which company is right, which is wrong?

[Carpathian]

Well that's certainly curious. I am an immigrant from Poland, very Polish on all sides. Yet when I uploaded my Ancestry's raw DNA to FTDNA I got the following result in My Origins:

European 96%
- East Europe 90%
- British Isles 6%

In reality, my liberal estimate is that maybe 6% of my ancestors knew what Great Britain was. In other words, it's complete baloney. Yet you, who have a British parent, are apparently 2 percent less British than me.
Ancestry DNA correctly estimates that I am 0% British and it makes sense.

Mine is essentially the same situation, and just as absurd. I'm entirely of Slavic ancestry. Both Ancestry and FTDNA have me reported as 99% Eastern European and 1% Siberian. That is essentially correct, and is provable by my family history for all generations over centuries. 23andme has me reported as 76% Eastern European with the remainder including some British in 8.5% "broadly European". This is not a matter of estimation, interpretation or plausibility - their report is just plain inaccurate and misleading. You can determine if a test is accurate if you have proof of your ancestry. It also helps if you are of little or no admixture, because then it's obvious what your ethnicity is. These reports are usually based upon models including comparison with people of other groups. But when the reports cite ethnic origins and regions in time frames that supposedly occurred many centuries ago, before there were written records, they are based upon guesstimates, not on fact.

The word "baloney" is an understatement. It's also been called "hogwash". The primary function of ethnic ancestry prediction is one of mass entertainment.

[John McCoy]

The statements about "many centuries ago" are indeed guesstimates, in my opinion, and I believe I see where they come from.

The admixture algorithms available today all rely on the development of relatively homogeneous clusters of samples from modern, probably still living humans. It was recognized early on that the modern samples are not actually the "ancestors" of the people we want to test. Some researchers called the reference groups "pseudo-ancestors", and noted that they serve as proxies for whoever the ancestors of those modern groups may have been, at some point in the indefinite past.

The algorithms also make the assumption that the SNP's that are used have been recombining in the "populations" that are represented by the modern reference groups, for such a long time that they are essentially unlinked (or, in linkage equilibrium). (We don't actually know that that assumption is met for a real group of modern samples, by the way! We say that the assumption has been met by imagining that the population is sufficiently old.) What this assumption means, I think, is that the "population" itself has been around, with essentially the same gene frequencies (or SNP frequencies) for hundreds or maybe even thousands of years (the numbers chosen are arbitrary) in order to achieve the assumed level of scrambling. I don't think there is a factual basis for citing anything more than a very broad range of years. In this limited, artificial, and nearly tautological sense only, the "population" that gave rise to the modern "reference group" is assumed to have existed for hundreds or thousands of years, and you are estimated with some probability to be a descendant of that "population". It does NOT mean that you have a particular ancestor from an identifiable historical population who lived centuries ago and who was, say, Scandinavian, only that you are statistically similar to a hypothetical, extrapolated population, and that some portion of your ancestry is therefore estimated to have come from that population at some point in the indefinite past, but presumably after that population was originally founded.

Now, we know that real European populations, or most of them, have been mixing with their neighbors almost routinely for as far back as European history itself. The statistical models, based on extrapolations from modern samples, cannot accommodate this fact directly. The algorithms presume static populations where gene frequencies haven't changed, and that must surely have some impact on the results. What impact? That important question doesn't appear to have a clear answer. We don't have enough, say, 16th Century samples to compare with our modern "reference groups", and we lack any solid external standard by which we can judge whether one algorithm is any better than the others. We can look at people's pedigrees, but we have to make the assumption that they are actually correct. We can look at consistency between parents and their children. There are a few other approaches to validation, but none of them will really answer the most important question, how accurate is the statement that I am 37% Scandinavian?

[Carpathian]

The statements about "many centuries ago" are indeed guesstimates, in my opinion, and I believe I see where they come from.

What difference does it make? Does everyone see and understand this? What purpose do these guesstimates serve?

...It does NOT mean that you have a particular ancestor from an identifiable historical population who lived centuries ago and who was, say, Scandinavian, only that you are statistically similar to a hypothetical, extrapolated population, and that some portion of your ancestry is therefore estimated to have come from that population at some point in the indefinite past, but presumably after that population was originally founded.

So how does this help the layperson who is reading the report about his "hypothetical, extrapolated population"? Who benefits here? Is it those who write and sell the hypothetical reports, or those reading them?

...The algorithms presume static populations where gene frequencies haven't changed, and that must surely have some impact on the results. What impact? That important question doesn't appear to have a clear answer. We don't have enough, say, 16th Century samples to compare with our modern "reference groups", and we lack any solid external standard by which we can judge whether one algorithm is any better than the others...

It is interesting that you defend this model with your repeated use of the word "we" in your lengthy explanation. Do you work for any of the GG sites in formulating these reports?

We can look at people's pedigrees, but we have to make the assumption that they are actually correct. We can look at consistency between parents and their children. There are a few other approaches to validation, but none of them will really answer the most important question, how accurate is the statement that I am 37% Scandinavian?

Historical records provide proof (or as you say, validation), if enough of them are available. Once anyone makes any "assumption" about anything, that assumption enters into the area of theory and speculation, not of fact. That's also entering into the realm of entertainment and belief. But you do have an aspect in your favor - if the predictions being made predate any earlier time frame than recorded historical record, then the prediction made becomes incapable of being proven or disproved, and therefore entirely one of assumption or conjecture. Once something cannot be proven as being factual or not, either is possible. But plausibility is one thing, and actual proof is another. The problem with these reports is that they are marketed and presented as being "scientific". From that point it is but a small step for the public consumer to assume that they are factual. In the face of advocates of these reports, such as yourself, it is often hard to convince them that the projections and prognostications have no factual basis, other then the projected statistical models that are their hypothetical foundation.

[jova99]

phasing

Does FTDNA offer a tool to phase a parent and child ?
Phasing with my parents improved my results at 23andme.

The ancestry results for my mother and myself became more accurate, with less categorized as "Broadly European" and increased the classification into the more specific categories.

It seems people who too concerned about getting specific Nations.
FTDNA would be more accurate if they tried not get to specific and had just 4 European categories, Northern Europe, Central Europe, Southern Europe and Eastern Europe. This would be more accurate and not as misleading.

FTDNA certainly could improve their ancestry origins report. FTDNA classified my mother as 74% Central Europe while they have me at 0%
this is not possible if the algo was working properly.

[Carpathian]

It seems people who too concerned about getting specific Nations.

Is there anything wrong with their seeking that information?

FTDNA would be more accurate if they tried not get to specific and had just 4 European categories, Northern Europe, Central Europe, Southern Europe and Eastern Europe. This would be more accurate and not as misleading.

FTDNA certainly could improve their ancestry origins report. FTDNA classified my mother as 74% Central Europe while they have me at 0%
this is not possible if the algo was working properly.

Until recently there was no classification of "Central Europe". There was Western Europe and Eastern Europe, based primarily upon culture and language. There was also Northern (usually Nordic) Europe and Southern (usually Mediterranean) Europe, but those geographical distinctions fell roughly into West or East, again depending upon culture and language. They were based upon broad, historically applicable regions, not merely those of geographic borders or nationalities.

So what is "Central Europe"? It is a newly devised label based upon a recent geo-political redefinition and is a newly imposed term. It ignores more than a millennium of European culture, perhaps in the hope of changing or diminishing it.

A relevant example of this is the old politically fabricated, artificial state of Czechoslovakia. The Czechs had Western European culture and language, and the Slovaks had Eastern European culture and language. The attempt to meld the two cultures into a nation state that was imposed upon them didn't work, and they are now better off having been divorced into two separate countries. This is a good example of where "East meets West" in Europe, quite literally.

But "Central Europe"? No. That's a label based upon mere geography, ignoring many centuries of the people who lived there, including their history, their language and their culture.

[jova99]

Western

i should have stated Western-Central Europe. The problem with ancestry reports, they imply they can narrow down where ones ancestors lived 500 years ago, but using Geographic areas is more accurate than specifying Nations, especially when some of these nations did not exist 500 years ago.

Nothing wrong with wanting to know the nation ones ancestors came from, but is makes the algorithms less accurate. often these nations no longer exist or did not not exist 400 years ago and the borders of many have changed.
Since specifying the nations of our ancestors is not as accurate as classifying the region. Thus they make mistakes by when they classify one as British instead of Western European or Northern European. With Eastern Europe they do to try and specify the nation, although the area is larger than Western and Northern Europe. it seems FTDNA has tried to hard to please people by fusing the nation your ancestors came from, instead of trying to be more accurate and just stating the region.
The nations can be misleading is partly due to historical events, and the fact that people did immigrate to neighboring nations. My German Great Grandfather was born Near the border of Poland, so it was not surprising that my Grandmother has some East German DNA...few Americans know where their great Grandparents were born, they may know they were born in Germany but not know the area...fewer know much about their GG grandparents.

Bratislava is now the capital of Slovakia...It started as a Celtic settlement 2,500 years ago...prior to WWI it was part of the Austrian-Hungarian Empire and the city was 42% German, 40% Hungarian, 5% Czech, 10% Slovak...While today is mostly Slovak, yet Slovaks they did not migrate to the areas until 1,000 years ago and were a minority until 1920.

So stating that your DNA indicates one is Western- Central European seems more accurate than German for my Great Grandmother who was born in Bratislava and had German and Hungarian parents. She did speak German, Slovak and Hungarian and lamented that her Nation no longer existed She was upset when they created Czecho-Slovakia, as the nation of her youth was gone.

[clintonslayton76]

Flawed marketing tools

Generation calculators and "origin" estimates are marketing tools that are essentially misleading. Read the molecular biology journals where you will see that there is little agreement. Do we compare with current populations or ancient ones? Does the term "European" really mean anything? Are "average" mutation rates truly known when some lines (such as mine) are so slow to mutate that 7th cousins are estimated as 2nd-4th? Of all of the people alive how many have tested DNA to allow for meaningful averages?
Rely on traditional genealogy. Do not expect the cold numbers of DNA to warm up and reveal a traceable geographic path. The recent re-evaluation of my "origins" on FTDNA is quite different from the previous one, so is the newer one more accurate? I do not think so. Do I have South-Central Asian in my line, as shown? Only through a Native American connection, but this is not explained on FTDNA.
Please check the writings and postings of geneticist Mark Thomas (developer of the Cohen Modal Haplotype) before swallowing the marketing tools of providers.

[Carpathian]

Generation calculators and "origin" estimates are marketing tools that are essentially misleading.

Marketing tools, indeed. Click this link for the explanation:

http://www.ahajokes.com/law045.html

[susan_dakin]

Exactly!

At the moment, and for the foreseeable future, there doesn't seem to be a rigorous way to validate admixture results.

Exactly! At this point, NO ONE should take admixture results from ANY company seriously. It's certainly not a reason to take a DNA test (contrary to what certain companies claim in their ads), unless you intend to use the results only as entertaining small talk at parties.

[MIKERH]

test variance

I've paid 3 companies to test my DNA. 23andme, Ancestry, and Living DNA. I've also uploaded these samples to as many sites as permit it. I think what surprised me was that the companies are all sampling varying areas of our DNA. My own results are relatively consistent in the high contribution percentages. Where I have seen wild variance is on the small contributions. I think some of this is based on each company's decision with regard to population grouping as well as population samples. Also each company choses what areas of our DNA to sample and compare to their own database. With the uploads I presume they only evaluate the portion of the competing test labs sample that overlaps their own. I have found the results and overall experience very satisfying and entertaining.

[jova99]

speculative verse conservative

it should be noted that 23andme has different settings. The current default is "speculative" and they state the speculative results have are 50% accurate.

The most conservative setting is 90% accurate. Most people do not like the conservative setting because the ancestry report will indicate a higher percentage of "broadly European" and thus lower percentages will be attributed to specific areas like Northern Europe or Southern Europe. But the report will be more accurate.

[Sherila]

I had a similar thing; my mother is 100% Norwegian, both of her parents come from Norway, and have through at least the early 1800's (probably further, but that is as far as I have gotten). 23andme has me down:

Scandinavian
32.5%
British & Irish
27.5%
French & German
2.6%
Finnish
1.4%
Broadly Northwestern European
23.8%
Ashkenazi Jewish
9.5%
Southern European
0.7%
Broadly Southern European
0.7%
Broadly European
2.0%

Which is pretty accurate to the best of my knowledge. The balance between 32.5 and 50 could easily be British/Broadly NW Euro overlap. My second largest percentage should be Dutch, so Broadly NW, and this also mixes often with British, and that should be third. Ashkenazi is accurate, my great great g-father was Ashkenazi. There were a few Germans waaaay back in Colonial times (but no French as far as I can tell.

Family Tree gave me this:

British Isles 87%
East Europe 6%
Ashkenazi 5%
Trace (Siberian, East African)

The trace stuff is possible I guess. 23 had me at 100 Euro and I have a known Native American ancestor but that person is so far back, about 1600, that I could see it not even showing as a trace. The only East Euro I had (unless it is hidden in the Norwegian pre 1800, which is possible) was the Ashkenazi g-g-grandfather. But *zero* Scandinavian?

It was the same sample. This confused me greatly.It is ironic I got this result, I was interested in joining a Scandinavian project, it is the main reason I submitted my sample to this site, but I can't. The project listing only shows me as compatible for a French project, and I am not even remotely French based on my geneaology. So you gotta laugh

But this thread has been helpful, and I will read the sources listed in the posts.