In 2016, I wrote a blog post comparing the ethnicity predictions between the three largest providers of DNA tests. Since then, some companies have changed their algorithms and I’ve tested with two more companies. Time for an update! I’ve kept the original article as-is so we can compare how the predictions change over time.
To recap, according to my tree, I’m mostly Dutch with a bit of German. DNA matches confirm that my tree is correct for the past several generations. I have some French Huguenot lines in the 1600s and some supposed noble lines in the 1200s and earlier from different parts of Europe. But in general, I’m about as Dutch as they come. How do the different testing companies show my ethnicity?
Ancestry’s ethnicity estimate hasn’t changed since 2016, and shows me as 94% Western European, and low confidence regions Great Britain (2%), Iberian Peninsula (2%), Ireland/Scotland/Wales (1%) and Scandinavia (<1%). I know that Ancestry is rolling out their new ethnicity prediction algorithm, but it hasn’t reached me yet so these predictions are the same as two years ago.
New since 2016 are “Genetic Communities.” Ancestry places me in the community “Germany, the Netherlands, Belgium & Luxembourg” which is spot on. When I click on that community, it shows that I most likely belong to the subgroup “Netherlands.” That’s pretty impressive!
23andMe predicts me to be French & German (39.3%) and even suggests that’s probably from the Netherlands. That’s an improvement from two years ago, when they did not go down to the country level. But it also predicts a significant British and Irish component at 28.9%, probably from the United Kingdom, which is not corroborated by my tree, and neither is the 6.8% Scandinavian or 1.4% Eastern European. The percentages are slightly different than those in 2016, but with the same two large predicted regions. Apart from the addition of the country-level predictions, the results are pretty similar to 2016, and similarly incorrect.
What’s new since 2016 is a timeline view, which shows how these predictions translate to my family tree. 23andMe predicts I “most likely had a parent, grandparent, or great-grandparent who was 100% British & Irish. This person was likely born between 1890 and 1950.” Similarly, it suggests I had a Scandinavian ancestor 3-5 generations ago and an Eastern European ancestor 5-8 generations ago. My tree is mostly complete back that far, and there are no Brits, Scandinavians, or Eastern Europeans in these generations.
FamilyTreeDNA had the worst prediction in 2016, completely missing my Western European DNA, and it’s gotten worse. It now thinks I’m 89% British, 4% Southern European and 2% Middle Eastern. It did not pick up any of my North-West European DNA.
In contrast, FamilyTreeDNA thinks my mother is 61% West and Central European, and my paternal uncle is 59% West and Central European, so there’s something about my kit which makes them completely miss the mark.
I uploaded my raw FamilyTreeDNA file to MyHeritage and got a prediction there too. Even though the raw data is the same, each company has different reference populations and ethnicity prediction algorithms, which explains the differences in the results.
MyHeritage predicts me to be 64.2% North and West European, 32.8% Irish, Scottish, and Welsh, 1.8% Ashkenazi Jewish, and 1.2% Middle Eastern. Except for Ancestry, it has the highest prediction of North-Western European DNA, which according to my research should be 100%. I have no idea why MyHeritage determined I have Ashkenazi Jewish or Middle Eastern ancestry, but these amounts are so low that they’re probably just noise. The Irish/Scotish/Welsh they predict is a subdivision of the higher-level North and West Europe group. At that higher level they predict me to be 97.0% North-Western European, which matches my tree.
LivingDNA predicts me to be 94.4% British/Irish, 1.1% North-Western European, 2.5% North Turkish, and 1.9% North-Western Caucuses. The British is way too high, the North-western European is way too low, and I have no idea where the Turkish and Caucasus comes from.
Living DNA even breaks down my supposed 94.4% British further into 21% South Central England, 17.3% South Yorkshire, 15.1% East Anglia, 13.6% South England, 13.1% Lincolnshire, 5.2% Central England, 3% Cornwall, 1.7% Northwest Scotland , 1.7% Aberdeenshire, 1.6% Devon and 1.2% Southeast England. It thinks my North and West European is Germanic (1.1%).
Clearly, LivingDNA suggests a level of detail for their ethnicity predictions they can’t deliver. If they get it this wrong on the parts-of-Europe level, there’s no use in splitting the groups up further into shires and regions.
Like in 2016, the ethnicity estimate that most closely resembles my tree was delivered by Ancestry. They not only got the largest group right, but even suggested that I might belong to the “Netherlands” genetic community.
MyHeritage and 23andMe recognized significant North-Western European contributions to my DNA, although they also attributed significant amounts to the British Isles.
The worst predictions for my ethnicity came from FamilyTreeDNA and LivingDNA, that both thought I was predominantly British. It is interesting that FamilyTreeDNA recognized Western European in my mother and paternal uncle’s DNA but not in mine, which shows that one person’s results should not be used as a benchmark.
The predictions haven’t changed that much since 2016: companies that got it right then still do, and companies that missed the mark then still miss it today. Of the two new companies I’ve tested with, MyHeritage’s prediction best matched my tree and picked up a large part of my North-West European DNA. LivingDNA’s incorrect prediction of me as a Brit was disappointing to me, since they marketed themselves as specialists in ethnicity predictions rather than finding matches.
I’m intrigued that FamilyTreeDNA, MyHeritage, and LivingDNA all picked up some Middle Eastern/Turkish/Caucasian DNA, while this was not picked up by 23andMe and Ancestry. My tree does not have any indication for ancestors from that area, apart from a Byzantine princess that supposedly married one of my ancestors in the Middle Ages, more than twenty generations ago. Her contribution, if that line proves to be correct, would be negligable after so many generations. For now, I will dismiss these results as an anomaly of the prediction algorithms.
I can see how these types of predictions could lead people astray who don’t already have a good tree. An adoptee getting these results could easily think they’re half British or more. People often think that low percentages are noise, but my results show that even 94.4% can be wrong. To quote Judy Russell, it’s still not soup. At the continent level, however, all companies agreed that I’m mostly European, so my recommendation is to take anything below that with a bucket of salt.
Have you had your DNA tested? How did your ethnicity predictions match the known information from your tree? Please leave a comment to share your experiences.