are unscaled coords shifted a lot towards East Asian?

[Nurzat]

what's your experience with scaled vs unscaled? I've tried some simple models and unscaled gives me very high East Asian % compared to scaled, but much better fits at the same time.


scaled:

Target: Nurzat
Distance: 3.7779% / 0.03777948
61.6 Lithuanian_VZ
34.6 Sardinian
3.8 Mongolian


unscaled (much better fit):

Target: Nurzat
Distance: 1.6909% / 0.01690858
61.8 Lithuanian_VZ
26.2 Sardinian
12.0 Mongolian

--------------------------------------------------------------------

scaled:


Target: Nurzat
Distance: 2.1686% / 0.02168623
58.0 Lithuanian_VZ
22.4 Sardinian
14.6 Georgian_Imer
3.0 BedouinB
2.0 Mongolian
0.0 Nganassan
0.0 Paniya


unscaled (much better fit again):

Target: Nurzat
Distance: 1.5295% / 0.01529517
57.6 Lithuanian_VZ
22.8 Sardinian
8.8 Georgian_Imer
7.8 Mongolian
3.0 BedouinB
0.0 Nganassan
0.0 Paniya


Target: Nurzat
Distance: 1.5418% / 0.01541760
58.2 Lithuanian_VZ
23.4 Sardinian
9.2 Georgian_Imer
5.8 Han_Jiangsu
3.4 BedouinB
0.0 Nganassan
0.0 Paniya


Target: Nurzat
Distance: 1.5122% / 0.01512217
58.2 Lithuanian_VZ
23.2 Sardinian
9.2 Georgian_Imer
6.2 Japanese
3.2 BedouinB
0.0 Nganassan
0.0 Paniya


Target: Nurzat
Distance: 1.4988% / 0.01498847
55.0 Lithuanian_VZ
21.6 Sardinian
8.2 Georgian_Imer
6.2 Korean
6.0 Irish
3.0 BedouinB
0.0 Nganassan
0.0 Paniya

---------------------------------------------------------

scaled:

Target: Nurzat
Distance: 2.1367% / 0.02136691
54.2 Lithuanian_VZ
20.8 Sardinian
14.0 Georgian_Imer
5.8 Irish
3.2 BedouinB
2.0 Mongolian
0.0 Nganassan
0.0 Paniya

[knez01]

Unscaled will always give you better fits, but that doesn't make it more accurate, I get loads of East Asian too which I believe is absurd.

[Ion Basescul]

what's your experience with scaled vs unscaled? I've tried some simple models and unscaled gives me very high East Asian % compared to scaled, but much better fits at the same time.


scaled:

Target: Nurzat
Distance: 3.7779% / 0.03777948
61.6 Lithuanian_VZ
34.6 Sardinian
3.8 Mongolian


unscaled (much better fit):

Target: Nurzat
Distance: 1.6909% / 0.01690858
61.8 Lithuanian_VZ
26.2 Sardinian
12.0 Mongolian

--------------------------------------------------------------------

scaled:


Target: Nurzat
Distance: 2.1686% / 0.02168623
58.0 Lithuanian_VZ
22.4 Sardinian
14.6 Georgian_Imer
3.0 BedouinB
2.0 Mongolian
0.0 Nganassan
0.0 Paniya


unscaled (much better fit again):

Target: Nurzat
Distance: 1.5295% / 0.01529517
57.6 Lithuanian_VZ
22.8 Sardinian
8.8 Georgian_Imer
7.8 Mongolian
3.0 BedouinB
0.0 Nganassan
0.0 Paniya


Target: Nurzat
Distance: 1.5418% / 0.01541760
58.2 Lithuanian_VZ
23.4 Sardinian
9.2 Georgian_Imer
5.8 Han_Jiangsu
3.4 BedouinB
0.0 Nganassan
0.0 Paniya


Target: Nurzat
Distance: 1.5122% / 0.01512217
58.2 Lithuanian_VZ
23.2 Sardinian
9.2 Georgian_Imer
6.2 Japanese
3.2 BedouinB
0.0 Nganassan
0.0 Paniya


Target: Nurzat
Distance: 1.4988% / 0.01498847
55.0 Lithuanian_VZ
21.6 Sardinian
8.2 Georgian_Imer
6.2 Korean
6.0 Irish
3.0 BedouinB
0.0 Nganassan
0.0 Paniya

---------------------------------------------------------

scaled:

Target: Nurzat
Distance: 2.1367% / 0.02136691
54.2 Lithuanian_VZ
20.8 Sardinian
14.0 Georgian_Imer
5.8 Irish
3.2 BedouinB
2.0 Mongolian
0.0 Nganassan
0.0 Paniya

Are you using a custom model or is this a run against the whole spreadsheet?

[Nurzat]

Are you using a custom model or is this a run against the whole spreadsheet?

I use only what you see, this is why you see the zeroes as well. but it's relevant, it's the usual source populations Eurogenes and others use (Lithuanian, Sardinian, Georgian, Irish/Orcadian etc). when I run it against the whole population list I get the same amount of East Asian, of which 4% is Japanese and 2-3% is shared between different other East Asian populations. so unscaled coordinates seem, from all possible components, to drag us all a bit towards East Asian

[Ion Basescul]

I use only what you see, this is why you see the zeroes as well. but it's relevant, it's the usual source populations Eurogenes and others use (Lithuanian, Sardinian, Georgian, Irish/Orcadian etc). when I run it against the whole population list I get the same amount of East Asian, of which 4% is Japanese and 2-3% is shared between different other East Asian populations. so unscaled coordinates seem, from all possible components, to drag us all a bit towards East Asian

Unscaled coordinates always had problems with smaller components, which is why they were scaled with eigenvalues on day one by a user in the comment section on Eurogenes and they pretty much became the standard since.

[Lucas]

what's your experience with scaled vs unscaled? I've tried some simple models and unscaled gives me very high East Asian % compared to scaled, but much better fits at the same time.


scaled:

Target: Nurzat
Distance: 3.7779% / 0.03777948
61.6 Lithuanian_VZ
34.6 Sardinian
3.8 Mongolian


unscaled (much better fit):

Target: Nurzat
Distance: 1.6909% / 0.01690858
61.8 Lithuanian_VZ
26.2 Sardinian
12.0 Mongolian

--------------------------------------------------------------------

scaled:


Target: Nurzat
Distance: 2.1686% / 0.02168623
58.0 Lithuanian_VZ
22.4 Sardinian
14.6 Georgian_Imer
3.0 BedouinB
2.0 Mongolian
0.0 Nganassan
0.0 Paniya


unscaled (much better fit again):

Target: Nurzat
Distance: 1.5295% / 0.01529517
57.6 Lithuanian_VZ
22.8 Sardinian
8.8 Georgian_Imer
7.8 Mongolian
3.0 BedouinB
0.0 Nganassan
0.0 Paniya


Target: Nurzat
Distance: 1.5418% / 0.01541760
58.2 Lithuanian_VZ
23.4 Sardinian
9.2 Georgian_Imer
5.8 Han_Jiangsu
3.4 BedouinB
0.0 Nganassan
0.0 Paniya


Target: Nurzat
Distance: 1.5122% / 0.01512217
58.2 Lithuanian_VZ
23.2 Sardinian
9.2 Georgian_Imer
6.2 Japanese
3.2 BedouinB
0.0 Nganassan
0.0 Paniya


Target: Nurzat
Distance: 1.4988% / 0.01498847
55.0 Lithuanian_VZ
21.6 Sardinian
8.2 Georgian_Imer
6.2 Korean
6.0 Irish
3.0 BedouinB
0.0 Nganassan
0.0 Paniya

---------------------------------------------------------

scaled:

Target: Nurzat
Distance: 2.1367% / 0.02136691
54.2 Lithuanian_VZ
20.8 Sardinian
14.0 Georgian_Imer
5.8 Irish
3.2 BedouinB
2.0 Mongolian
0.0 Nganassan
0.0 Paniya

It looks impossible to have such difference with so distant component. If it was any test outside G25 with unscaled, where you have more than 10% East Asian?
I'm loosing confidence in G25 when I see this...

[Kaspias]

Not necessarily apparently. It is just stupid. Scaled is always better while looking for admixture proportions. Unscaled works only in single populations, in some cases.


Target: Kaspias_scaled
Distance: 3.4530% / 0.03452985
60.2 Cypriot
30.6 Lithuanian_VZ
9.2 Nganassan

Target: Kaspias
Distance: 2.3177% / 0.02317701
68.6 Cypriot
26.4 Lithuanian_VZ
5.0 Nganassan

[RyoHazuki]

In my case yeah. I always get like 5% Jomon.