About Proto-Germanic

[Rodoorn]

A little autosomal math. I believe I posted qpAdm models a while ago with Swedish Margaryan Iron groups as targets, and sources a combination of Volga_Oka_IA and Sweden_LN. Like this one:
left pops:
Stockholm_IA
Sweden-LN-orig
VolgaOka_IA

best coefficients: 0.907 0.093
totmean: 0.907 0.093
boot mean: 0.910 0.090
      std. errors: 0.102 0.102
     
     
fixed pat wt dof chisq tail prob
            00 0 15 11.906 0.686098 0.907 0.093
            01 1 16 12.808 0.686722 1.000 0.000
            10 1 16 104.543 4.84389e-15 0.000 1.000
best pat: 00 0.686098 - -
best pat: 01 0.686722 chi(nested): 0.902 p-value for nested model: 0.342286

Of course, the lack of decent-quality Swedish Bronze Age genomes is sorely felt, but we make do with what we have. I had the crazy idea to see what it would be like if I added the Germans from Unetice (Esperstedt) as sources. I say that this idea is far-fetched because inevitably we must expect to see standard errors explode. But on the other hand, crazy ideas can be worthwhile, if you keep in mind that they are crazy. To have something of good quality as a source and not to use post-imputation data, I replaced the Allentoft group with my old friend ber1M. So here it is:

Stockholm_IA
Sweden_LN_ber1
Germany_EBA_Unetice
VolgaOka_IA

fixed pat  wt  dof    chisq      tail prob
          000  0    14    14.301        0.427568    0.809    0.119    0.072
          001  1    15    14.633        0.47819    0.802    0.198    0.000
          010  1    15    14.430        0.493173    0.908    0.000    0.092
          100  1    15    27.687      0.0236155    0.000    0.956    0.044
          011  2    16    14.944        0.528753    1.000    0.000    0.000
          101  2    16    27.841        0.033033    0.000    1.000    0.000
          110  2    16  106.012    2.55702e-15    0.000    0.000    1.000
best pat:          000        0.427568              -  -
best pat:          010        0.493173  chi(nested):    0.130 p-value for nested model:        0.718658
best pat:          011        0.528753  chi(nested):    0.514 p-value for nested model:        0.47361

If we only look at the tail_probs the best models are those without Unetice (010 and 011). If we now look for the nested model with the best p-value, it is undoubtedly 010, the one which only keeps ber1M and Volga_Oka as sources. By the way, 0.71 is a really good p-value (think of all those studies that let pass p-values in the 0.051 range!). I'll let you compare the tail_probs of single source models 011 and 101. I think it's clear.

This is imo a clear map from Ph2ter. 

What is clear is that your BAC favorite scores very red in the Unetice koiné from VandKilde (2017). Baltics: nada.