I think the biggest obstacle to accurate ethnicity calculators is identifying an appropriate level of aggregation at which to perform the analysis.
Based on personal experience, with known ancestry from 7 micro-regions within the last 8 generations, I've noticed that the most accurate results are returned by Gedmatch's 4-component analyses of my phased results. That is, when my parental contributions are analyzed separately for their actual 3- or 4- component admixtures.
It's not perfect. Nobody receives a perfectly even 25% contribution from all four grandparents, so having to assign a single label to a 25% component that may be 1/3 Silesian and 2/3 Irish is going to be tricky. But having multiple scenario results to compare will clarify the actual regional components.
I think the relative success of this 4-component analysis in my case is that it roughly reflects the mathematical level of my parents' ethnic diversity. It probably works reasonably well for people less diverse, and much less well for people who are more diverse.
I don't understand how MyOrigins is expected to return a meaningful result when they don't commit to a discrete number of factors in their analysis. They tell you up front that the base genealogical product, Family Finder, can only be expected to detect ancestry for five generations. What then would be the point of reporting ethnicity in 1% increments, which would imply a lookback of 6 or 7 generations (e.g., 2^6=64, 2^7=128 ancestors)? Especially considering the relaxation of matching precision implied by moving from strictly genealogical criteria to an ethnic one.
Based on personal experience, with known ancestry from 7 micro-regions within the last 8 generations, I've noticed that the most accurate results are returned by Gedmatch's 4-component analyses of my phased results. That is, when my parental contributions are analyzed separately for their actual 3- or 4- component admixtures.
It's not perfect. Nobody receives a perfectly even 25% contribution from all four grandparents, so having to assign a single label to a 25% component that may be 1/3 Silesian and 2/3 Irish is going to be tricky. But having multiple scenario results to compare will clarify the actual regional components.
I think the relative success of this 4-component analysis in my case is that it roughly reflects the mathematical level of my parents' ethnic diversity. It probably works reasonably well for people less diverse, and much less well for people who are more diverse.
I don't understand how MyOrigins is expected to return a meaningful result when they don't commit to a discrete number of factors in their analysis. They tell you up front that the base genealogical product, Family Finder, can only be expected to detect ancestry for five generations. What then would be the point of reporting ethnicity in 1% increments, which would imply a lookback of 6 or 7 generations (e.g., 2^6=64, 2^7=128 ancestors)? Especially considering the relaxation of matching precision implied by moving from strictly genealogical criteria to an ethnic one.
Comment