Sunday, September 2, 2018

Review of AncestryDNA's New Ethnicity Estimate Update (Beta) - Continued

My paternal grandfather's new AncestryDNA ethnicity map
Previously, I went over my new, updated results with AncestryDNA's ethnicity estimate and detailed what's different in general and with my personal report. I was mostly disappointed with my update because although the regions were more specific areas, it became less accurate in comparison with my tree, but now I want to get into the results of the other kits I manage because a couple of them are excitingly very accurate.

My theory is that the update seems to work best for people who are less mixed. Unfortunately, this doesn't bode well for most Americans since most of us are fairly mixed and come from multiple backgrounds. I believe this is why my report became less accurate with the update - I am a mixture of Italian, British, German, Norwegian, and a little bit Dutch and French. But my paternal grandfather, on the other hand, isn't very mixed at all. He's basically around 60% German and 40% Scottish or Scots-Irish, and although his results don't exactly match his tree, the new update is much more accurate than it was before.

My paternal grandfather's original AncestryDNA ethnicity
His original results (also see image right):
Great Britain 67% (range 43-87%)
Europe West 15% (range 0-35%)
Caucasus 10% (range 5-16%)
(Low Confidence Regions)
Italy/Greece 5% (range 0-14%)
Scandinavia 1% (range 0-7%)
European Jewish 1% (range 0-4%)
Iberian Peninsula < 1% (range 0-4%)
Europe East < 1% (range 0-3%)

With the update (also see map above):
England, Wales & Northwestern Europe 67% (range 62-100%)
Germanic Europe 23% (range 22-23%)
France 6% (range 0-9%)
Ireland and Scotland 4% (range 0-5%)

Remember, as I pointed out before, even though the category is now called "England, Wales, & Northwestern Europe, it covers the same area as before, including Scotland. The update is still leaning more towards his British background when his tree suggests he is more German, but the loss of all those low percentages in regions that didn't match his tree make it much more accurate than before. They also eliminated the unexpected 10% Caucasus, which wasn't even a low confidence estimate and so simply dropping that has made it much more consistent with his tree. Also, keep in mind that while we inherit 50% from each parent, we do not inherit exactly 25% from each grandparent, or 12.5% from each great grandparent, etc. So when I say my grandfather's tree suggests he's 60% German and 40% Scottish, that is very much an estimate because it's based on simply dividing each generation by half, which isn't actually how inheritance works. It's entirely possible he inherited more DNA from his Scottish ancestors than his German ones, even though he had more German ancestors than Scottish, and therefore the lean towards British over German in his ethnicity report could be entirely accurate.

My paternal grandfather's new AncestryDNA ethnicity
You may notice something weird about his new ranges though. As far as I know, they are still calculated by running 40 different analyses and than averaging out the results to get your final percentage, with the range showing the lowest and highest scores you got out of those 40 analyses. For England/Wales, my grandfather's range is as high as 100%, but that means every other category must have a range as low as 0%, and they don't. For Germanic, his range is 22-23%. It seems as though AncestryDNA are excluding the 0% in Germanic as an outlier - which I wouldn't have a problem with, but if they are going to do that, it means they should also be excluding the 100% England/Wales outlier too. Granted, it is still supposed to be in beta mode, so this could change as they continue to tweak things.

My husband is even less mixed than my grandfather. He is from England, born and raised, with a Catholic Irish father and Protestant English mother (which apparently caused quite the problem in their family when they married). On his mother's side, he actually has a Scottish branch and another Irish branch if you go back far enough, so he's about 60% Irish/Scottish and 40% English, and of course 100% British Isles.

Husband's update and comparison with old report
Here's his old results:
Ireland/Scotland/Wales 67% (range 50% - 79%)
Scandinavia 12% (range 0% - 27%)
Europe South 9% (range 1% - 16%)
Europe West 6% (range 0% - 23%)
Low Confidence Regions:
Iberian Peninsula 5% (range 0% - 13%)
Great Britain 1% (range 0% - 8%)

And with the new update:
Ireland and Scotland 61% (range 45% - 76%)
England and Wales 38% (range 34% - 38%)
Benin Togo 1% (range 0% - 1%)

Wow! Apart from that weird 1% in Africa, which is obviously just noise, the new results are almost exactly what his tree says and support the idea that the less mixed you are, the more accurate the update may be for you.

Here again, were seeing inconsistencies with the ranges. How can one category range by 31% when the only other two categories only range by 1% and 4%? In order to have gotten 45% as the lowest score for Ireland/Scotland, the rest of his categories must have added up to 55% to make 100%, but neither of them have that high of a result as their maximum in the range. This again suggests that they are sometimes excluding outliers, but oddly aren't doing it consistently. I will be eager to see more details when they finally release them. Understandably, they have not given us much information (none at all, really) on their methods for the update because it's still in beta mode.

Dad's update and comparison with old report
My dad's new results are aren't quite as accurate because he's still seeing a number of low percentages in various regions. His background isn't hugely mixed, but is 50% Italian, and his other half is about 30% German and 20% British (again, Scottish/Scots-Irish).

Previously:
Italy/Greece 44% (range 31-55%)
Europe West 19% (range 0-43%)
Great Britain 15% (range 0-35%)
Ireland 7% (range 0-19%)
Low Confidence Regions:
Middle East 5% (range 0-10%)
Scandinavia 4% (range 0-16%)
Caucasus 4% (range 0-10%)
Finland/Northwest Russia 1% (range 0-5%)
Asia South < 1% (range 0-2%)

Updated:
Italy 44% (range 43-58%)
England and Wales 24% (range 23-24%)
France 12% (range 0-13%)
Ireland and Scotland 10% (range 0-10%)
Germanic Europe 3% (range 0-27%)
Greece and The Balkans 3% (range 0-3%)
Sweden 2% (range 0-2%)
Turkey and the Caucasus 2% (range 0-2%)

It's good to see his Italian/Southern Europe results didn't change, considering mine changed so drastically (and wrongly). His are still 44% and now narrowed down to Italy, not just Europe South. And the 24% England/Wales is fairly accurate too considering, again, that despite the name it does primarily include Scotland as well. But they did fail to identify much German and he's still getting several noise-level results. Noteworthy though is the fact that if they were still using the old low confidence regions, everything apart from Italy, England/Wales, and Germanic Europe would be in low confidence, which is consistent with his tree. The criteria for it is described as:

"When an ethnicity has a range that includes zero (meaning that in at least one of the 40 tests, that ethnicity didn’t appear) and doesn’t exceed 15%, or when the predicted percentage is less than 4.5%, the ethnicity is included in an estimate as a low confidence region." - Full article here

So despite his low average for Germanic, the range is as high as 27% and that would be above the criteria for the low confidence regions. Looking at it that way, his new results for mainly Italy, England/Wales, and Germanic are actually kind of accurate.

My mom's previous AncestryDNA ethnicity report
Lastly, we come to my mom's new results. Her background is more mixed than my dad's, so it's not surprising that her update isn't quite as accurate. She is approximately 50% British (English and Scots-Irish), 25% Norwegian, 20% German/Swiss, and possibly a tiny 2-3% Dutch and 2-3% French.

Her previous results (show left):
Great Britain 46% (range 10-81%)
Scandinavia 29% (range 2-57%)
Europe West 16% (range 0-44%)
Ireland 4% (range 0-14%)
Italy/Greece 3% (range 0-8%)
Iberian Peninsula 2% (range 0-7%)

With the update (shown below):
England, Wales & Northwestern Europe 45% (range 43-52%)
Norway 40% (range 38-40%)
Ireland and Scotland 11% (range 0-12%)
Sweden 4% (range 0-4%)

My mom's new AncestryDNA ethnicity report
Her British results haven't changed much, but her Norwegian/Scandinavian results are now much, much higher than before, a drastic deviation from the fact that she only had one Norwegian grandparent. While the amounts we inherit from a grandparent may vary, it's unlikely to be as high as 40%, much less 44% if you add in the Swedish results. But most importantly, where did her German ancestry go? The update definitely seem to be underestimating German results in general, at least for my family. At least my dad got small results in Germanic - but none at all for my mom?

The ranges are generally smaller than they used to be, which suggests better confidence or more consistency in the results, but is that because in some cases, they are excluding outliers? Hard to say until we get more information on their methodology.

6 comments:

  1. My update changed drastically! It completely removed all low confidence regions, and even eliminated Eastern Europe which was my third highest region. My grandpa's family migrated from the former Bohemia so I was very surprised to see this removed. Also, on my grandma's they removed her formerly 40% german altogether! Like whattt? I'm not really feeling that this update is very reliable or accurate.

    ReplyDelete
  2. This update is hardly an improvement. Well I suppose it could be considered an improvement if you know next to nothing about your ancestry & any shot-in-the-dark ethnic guesswork seems legit.

    Before the update I possessed 14% Iberian & 8% South European. Accurate to my known ancestry. Now it is 15% French. I *do* have French ancestry but it's predominantly southern French ancestry which should be closer to southern euro/iberian - on the free website gedmatch I am almost 25% "Spanish" on some of their calculators/oracles, on My Heritage I am 30% southern European, etc. - than generic whatever French.

    ReplyDelete
  3. I just had my DNA report through after the update. I was surprised by the level of accuracy, even pinpointing my genetic heritage to the South East and West Midlands regions of England, where my grandparents originate. I had 12% Ireland and Scotland and it is known that one of my great great grandfathers was from Ayrshire.

    From seeing other people's updated results and their original results, I think you may be right that the updated results are more accurate and specific for people who don't have a lot of mixture.

    ReplyDelete
  4. I just don't like AncestryDNA's estimate I feel like AncestryDNA is not primarily scientific but almost like a facebook page for geneology. They recently increased my Scottish/Welsh/Irish estimate, lowered my British and split my Scandinavian to Swedish and Norwegian. I have no known Scandinavian ancestry and that ancestry could easily just be from Scotland (Norway) and England (Danish). Furthermore, the BBC reported that that Scotland, Wales and Ireland have different genetic profiles. A dog on the street could tell you that the Irish and the Scottish look similar (hyberbole) but really mainland Scots are halfway between the English/Germanics and the Irish more or less with maybe some Welsh.

    Nevermind the Cornish ! Their British sample comes from East Anglia because Americans ignorantly associate the English with the British. If you are going to do that then middle England would be better. Ah, you shouldn't have gotten me started I am ranting now.

    I prefer GEDmatch !

    ReplyDelete
    Replies
    1. AncestryDNA's ethnicity estimate is no less scientific than any other company. The accuracy of it, or the consistency of it with your known ancestry really depends on the individual, as is true for any other company. You can read about the science behind it in their White Paper: https://www.ancestrycdn.com/dna/static/pdf/whitepapers/WhitePaper_2018_1130_update.pdf

      Gedmatch's admixture calculators are fun and interesting to explore, but they are not necessarily any better than AncestryDNA's estimate, and are arguably getting a bit outdated.

      The genetic differences or similarities among the British Isles is debatable.

      Delete
  5. Really interesting page, It might interest you to know that 53% of white Americans today are descended from Colonial Settlers (80% of whom were British, mostly English) whilst 47% are more recent (Ellis island etc) immigrants. Most Americans cannot trace their roots back to colonial times so it’s probable that you do have a lot of English Ancestry as do most Americans which you don’t know about. 7% of Americans said they were English on the US 2010 census when it’s more like 35%+ German was incorrectly given as the highest Ancestry Also Scots-Irish are of Scottish/Northern English Decent and many “Irish” Americans are their decedents not Catholic Irish fleeing the famine in the mid 1800s. Ancestry did a “Genetic Census of America” which showed most states were 30 - 55% British. The Average Englishmen has only 60% English DNA with about 20% Irish DNA and differing amounts of German and Norwegian DNA percentages increasing the further east you go. Hope this helps.

    ReplyDelete