About population averages in Gedmatch and Vahaduo

[reboun]

In Gedmatch and Vahaduo, population averages are calculated by taking the numeric averages of the results from a certain ethnicity. However, how accurate is to take averages of DNA results of two or more human-beings? In genetically diverse populations, taking the averages and presenting it to Gedmatch and Vahaduo would be really meaningless. Say, 10 samples from ethnicity X is obtained but those 10 people are really different from each other in terms of DNA. The average of these 10 DNA samples would not be really close to neither of these 10 people and therefore taking the average would not be accurate at all.

What should be done?
In my opinion, the best solution is not to take average but represent each individual result separately and name as X1, X2, X3 and so on. Here, X is the name of the person's ethnicity.

Also, you cannot take the "average" of human-beings since humans are not numbers.

[gixajo]

You can do both things, they are not mutually exclusive.

Anyway I agree that averages can present some "problems" because individual samples are screened to be validated to form part of a really representative average, by doing clusters , watching positions on PCAs,comparing results of admixture in "standard models" etc...and individuals considered "non-representative", "possibly mixed", "outliers" are eliminated from the group with which the average will be make.

I see a problem there, because everyone has a different criterium considering what is the essence of the representativeness of a certain group.

if the person who does averages has any bias (we all have it to a greater or lesser extent, consciously or unconsciously, I mean those who deliberately cheat) , doing this the averages can be relatively directed so that when interacting with other averages in a model, that model will give the results that we want to show for a certain group, and it would not show a reality as objective as possible, but rather a subjective reality, subordinated to the bias of the creator.

Since that possibility is real, we shouldn't take all of this too seriously.

[Luke35]

You can do both things, they are not mutually exclusive.

Anyway I agree that averages can present some "problems" because individual samples are screened to be validated to form part of a really representative average, by doing clusters , watching positions on PCAs,comparing results of admixture in "standard models" etc...and individuals considered "non-representative", "possibly mixed", "outliers" are eliminated from the group with which the average will be make.

I see a problem there, because everyone has a different criterium considering what is the essence of the representativeness of a certain group.

if the person who does averages has any bias (we all have it to a greater or lesser extent, consciously or unconsciously, I mean those who deliberately cheat) , doing this the averages can be relatively directed so that when interacting with other averages in a model, that model will give the results that we want to show for a certain group, and it would not show a reality as objective as possible, but rather a subjective reality, subordinated to the bias of the creator.

Since that possibility is real, we shouldn't take all of this too seriously.

Brilliant post, gix

Modeling using individual samples is perhaps underutilized, on anthrofora.

[Jingle Bell]

When i use heterogenous populations like Imperial Romans i always select some representative individuals and use they, its better than use a average of a population made of diferents levels of levantine, Italic, Greek, Anatolian and North African admixture as it was representative in real life (A Average of a truly heterogenous population means nothing).
But in for example more homogenous populations like for example Al-Andalus, where of 12 samples, just 2 or 1 can rly overlap with Christian samples i prefer use a average.
https://i.imgur.com/k6XJKjH.png


Depends on the occasion and ur use