[MyOrigins Strange Results Log]

[malchik]

Hello,

Is there any list available where people report 'strange results' from MyOrigins?

I thought it would be interesting to refer to common issues by a code or something, for example MOSR MyOriginsStrangeResults...

Browsing through the fora I come accross issues like the following below:

#MOSR001 - Only one parent has X% of an ethnic group, but son/daughter has actually higher percentage.

#MOSR002 - Only one parent has 50% of an ethnic group, but son/daughter has 0 (zero) percentage.

#MOSR003 - Individual has double digit percentage of an ethnic group, but his matches to this group are ridiculously scarce, e.g. less than 5%.

#MOSR004 - Individual has small percentage of an ethnic group, but none of the parents have it.

#MOSR005 - Individual has 100% of an ethnic group, but matches have 0% of it.

And so on...

Which other 'apparently strange results' would you like to share here?

[malchik]

More...

#MOSR006 - Individual with well known very diverse ancestry has simplified (lumped together) ethnic group composition, missing one or more ancestry groups.

#MOSR007 - Individual has different ethnic groups shown by other DNA tests.

And so on...

Which other 'apparently strange results' would you like to share here?

[malchik]

More...

#MOSR006 - Individual with well known very diverse ancestry has simplified (lumped together) ethnic group composition, missing one or more ancestry groups.

#MOSR007 - Individual has different ethnic groups shown by other DNA tests.

And so on...

Which other 'apparently strange results' would you like to share here?

#MOSR008 - Individual has results showing only 99%. Where is the missing 1%?

[192971]

#MOSR008 - Individual has results showing only 99%. Where is the missing 1%?

It is the rounding error. FTDNA could avoid these complaints by forcing the sum always be 100%, but it would be faking and less accurate.

[malchik]

It is the rounding error. FTDNA could avoid these complaints by forcing the sum always be 100%, but it would be faking and less accurate.

Thanks for replying.

The fact that others get 100% in their results does not necessarily mean that FTDNA is not faking or that the results are more accurate.

If you run the results into any GEDMatch calculator, you will notice how much of the percentages break down into many other ethniticies, which in FTDNA are simply lumped together within a single label.

Each sample contains the whole 100% DNA from each customer, who in turn expect 100% revelation about his/her ethnic background.

Since the accuracy of such estimates is NOT EVEN CLOSE to 1%, there is nothing else to expect other than speculative results (specially for that particular missing part).

FTDNA has no excuse for displaying 99% other than ridiculously badly written software. Didn't decode 1%? Simply state 1% not-identified.

It is that simple.

[malchik]

#MOSR001
Only one parent has X% of an ethnic group, but son/daughter has actually higher percentage.

#MOSR002
Only one parent has 50% of an ethnic group, but son/daughter has 0 (zero) percentage.

#MOSR003
Individual has double digit percentage of an ethnic group, but his matches to this group are ridiculously scarce, e.g. less than 5%.

#MOSR004
Individual has small percentage of an ethnic group, but none of the parents have it.

#MOSR005
Individual has 100% of an ethnic group, but matches have 0% of it.

#MOSR006
Individual with well known very diverse ancestry has simplified (lumped together) ethnic group composition, missing one or more ancestry groups.

#MOSR007
Individual has different ethnic groups shown by other DNA tests.

#MOSR008
Individual has results showing only 99%. Where is the missing 1%?

MOST RECENT ONE!
#MOSR009
Individual has results showing only below 100% European descent, but Ancient Origins show 100% European. A contradiction?

You gotta love this mess...

[192971]

If you run the results into any GEDMatch calculator, you will notice how much of the percentages break down into many other ethniticies, which in FTDNA are simply lumped together within a single label.

Have you noticed that, for example, under label "European" there might be many population components in myOrigins?

Each sample contains the whole 100% DNA from each customer, who in turn expect 100% revelation about his/her ethnic background.

Since the accuracy of such estimates is NOT EVEN CLOSE to 1%, there is nothing else to expect other than speculative results (specially for that particular missing part).

FTDNA has no excuse for displaying 99% other than ridiculously badly written software. Didn't decode 1%? Simply state 1% not-identified.

It is that simple.

OK, then show me your own simple solution which satisfies your demands. Tell me the percentages which values for components A, B and C and their sum FTDNA should show to a customer, when component A is 33.44444...% (up to the maximum accuracy of the used floating point variable type in the software) and component B is 33.33333...% and C is 33.22222...%

Your solution may utilize any rounding accuracy coarser than 1% level you complained of being too accurate for such estimates. Also explain why your solution is more truthful than FTDNA's simplest possible 1% accurate solution A = 33%, B = 33%, C = 33% and sum = 99%, please.

I know that FTDNA's programmers (or at least one of them) suck in programming, but in this particular case FTDNA has done it right. (Probably because the right solution is the most obvious one.)

[malchik]

Have you noticed that, for example, under label "European" there might be many population components in myOrigins?

No, really!?

You missed my point...

I repeat.

If you run the results into any GEDMatch calculator, you will notice how much of the percentages break down into many other ethniticies, which in FTDNA are simply lumped together within a single label [i.e. COMPONENT].

For example, in FTDNA you have Western Central Europe, but in GEDMatch you might have this COMPONENT (from the label EUROPEAN) broken down even further into many other ethniticies... that is it, no big deal.

OK, then show me your own simple solution which satisfies your demands. Tell me the percentages which values for components A, B and C and their sum FTDNA should show to a customer, when component A is 33.44444...% (up to the maximum accuracy of the used floating point variable type in the software) and component B is 33.33333...% and C is 33.22222...%

Maybe you missed my point...? I dont know.

Please tell me the accuracy of their test by a) stating the total number of markers tested and b) the minimum number of markers considered in a comparison to be able to associate them with each ethnicity.

I dont think that every single marker can be exclusively associated to an ethnicity so that you never have 0.0000001% of anything.

[Boudicca1]

I get 13% Eastern European but my parents get 0%. I did an Ancestry autosomal transfer, my Dad's is a 23andme transfer and my Mum did FF a few years ago.

Dad's results 92% European 37% British, 35% Scandinavian, 11% Western/Central European and 9% Southern Europe and 8% Jewish.

Mum's results: 97% European 41% British, 34% Western/Central European and 22% Scandinavian and 3% Central and South Asian.

My results: British Isles 40%
Western and Central Europe 29%
Eastern Europe 13%
Southern Europe 11%
Scandinavia 7% spot the obvious mistake!!!

Dad gets 1.1% Eastern European, 0.3% MENA and 0.1% AJ at 23andme, Mum gets 0% for all three groups.
Mum gets 8% Eastern European at Ancestry, Dad hasn't tested, I get 4% (I also get 3% Jewish but Mum gets none). How can I show a population that neither parent does?!

[malchik]

I get 13% Eastern European but my parents get 0%. ... How can I show a population that neither parent does?!

_________________________Father..Mother....You

British______________________37%;41%;40%
Scandinavian_______________35%;22%; 7%
Western/Central European_11%;34%;29%
Southern Europe___________ 9%; 0%;11%
Central and South Asian____ 0%; 3%; 0%
Jewish_______________________ 8%; 0%; 0%
Eastern Europe_____________ 0%; 0%;13%

Good question.

I would try the GEDMatch phasing test and investigate further.

[Boudicca1]

_________________________Father..Mother....You

British______________________37%;41%;40%
Scandinavian_______________35%;22%; 7%
Western/Central European_11%;34%;29%
Southern Europe___________ 9%; 0%;11%
Central and South Asian____ 0%; 3%; 0%
Jewish_______________________ 8%; 0%; 0%
Eastern Europe_____________ 0%; 0%;13%

Good question.

I would try the GEDMatch phasing test and investigate further.

What should I be looking for? Is there a specific test? I have phased my kit against both parents at Gedmatch. Someone on another thread has suggested it's actually Scandinavian rather than EE.

[malchik]

What should I be looking for? Is there a specific test? I have phased my kit against both parents at Gedmatch. Someone on another thread has suggested it's actually Scandinavian rather than EE.

You can read about that here:

http://isogg.org/wiki/Phasing

I think your EE is not real, but a result of the scrambled markers from your Scandinavian markers that somehow get read as EE.

[192971]

in GEDMatch you might have this COMPONENT (from the label EUROPEAN) broken down even further into many other ethniticies...

You mean in the Oracles? Well myOrigins do not have them. But if you run all the admix calculators in Gedmatch, you will notice how wildly the results vary. But we all like Gedmatch because there we can choose the admix result that pleases us the most. I think FTDNA cannot do it like Gedmatch does.

I dont think that every single marker can be exclusively associated to an ethnicity so that you never have 0.0000001% of anything.

Whatever the upper limit or accuracy of the percentages there is no way out of rounding errors. Gedmatch admix calculators also give as sum of components something near 100%, but gives component percentages with two decimals, so that the rounding error is almost invisible. There is no fundamental difference between FTDNA and Gedmatch in this sense.

[malchik]

You mean in the Oracles? Well myOrigins do not have them. But if you run all the admix calculators in Gedmatch, you will notice how wildly the results vary. But we all like Gedmatch because there we can choose the admix result that pleases us the most. I think FTDNA cannot do it like Gedmatch does.

Are we even talking about the same thing...? Nobody likes GEDMatch 'because there we can choose the admix result that pleases us the most'. It is a fact that GEDMatch gives us a more detailed breakdown, therefore it offers MORE information (quantitative), not BETTER information (qualitative). Both GEDMatch (including ALL ITS CALCULATORS) and FTDNA differ in the quantity of information, but the quality of the information is similar. FTDNA can in this sense be considered as the simplest GEDMatch available.

I think FTDNA cannot do it like Gedmatch does.

Do you mean offering more information for free?

Whatever the upper limit or accuracy of the percentages there is no way out of rounding errors.

You missed the point again... I bring you back to the original question in a different form. Suppose you have test results 99.99% Zulu. Then you ask yourself what is the real information meaning behind 0.01% of any ethnic percentage?

(a) If you know it is in fact meaningless, then you can add this 0.01% to the result and display 100% Zulu WITHOUT ANY CONSEQUENCES TO INFORMATION CREDIBILITY. IT WOULDN'T MATTER BECAUSE DISPLAYING 99.99% IS EQUAL TO DISPLAYING 100%.

(b) If you know it is in fact meaningful, then you might actually be 99.98% Zulu and 0.02% SomethingElse, 99.99% 0.01% SomethingElse or actually 100% Zulu. IT WOULD MATTER BECAUSE DISPLAYING 99.98% IS DIFFERENT THAN DISPLAYING 100%, and you should then assume some ERROR TOLERANCE EQUAL OR LESS 0.01%.

In case (a), any junior programmer would write a simple 'ceiling function' to force 99.99% to become 100%.

In case (b), any junior programmer would write a simple 'truncate function' to force a.bcX% to become a.bc% and label the delta to 100% as 'not-identified'.

Case (b) is very likely not true and this leads us back to my original statement...

Since the accuracy of such estimates is NOT EVEN CLOSE to 1%, there is nothing else to expect other than speculative results (specially for that particular missing part). FTDNA has no excuse for displaying 99% other than ridiculously badly written software. Didn't decode 1%? Simply state 1% not-identified.

Case (a) is true, both in GEDMatch and FTDNA.

Gedmatch admix calculators also give as sum of components something near 100%, but gives component percentages with two decimals, so that the rounding error is almost invisible. There is no fundamental difference between FTDNA and Gedmatch in this sense.

The fact that others get 100% in their results does not necessarily mean that FTDNA is not faking or that the results are more accurate.

[malchik]

#MOSR001
Only one parent has X% of an ethnic group, but son/daughter has actually higher percentage.

#MOSR002
Only one parent has 50% of an ethnic group, but son/daughter has 0 (zero) percentage.

#MOSR003
Individual has double digit percentage of an ethnic group, but his matches to this group are ridiculously scarce, e.g. less than 5%.

#MOSR004
Individual has small percentage of an ethnic group, but none of the parents have it.

#MOSR005
Individual has 100% of an ethnic group, but matches have 0% of it.

#MOSR006
Individual with well known very diverse ancestry has simplified (lumped together) ethnic group composition, missing one or more ancestry groups.

#MOSR007
Individual has different ethnic groups shown by other DNA tests.

#MOSR008
Individual has results showing only 99%. Where is the missing 1%?

#MOSR009
Individual has results showing only below 100% European descent, but Ancient Origins show 100% European. A contradiction?

You gotta love this mess...