Goths, Visigoths, etc., who populated Europe around the time of the fall of the Roman Empire.

Answer 4

Mine was pretty much right on target with my Ancestry. com  pie chart.  I guess it has a lot to do with how many samples they have from certain regions.