MyOrigins - MANY Discrepancies - please list yours here

[Ftdnauser]

Hopefully, FTDNA will listen to us.

I've noticed quite a few discrepancies in the new MyOrigins among many users. People are reporting things like, one sibling having Jewish, and all siblings and parents show none. People are also reporting Jewish results they did not have before, and people who had Jewish in their Population Finder, with with less than 0.5% margin of error, and now there is no Jewish. People are also reporting things like having Central European Coastlands, even though they are of Irish descendancy, so you would think that it would be European Coastal Islands instead.

So, will there be any fix to this anytime soon? I ask this, because Population Finder remained in "Beta" stage since it's inception, and yet as far as I know, no changes were ever made to results.

[dtvmcdonald]

I can answer this. I was the author of the code
for the original Pop Finder.

There was never anything wrong per se with the Pop Finder
code. It simply HAD to use a database that was available
when it was written, and changing it would have been a nightmare. I changed the version I used three times and
it WAS a nightmare. And this database had to use Orkney instead of England, as there was at the time no
available England database.

We don't yet know how My Origins works, or what they used
as database. I have not looked at a wide variety of people
with it since I don't have access to the needed projects.
I have looked at all the Clan Donald people who have it.
Of course most of these are English/Irish/Scottish, but
there are a very few people who are McDonalds in the male line and are actually Italian, Spanish, and Mexican.

The new Pop finder is just fine, niot broken in any way.
It is getting, in our list, 100% sane answers.

ALL these things, including 23andme and not quite
as good Ancestry.com, and almost all of Gedmatch, are
also sane and correct.

One simply cannot expect different programs to give
identical answers for everybody. One must expect small bits
and pieces of the "edge" of a person's general region.
One must expect siblings to sometimes test quite differently. In the latter case of course if the parents are
different origins but both pure (e.g. English and
Vietnamese) then the children will in fact all be 50-50.
But if both parents are generic mixed Mexican, even both
being 50% Mayan, 40% Spanish, 5% Jewish and 5% Nigerian, which is possible, then one child could be 0% Jewish and
the other 10% Jewish, or one 70% Mayan and one 30%.

This applies to Europe too.

And different programs, even using the same data, and
give APPARENTLY different answers, exactly as you report,
as still be correct. It just depends on how the database
makes "cuts" between different comparison panels,
and whether they use modern populations (the old Pop Finder)
or "Zombies" (DIYDodecad). Neither is more correct.

Now if the average spot on the map of a purely
European/Jewish/Armenian(etc.)/Turkish person
differs by more than 1000 miles between programs,
then you have a really bad and exceptional case.
But 600 mile error is common and to be expected!

In the case of people who are listed as a mix of Coastal Plain/ Islands (Britain)/Scandinavian/"Euro Peneplan"
and small amounts of Mediterranean, siblings and other close relatives should do very well in the "average position on the map" test. I and my aunt are a good
test of this, and ALL the programs pass the test nicely.
We sometimes test as French, sometimes as English/German/Italian ... but the average is always the same spot in
Normandy, which is wrong by the distance from Calais
to say London.

Doug McDonald

[Germanica]

I certainly don't expect identical answers from each program and I understand it's a complex topic, but when I get ZERO British results (European coastal islands) from myOrigins even though I not only have several known British branches in my tree but also, AncestryDNA are putting my British results as high as 55% something is obviously wrong. Additionally, myOrigins has my Scandinavian (European northlands) and German (European coastal plains) results significantly higher than AncestryDNA (34% and 26% versus 2% and 5%). It's becoming more and more obvious that myOrigins is putting my British results into northlands and coastal plains instead, because many people are reporting the same or opposite problem (people with German or Scandinavian but no British ancestry showing results for coastal islands).

I accept that AncestryDNA could be a little off as well - perhaps they are overestimating my British results and underestimating my German and Norwegian ancestry. And if myOrigins was showing at least SOME coastal islands results, I wouldn't be complaining. But fact of the matter remains that I clearly have British DNA that isn't registering AT ALL on myOrigins and that is a significant issue.

[Ftdnauser]

I can answer this. I was the author of the code
for the original Pop Finder.

There was never anything wrong per se with the Pop Finder
code. It simply HAD to use a database that was available
when it was written, and changing it would have been a nightmare. I changed the version I used three times and
it WAS a nightmare. And this database had to use Orkney instead of England, as there was at the time no
available England database.

We don't yet know how My Origins works, or what they used
as database. I have not looked at a wide variety of people
with it since I don't have access to the needed projects.
I have looked at all the Clan Donald people who have it.
Of course most of these are English/Irish/Scottish, but
there are a very few people who are McDonalds in the male line and are actually Italian, Spanish, and Mexican.

The new Pop finder is just fine, niot broken in any way.
It is getting, in our list, 100% sane answers.

ALL these things, including 23andme and not quite
as good Ancestry.com, and almost all of Gedmatch, are
also sane and correct.

One simply cannot expect different programs to give
identical answers for everybody. One must expect small bits
and pieces of the "edge" of a person's general region.
One must expect siblings to sometimes test quite differently. In the latter case of course if the parents are
different origins but both pure (e.g. English and
Vietnamese) then the children will in fact all be 50-50.
But if both parents are generic mixed Mexican, even both
being 50% Mayan, 40% Spanish, 5% Jewish and 5% Nigerian, which is possible, then one child could be 0% Jewish and
the other 10% Jewish, or one 70% Mayan and one 30%.

This applies to Europe too.

And different programs, even using the same data, and
give APPARENTLY different answers, exactly as you report,
as still be correct. It just depends on how the database
makes "cuts" between different comparison panels,
and whether they use modern populations (the old Pop Finder)
or "Zombies" (DIYDodecad). Neither is more correct.

Now if the average spot on the map of a purely
European/Jewish/Armenian(etc.)/Turkish person
differs by more than 1000 miles between programs,
then you have a really bad and exceptional case.
But 600 mile error is common and to be expected!

In the case of people who are listed as a mix of Coastal Plain/ Islands (Britain)/Scandinavian/"Euro Peneplan"
and small amounts of Mediterranean, siblings and other close relatives should do very well in the "average position on the map" test. I and my aunt are a good
test of this, and ALL the programs pass the test nicely.
We sometimes test as French, sometimes as English/German/Italian ... but the average is always the same spot in
Normandy, which is wrong by the distance from Calais
to say London.

Doug McDonald

OH! C'mon! That's such a generic answer. I listed just a very basic problem that many users are seeing. But that's not the only problem. There's people who are Jewish, that are no longer showing Jewish, and people who were never Jewish and are now showing Jewish when no one else in their family results does. And then when you can go to 4 other places and have it verified, then I'm sorry, but that's a problem. I realize things like "center", "spatial", "border", but when a company stands out as having wildly different results than the other companies around them, than that really says something. I can take a sample, run it through Ancestry, 23andme, GedMatch, and DNATribes, and come up with almost the same results. But then, come to Population Finder, and find it wildly different. And now with MyOrigins I see that on a basic general level, the results are in line with the others, BUT on the more specific level is where I'm seeing the problems.

So then question......how come Ancestry can tell me British Isles for that same sample above, and 23andme can tell me British Isles, and DNATribes can tell me British Isles, and then when you come here, you can't get that same answer. You get Central European. An explanation of "different data set" isn't good enough. Other companies seem to be doing a far better job of pinpointing DNA than FTDNA is right now. If I run my sample at other companies I get "Portuguese", "Iberian", "Basque", "Spanish" which guess what, I'm half Mexican with Portuguese descendancy. But at FTDNA, Population Finder only gave a super regional level answer, and now with MyOrigins, I'm "Mediterranean Basin" which can include anything from Greece/Turkey all the way to Spain, and northward into Southern France/Germany/Austria. And yes, I understand DNA has no borders. Yes, I'm a product of the many peoples from the Mediterranean basin, so in that sense FTDNA is correct. But many false negatives, many false positives, and other companies being able to pinpoint data more correctly, says a lot about what FTDNA is doing. (And no, I'm not the Irish sample)

I respect FTDNA more than any other company. They have some of the top scientists working for them. And I just love Bennett, he's a wonderful guy. He will answer any question I ask him. I don't know any other CEO that will do that for their companies users. But to me, and many others, MyOrigins is an EPIC FAIL. We expected more, and only got these overly regional descriptions. AGAIN. And to top it off, I'm sorry, but I'm a programmer, and EVERY algorithm is gonna have bugs. Period. And as users, we are finding them. Please don't give us a generalized explanation and ignore them.

PS - saying you only had Orkney at the time, and no specific "England" data set is no excuse. Population Finder was in Beta during its entire run. New data sets could've been added at any time, thus changing the results. But new data sets were never added to the program.

[Ftdnauser]

I certainly don't expect identical answers from each program and I understand it's a complex topic, but when I get ZERO British results (European coastal islands) from myOrigins even though I not only have several known British branches in my tree but also, AncestryDNA are putting my British results as high as 55% something is obviously wrong. Additionally, myOrigins has my Scandinavian (European northlands) and German (European coastal plains) results significantly higher than AncestryDNA (34% and 26% versus 2% and 5%). It's becoming more and more obvious that myOrigins is putting my British results into northlands and coastal plains instead, because many people are reporting the same or opposite problem (people with German or Scandinavian but no British ancestry showing results for coastal islands).

I accept that AncestryDNA could be a little off as well - perhaps they are overestimating my British results and underestimating my German and Norwegian ancestry. And if myOrigins was showing at least SOME coastal islands results, I wouldn't be complaining. But fact of the matter remains that I clearly have British DNA that isn't registering AT ALL on myOrigins and that is a significant issue.

Exactly! Well said, and you, me, and many others agree.

[Ftdnauser]

I can answer this. I was the author of the code
for the original Pop Finder.

There was never anything wrong per se with the Pop Finder
code. It simply HAD to use a database that was available
when it was written, and changing it would have been a nightmare. I changed the version I used three times and
it WAS a nightmare. And this database had to use Orkney instead of England, as there was at the time no
available England database.

We don't yet know how My Origins works, or what they used
as database. I have not looked at a wide variety of people
with it since I don't have access to the needed projects.
I have looked at all the Clan Donald people who have it.
Of course most of these are English/Irish/Scottish, but
there are a very few people who are McDonalds in the male line and are actually Italian, Spanish, and Mexican.

The new Pop finder is just fine, niot broken in any way.
It is getting, in our list, 100% sane answers.

ALL these things, including 23andme and not quite
as good Ancestry.com, and almost all of Gedmatch, are
also sane and correct.

One simply cannot expect different programs to give
identical answers for everybody. One must expect small bits
and pieces of the "edge" of a person's general region.
One must expect siblings to sometimes test quite differently. In the latter case of course if the parents are
different origins but both pure (e.g. English and
Vietnamese) then the children will in fact all be 50-50.
But if both parents are generic mixed Mexican, even both
being 50% Mayan, 40% Spanish, 5% Jewish and 5% Nigerian, which is possible, then one child could be 0% Jewish and
the other 10% Jewish, or one 70% Mayan and one 30%.

This applies to Europe too.

And different programs, even using the same data, and
give APPARENTLY different answers, exactly as you report,
as still be correct. It just depends on how the database
makes "cuts" between different comparison panels,
and whether they use modern populations (the old Pop Finder)
or "Zombies" (DIYDodecad). Neither is more correct.

Now if the average spot on the map of a purely
European/Jewish/Armenian(etc.)/Turkish person
differs by more than 1000 miles between programs,
then you have a really bad and exceptional case.
But 600 mile error is common and to be expected!

In the case of people who are listed as a mix of Coastal Plain/ Islands (Britain)/Scandinavian/"Euro Peneplan"
and small amounts of Mediterranean, siblings and other close relatives should do very well in the "average position on the map" test. I and my aunt are a good
test of this, and ALL the programs pass the test nicely.
We sometimes test as French, sometimes as English/German/Italian ... but the average is always the same spot in
Normandy, which is wrong by the distance from Calais
to say London.

Doug McDonald

I forgot to add a couple other quick examples,


Things like one sibling shows Jewish @10% and yet the brother and both parents show none at all. That's another problem. A Y DNA Jewish Cohen line.......was Jewish in Pop Finder, is Jewish elsewhere, but now, is no longer Jewish in MyOrigins.......One second and another second cousin showing completely different subsets and their common ancestors subset doesn't show in both samples even though everyone is from the same area. Those are problems Dr. McDonald, and I'm hearing of a lot more from other users.

[John McCoy]

I remain unimpressed by the entire effort to detect ethnic origins. The fact that there is so much variation in the results obtained from the various admixture tools tells me that there is not yet a consensus on the data used for comparison, nor probably the methodology as well. Looks to me like many people just try all the tools until they find one that gives them the answer they want, hardly a robust procedure.

I might buy into the methodology if it could be demonstrated that there are statistically significant suites of SNP variants that characterize regional groups of samples from isolated communities where the history and surnames suggest very limited outbreeding. It would help, too, if there were some effort to measure the predictive value of these tools, based on large samples selected by some means other than anecdotal reporting. In other words, I haven't yet seen metrics for the reference populations, nor for the power of the tools.

[Ftdnauser]

I remain unimpressed by the entire effort to detect ethnic origins. The fact that there is so much variation in the results obtained from the various admixture tools tells me that there is not yet a consensus on the data used for comparison, nor probably the methodology as well. Looks to me like many people just try all the tools until they find one that gives them the answer they want, hardly a robust procedure.

I might buy into the methodology if it could be demonstrated that there are statistically significant suites of SNP variants that characterize regional groups of samples from isolated communities where the history and surnames suggest very limited outbreeding. It would help, too, if there were some effort to measure the predictive value of these tools, based on large samples selected by some means other than anecdotal reporting. In other words, I haven't yet seen metrics for the reference populations, nor for the power of the tools.

Which "tools" are you referring to?

[vinnie]

I might buy into the methodology if it could be demonstrated that there are statistically significant suites of SNP variants that characterize regional groups of samples from isolated communities where the history and surnames suggest very limited outbreeding. It would help, too, if there were some effort to measure the predictive value of these tools, based on large samples selected by some means other than anecdotal reporting. In other words, I haven't yet seen metrics for the reference populations, nor for the power of the tools.

Exactly, and in the same spirit as what I posted in the ISOGG FB forum.

"As thankful as I am for Gedmatch, Dienekes, and the others who've contributed there, I don't believe there is any empirical evidence demonstrating that any one of those calculators is more or less accurate than any other, or than any of the testing companies. In other words, while individuals can anecdotally claim that this tool or that tool fits their known origins, the best that most of us can go back on the paper trail is 200-300 years, and we all know the various problems associated with paper trails. So the bottom line is that while a particular analysis may look "sexier" than another, perhaps its the rather dull one that is actually more historically accurate. Let us know when someone invents a reliable time-travel machine so we can get this figured out."

[Ftdnauser]

Exactly, and in the same spirit as what I posted in the ISOGG FB forum.

"As thankful as I am for Gedmatch, Dienekes, and the others who've contributed there, I don't believe there is any empirical evidence demonstrating that any one of those calculators is more or less accurate than any other, or than any of the testing companies. In other words, while individuals can anecdotally claim that this tool or that tool fits their known origins, the best that most of us can go back on the paper trail is 200-300 years, and we all know the various problems associated with paper trails. So the bottom line is that while a particular analysis may look "sexier" than another, perhaps its the rather dull one that is actually more historically accurate. Let us know when someone invents a reliable time-travel machine so we can get this figured out."

Yeahhhhhhh, I'm sorry, but I tell everyone not to use GedMatch, or to take the results on a very general level and not a specific or literal level.

[dtvmcdonald]

I remain unimpressed by the entire effort to detect ethnic origins. It would help, too, if there were some effort to measure the predictive value of these tools, based on large samples selected by some means other than anecdotal reporting. In other words, I haven't yet seen metrics for the reference populations, nor for the power of the tools.

This is anecdotal, mostly, but in my experience, and I
have looked at 5000 (mostly mixed) people, people ENTIRELY from
isolated populations are extremely well identified
if there is the correct reference. For example, Basque
or pure Ethiopians. I can, for example, RELIABLY
tell Oromo Ethiopians from Tigray or Amhara, IF
they really are from the isolated part. I have seen
quite a few of these. But I can't tell, reliably,
Tigray from Amhara.

But VERY few Europeans are from isolated populations.

Doug McDonald

[genis]

I can answer this. I was the author of the code
for the original Pop Finder.

There was never anything wrong per se with the Pop Finder
code.

Doug McDonald

Mr. McDonald,

According to the old PF, my mother had a 44.22% Jewish ancestry with an error +-0.1% while i had ZERO Jewish.

This is no wrong ?

MyOrigins located both us in Turkey which is (at least) geograpichally correct though.

[vinnie]

Yeahhhhhhh, I'm sorry, but I tell everyone not to use GedMatch, or to take the results on a very general level and not a specific or literal level.

I think Gedmatch is the best available at this point because the tools have been developed by different people using (somewhat) different databases and methodologies. When I receive congruent results from a number of different sources, I believe there's some truth to the results, especially when they can be associated with known world and family history, my comments above not withstanding.

[vinnie]

One thing that bothers me is the relatively low number of samples that MO is based on. I'm not a stats whiz, and I can't say that I completely understand how they use the small samples they do to generalize. However, in polling/survey research, it takes 1k+ of randomly sampled subjects to generalize to the population in question. I don't know to what degree having 1k samples from a given population would improve MO, and perhaps that many wouldn't be needed because this isn't about polling attitudes, but I don't understand why FTDNA couldn't dedicate some of its resources to subsidize testing in key areas.

For example, the DNA from only 30 Italians were used for MO. What part(s) of Italy did they come from? Were all four of their lines Italian? How far back could they document their genealogy? While not being privy to FTDNA's finances, I would think that an investment in testing 1k Italians would be worth it; offer it for $50 U.S. instead of $99 to the desired number of people. If they could also test a similar number of Ashkenazim and Sephardim, we might be able to come to some reasonable conclusions about the shared heritages. I know that this situation is very complicated, but I just don't trust the statistical inferences that are being made based on such small sample sizes, despite the fact that both PF and MO are fairly accurate for my parents and me, at least in relationship to how they compare to the Gedmatch tools.

In the meantime, I'll continue to encourage DNA testing among my Italian community, and perhaps those with access to Sephardic communities can do the same.

[dtvmcdonald]

One thing that bothers me is the relatively low number of samples that MO is based on. I'm not a stats whiz, and I can't say that I completely understand how they use the small samples they do to generalize. However, in polling/survey research, it takes 1k+ of randomly sampled subjects to generalize to the population in question.
For example, the DNA from only 30 Italians were used for MO.

You need 1000 Italians to tell ACCURATELY where somebody is from in ITALY. Well, maybe 600.

But to tell over the WORLD, you only need 1000 people
over the whole world to the same RELATIVE accuracy.