Davidski said... Matt, How would you interpret these sets of f4 and D statistics? The f4 stats are from Hofmanova et al., while the D stats were run by me. The first set of D stats uses the highest quality Anatolia Neolithic sample from Barcin from Mathieson et al. and CHG genotypes from Fu Q et al., and the second uses the same Barcin sample plus CHG genotypes from Jones et al. Also, keep in mind that, as far as I can tell, the Barcin genomes from Hofmanova et al. and Mathieson et al. date to the same period. f4 Corded_Ware_LN Bar8 Satsurblia Khomani -0.0367 -8.145 f4 Corded_Ware_LN Bar8 Kotias Khomani -0.0193 -3.437 f4 Spain_MN Bar8 Satsurblia Khomani -0.0327 -5.385 f4 Spain_MN Bar8 Kotias Khomani -0.0182 -3.136 versus... D Corded_Ware_Germany BAR20_I0709 Satsurblia Khomani 0.0215 4.299 D Corded_Ware_Germany BAR20_I0709 Kotias Khomani 0.0205 4.408 D Iberia_MN BAR20_I0709 Satsurblia Khomani -0.0003 -0.06 D Iberia_MN BAR20_I0709 Kotias Khomani -0.0017 -0.339 ... D Corded_Ware_Germany BAR20_I0709 Satsurblia2 Khomani 0.0224 4.38 D Corded_Ware_Germany BAR20_I0709 Kotias2 Khomani 0.0226 5.168 D Iberia_MN BAR20_I0709 Satsurblia2 Khomani 0.0068 1.222 D Iberia_MN BAR20_I0709 Kotias2 Khomani 0 0.004 Clearly, something's horribly wrong. If I made a mistake, my apologies. But I'm pretty sure I didn't make any mistakes. I checked the datasets that I'm using for consistency with the f4 and D stats published in Mathieson et al. and Fu et al., so I can say with confidence that my D stats should not be much different from correctly run f4 stats using the same ancient samples. June 8, 2016 at 7:06 PM
Matt said... @ Davidski, yeah I see what's going on there with the D stats giving a result we would expect from previous work - Anatolia Neolithic and Iberia_MN equally related to CHG, Corded Ware more to CHG - with the stats from this paper being different - Bar8 being more related to CHG than Iberia_MN, and Bar8 even more strongly related to CHG relative to Corded_Ware, also implying Iberia_MN more related to CHG than Corded Ware is. I don't know that there's anything about f4 vs D stats themselves that would explain that difference, and as you say, yours are consistent with the previously published. This is really stuff that should have been in and the resolved in the early print. That's the whole point of the process! June 9, 2016 at 12:13 AMUpdate 12/06/2016: Here are f4 stats using the same data as for the D stats above. They look basically the same as the D stats.
Corded_Ware_Germany BAR20_I0709 Satsurblia Khomani 0.002073 4.294 Corded_Ware_Germany BAR20_I0709 Kotias Khomani 0.001997 4.402 Iberia_MN BAR20_I0709 Satsurblia Khomani -0.00003 -0.06 Iberia_MN BAR20_I0709 Kotias Khomani -0.000157 -0.34 Corded_Ware_Germany BAR20_I0709 Satsurblia2 Khomani 0.0021 4.388 Corded_Ware_Germany BAR20_I0709 Kotias2 Khomani 0.002031 4.971 Iberia_MN BAR20_I0709 Satsurblia2 Khomani 0.000613 1.221 Iberia_MN BAR20_I0709 Kotias2 Khomani 0.000002 0.004