Thanks for the link. It's important that consumers understand these results are merely estimates based upon undisclosed data and assumptions.
I can understand the variance between results from different companies. Each company uses their own database and their proprietary algorithms which are disclosed to the public. They also don't disclose the degree of error associated with each estimation.
The variations within each company is probably a good estimation of the calculation errors inherent to their data and algorithms. I would like to see the variations broken down according to magnitude. One would expect that results corresponding to smaller percentages should have the largest systematic errors.
The author failed by omission to address Admixture Analysis, like that freely available at gedmatch.com, which I personally prefer. Admixture analysis uses fully disclosed details, plus it gives users an opportunity to see how model-dependent their results really are. The commercial labs chose the model themselves, without providing any details to consumers.
