Attachment 145813
Attachment 145814
Attachment 145815
My MENA is pretty high, but is not that bad.
Printable View
Attachment 145813
Attachment 145814
Attachment 145815
My MENA is pretty high, but is not that bad.
Nice models for Miroiço (closest ancient sample to modern portugueses):
Attachment 145818
Attachment 145819
I did two 2-way models with Ireland EBA and Spain EBA.
Attachment 145906
This one was with drb234's rights plus some additions:
Attachment 145907
Attachment 145908
Attachment 145909
The second model:
Attachment 145910
References used:
Attachment 145911
Attachment 145912
Edit: did a third model with Ireland EBA and Spain BA (with the above references and Spain EBA in the rights)
Attachment 145913
I mean, is a nice model. But, theres not too many outgroups? Especially with very similar profiles like La Tene and Bell Beaker, France IA and England EIA, etc . . . Is good to base your riight pops in the ones used in articles, especially from the one taht study your background populations.Quote:
Originally Posted by Gannicus
https://www.nature.com/articles/s41586-022-05247-2
I have run into situations where the standard errors are higher when less outgroups are used. It doesn't happen all the time though.Quote:
Originally Posted by Jingle Bell
This model has the same outgroups as the previous model's that used 30 references. With the exception of Spain BA or Spain EBA not included. SE's are nice.
Attachment 145914
So far all of the various outgroup lists I have used are from AndreiDNA, others like drb234, and my own modifications to those lists. Someone earlier in the thread linked a reference list from the Genarchivist forum too. And I experimented with that.
AndreiDNA did a video on the Durotriges. In the video, there was a qpAdm model he did with WBK07. He scored 54% Latvia LN CordedWare and Scotland N 46% pval=0.654.
Here's what I get:
Attachment 145923
I also did a model with CordedWare + Scotland Megalithic:
Attachment 145924
Wales IA + Denmark IA
Attachment 145932
Same rights were used for this one:
Attachment 145947
Attachment 145948
I'm about to go out to eat some Italian. This gave me an idea. Can I model myself with Italy Imperial samples?
Attachment 146003
How Scythian are the Tatars? AT1 model
Attachment 146015
I had AndreiDNA do a few Celto-Germanic qpAdm models for me with Admixtools2.
Durotrigian + FranceGrandEst IA
Attachment 146024
Wales IA + Durotrigian
Attachment 146025
Durotrigian + Norway IA
Attachment 146028
France Yonne IA + Denmark IA
Attachment 146030
France GrandEst IA2 + Denmark IA
Attachment 146031
Spain IA Celt + Durotrigian + Denmark IA
Attachment 146032
Edit: For those viewing, Durotrigians are IA Celtic Britons.
Tried a few Models of my own; they seem pretty similar to my G25 results, very insular Celtic, along with 10-20% Germanic. But I'm finding it impossible to get a model with low statistical error. What do I do to fix this?
https://i.postimg.cc/8sGtDqgx/Screenshot-(11).png
https://i.postimg.cc/yWKvBwzq/Screenshot-(12).png
Someone on Twitter did this one for me. My Steppe is pretty high.
https://i.postimg.cc/d7BDCkDj/45069a...9cd16e847b.jpg
Attachment 146035
Ranking z scores for f4 models on russians.
Peak z score (Best model): Mordovian as base, belarusian as source, mbuti as outgroup
Worst z score (worst model): Finnish as base, Bulgarian as source, Dinka as outgroup
For fun also I asked AndreiDNA to model me with NUE001 (Old Kingdom Ancient Egyptian). Can it be done?
Attachment 146044
Attachment 146046
Attachment 146047
Attachment 146045
The answer is yes it can be done. Quite incredible I think.
I've been able to do it with the TIP Egyptians also:
Attachment 146041
Attachment 146042
AndreiDNA did a model with EGY TIP for me:
Attachment 146043
Edit: as a reminder 0.05 is the conservative cutoff for p values in qpAdm.
Attachment 146049
France_Aude_IA.SG + Denmark_IA.SG + England_LIA_o.AG
Probably used too few references but I’m just getting started lol.
Try Scotland MIA LIAQuote:
Originally Posted by Birchy
Try these references from drb234:
Attachment 146053
Attachment 146054
I have gotten good results for Celto-Germanic models with that list.
100% Wales MIA
Attachment 146055
references:
Attachment 146056
100% Norway IA
Attachment 146057
Attachment 146058
Attachment 146059Quote:
Originally Posted by Gannicus
Scotland MIA looks stronger indeed
deleted
Quote:
Originally Posted by Albannach
Do you have the reference list you used for the first two models ?
Looks good :thumb001:Quote:
Originally Posted by Birchy
Did you use drb's references?
Attachment 146060 < What I get with the references that I linked.
Would you mind trying to see if you can model yourself with some ancient Egyptians?
This one is apparently and ancient Egyptian buried in Lebanon during the Achaemenid period:
Attachment 146061
try with these references:
Attachment 146062
Attachment 146063
References:
Attachment 146064
Attachment 146065
Attachment 146066
References for this one:
Attachment 146067
Attachment 146068
Yep exact copy and paste because I only used 5 references before lol. I do agree with your theory that at least some of the Celtic in both our models is being absorbed by the Iron Age Germanic.Quote:
Originally Posted by Gannicus
I'll definitely try some of those models for you tomorrow. :thumb001:
https://i.postimg.cc/KRVBPYDb/Screenshot-(14).pngQuote:
Originally Posted by Xavier727
Which right pops he used? 50% of SE is just crazy, most of these models are basically a fail bcs the sources have a z-value < 3. Generally, a SE < 5 is ideal, but - depending of how much overlapping are the samples - until 15% is ok, anythng above this - especially way above - is too unsure to get any conclusion off.Quote:
Originally Posted by Gannicus
Thanks — is that all of it, it looks as though there's only part of the list, (7/30), do you have the rest of it ?Quote:
Originally Posted by Albannach
That's all I added. Are you supposed to have 30 reference populations for it to be accurate? I thought 30 was just the maximum allowed. I read that the optimal range for reference populations is 5-15, but maybe that explains why I have such a high statistical error on any model I run. I'm new to QpAdm, so still trying to find my bearings.Quote:
Originally Posted by Xavier727
Quote:
Originally Posted by Albannach
Ah got it, my bad, I thought you might’ve had more than 7. Yeah, 30’s just the max, it’s not required, 5–15 is totally fine, though I usually find my SE drops a bit when I add a few more reference populations.
You could ask him directly about his view on the SEs and model context. Sometimes a model has to be interpreted cautiously rather than dismissed outright, but p-values are still a key indicator of model validity. Also, I have done several Celto-Germanic models of myself with cleaner SE's. These are telling the same story for the most part.Quote:
Originally Posted by Jingle Bell
Attachment 146070
OpenAI:
1. Model validity (global fit)
p-value = 0.761
This is a strong pass. There is no statistical reason to reject the model.
In qpAdm terms, anything comfortably above ~0.05 is acceptable; values this high indicate the references jointly explain the target very well.
2. Standard errors (SE)
SE ≈ 0.254 (≈25%)
Yes, this is not “tight,” but it is not pathological either—especially for:
Two closely related Iron Age NW European sources
Likely overlapping ancestry and shared drift
Small reference sample sizes
This is exactly the kind of situation where SE inflation is expected and does not invalidate the model, provided the p-value remains strong (which it does).
3. Z-scores
Z ≈ 1.8–2.1
This is normal for highly correlated sources.
Low |Z| here reflects collinearity, not model failure.
Importantly:
qpAdm does not require |Z| > 3 for ancestry components
|Z| thresholds are more relevant for f4 tests of asymmetry, not mixture weights between near-identical populations
4. Interpretation of the result
~45.6% Durotrigian
~54.4% France_GrandEst_IA2
This is a sensible, biologically coherent outcome:
Both are Celtic / NW European Iron Age groups
The model is testing relative affinity, not exotic ancestry
The weights are stable and symmetric
Nothing here suggests a “failed model.”
Bottom line (forum-ready)
High p-value → model is valid
Moderate SE → expected given source overlap
Z < 3 → not disqualifying in this context
Interpretation → reasonable and consistent
If someone is calling this “basically a fail,” they are over-applying heuristics (SE/Z rules of thumb) without respecting qpAdm’s actual decision criterion: overall model fit.
I showed it the 50% Standard Error model:
This second model reinforces the same point and, again, does not support the claim that these are “basically fails.”
Here is how it stands on qpAdm’s actual criteria:
1. Global model fit
p-value = 0.432
This is a clean pass. It is comfortably above rejection thresholds and indicates the reference pair adequately explains the target.
2. Standard errors (SE)
SE ≈ 0.495 (~50%)
This is large, but it is expected in this specific setup:
Durotrigians and Norway_IA are both NW European Iron Age populations
They share substantial steppe + farmer ancestry
qpAdm struggles to resolve proportions cleanly when sources are this genetically close
Large SE here reflects source collinearity, not a broken model.
3. Z-scores
Z ≈ 1.0
Again, this is normal for near-identical or highly correlated sources.
Low |Z| in this context indicates qpAdm cannot strongly distinguish between them—not that the model is invalid.
Crucially:
|Z| > 3 is not a requirement for qpAdm mixture weights
Z is not the pass/fail criterion; p-value is
4. Interpretation (what this model is actually testing)
It is testing:
“Does your ancestry fall between southern British IA and Scandinavian IA?”
The answer is yes, and qpAdm cannot strongly prefer one over the other—hence the ~50/50 split and inflated SEs.
That is a meaningful result, not noise.
Bottom line (concise forum framing)
p = 0.432 → model passes
Large SE → expected due to overlapping NW European sources
Low Z → reflects collinearity, not failure
Result → confirms intermediate positioning within NW European IA space
Calling this a “fail” misunderstands how qpAdm behaves when modeling closely related populations.
This one looks good
Attachment 146076
Wales_IA.AG + France_Aisne_IA.SG + Norway_IA.SG
What reference populations did you use?Quote:
Originally Posted by Birchy
Attachment 146077Quote:
Originally Posted by Albannach
Attachment 146078
Attachment 146079
These are Drb’s, they seem to give good results for NW Euros.
With Drb's rights. Not a good combo for me.
Attachment 146083
Attachment 146085
One of your Egyptian models did work on me, over two turned out false :(
Try theseQuote:
Originally Posted by Birchy
Attachment 146086
Attachment 146087
References used for those:
Attachment 146088
Also did you try SFI 43 + Scotland BlackIsle ?
Edit: here is what these models look like for me in G25
Target: Gannicus_MergedFile_officialDavidski_scaled
Distance: 0.0300% / 0.03002681
61.4 Scotland_BlackIsle_IA:KD001__AD_569__Cov_54.79%
29.4 Denmark_IA.SG
9.2 Egyptian_Late_Kingdom_Period:EGY_Late_Period:JK291 1
Target: Gannicus_MergedFile_officialDavidski_scaled
Distance: 0.0294% / 0.02938280
61.7 Scotland_BlackIsle_IA
28.8 Denmark_IA.SG
9.5 Egyptian_LateKingdomPeriod_JK2134_JK2911_avg
Scotland_BlackIsle_IA:KD001__AD_569__Cov_54.79%,0. 135449,0.125926,0.061094,0.052326,0.034776,0.02063 8,-0.00658,-0.000231,0,-0.003098,-0.013641,0.004496,-0.01442,-0.011973,0.031351,0.007823,-0.014342,-0.000887,0.005154,0.011756,0.008984,0.004204,-0.000246,0.001084,-0.002994
Denmark_IA.SG,0.127482,0.131003,0.066185,0.066538, 0.042931,0.0214745,0.0027025,0.0129225,0.00859,-0.010934,-0.003816,0.0080175,-0.0043855,-0.004542,0.0217835,0.007027,-0.0131685,0.008488,0.005091,0.0075035,0.0086725,0. 007172,0.001171,0.013315,0.0015565
Egyptian_LateKingdomPeriod_JK2134_JK2911_avg,0.050 082,0.146744,-0.042426,-0.1183805,-0.0020005,-0.049782,-0.015628,-0.004731,0.0474495,0.006196,0.0090125,-0.011465,0.029732,-0.005849,-0.0008145,-0.0020555,-0.0074315,-0.0019635,-0.0035195,0.0120055,0.005615,0.0016695,0.0036355,0 .003916,-0.0005985
Egyptian_Late_Kingdom_Period:EGY_Late_Period:JK291 1,0.053497,0.141159,-0.04714,-0.115635,-0.004001,-0.04518,-0.020681,-0.003231,0.046427,0.005103,0.017863,-0.012739,0.0333,-0.011836,-0.001629,0.004375,-0.00691,0.001267,-0.005279,0.017133,0.005241,0.002102,0.002218,-0.005302,-0.007424
Also did you try SFI 43 + Scotland BlackIsle ?[/QUOTE]
I tried it with the exact one you did, I think it's ID was labelled SFI 43.
Attachment 146089Quote:
Originally Posted by Gannicus
Quote:
Originally Posted by Birchy
Have you got your G25 cords yet? If so, would you mind sharing your raw scaled modern pop results ?
Attachment 146091
Attachment 146092
https://imgur.com/a/BglRBTO
AT1 models and F4 runs on a 8000 year old sample from Serbia
Looks like I’ve got more east med than you. That’s what my goal was. It was to get other northwest Euros to try these models for a control.Quote:
Originally Posted by Birchy
Hey Xavier, would you mind trying the model I did in post #354 with the same references? I’m curious to see how it turns out for you.Quote:
Originally Posted by Xavier727