View Single Post
Old 12th November 2017, 07:37 PM
Frederator Frederator is offline
FTDNA Customer
Join Date: Jul 2010
Posts: 819
I think the main problem with ethnicity calculators is knowing which level is best to aggregate at.

The only calculator that I am impressed with, just based on personal experience, is the MDLP Project K16 Modern 4-factor. It pretty much nailed my unphased and phased paternal kits, although I'm having trouble making sense of my brother's kit and my own phased maternal readings.

I'm pondering the notion that my success with MDLP is attributable to the fact that it employs a 4-factor format, which comes closest to mirroring my specific genealogy, which is comprised of lines from two major ethnic groups (Irish, German) and five distinct sub-groups: Baden-Wuerttemberg, Swiss, Ukrainian, Bohemian, and Irish, which due to the small size of that country I mostly don't bother to analyze on a more granular basis any more.

It's not perfect, but being able to make clear connections between each of the 4 MDLP factors and my paper trail gives me a greater (maybe illusory?) sense of confidence in the algorithm's rationality.

But I have real problems with my brother's results. At Gedmatch I can only make a clear correspondence for one grandparent on each side of our family. I even feel like I'm really reaching to do so for my own maternal phased results, although my mother's own unphased results match pretty nicely, and somehow, despite the greater complexity of my personal admixture, so do my own unphased results.

On the other hand, MyOrigins has just been an unmitigated failure for all of my family's kits. I'm not really surprised given the complexity of the admixtures involved. But this first taste of (relative) success with Gedmatch encouraged me to believe that there are some fundamental design principles to be derived from the contrast. Maybe like dis-aggregating at an appropriate level.

My thinking was that, in order to build on that foundational observation, I should try to look at 4-figure results for mono-ethnic individuals, like these Irish-born people, and see if there are any patterns in the way their individual components relate to one another.

Their one-factor analyses all return them as Irish, Scottish or, less often, English, which seems good enough as far as I'm concerned.

A handful of these people consistently show one factor out of four as Belarusian or Polish--usually paired off with another factor as French or Basque, with the remaining two Scottish or Irish. I know from those with posted pedigrees that they're pretty vanilla Catholics from the west with no hint of anything as exotic as the Baltics.

At the moment I don't know if those are just random noise or if there really are some important sub-national regional patterns at work. And it bothers me a bit. I don't have a unified theory on how ethnic calculators should work.

I can't have much confidence in my own obviously inaccurate one-factor results or the dubious 4-factor results of these Irish-born people if there is no consistent, objective basis for knowing which model to apply. It seems like nonsense to say that people in Europe should all use one-factor models. There is plenty of ethnic diversity within each country's population, although confusingly, not within my pilot sample.
Reply With Quote