GENEALOGY-DNA-L Archives

Archiver > GENEALOGY-DNA > 2006-01 > 1138544962


From: Doug McDonald <>
Subject: Re: [DNA] autosomal testing -- BGA test DNAPrint 2.5 and Euro 1.0tests offered by DNAPrint.com
Date: Sun, 29 Jan 2006 08:29:22 -0600
References: <43DA63A5.8050307@neyersoftware.com> <43DA699A.2010908@kerchner.com> <REME20060127214408@alum.mit.edu> <4e393532de42e8cba0dfd832fdf90cbc@midcoast.com> <REME20060128180438@alum.mit.edu>
In-Reply-To: <REME20060128180438@alum.mit.edu>


John Chandler wrote:

>> could they not also improve
>> usefulness of results by increasing the number of categories into which
>> results are sorted?
>
> The results are not "sorted" into categories. What DNAprint does is
> estimate the percentages of admixture of the reference populations
> in each test result. The statistical uncertainty of the estimates
> increases dramatically as the number of populations included rises.
> The problem is that the more populations are considered, the easier
> it is to "imitate" one population by some blend of the others.

This is not necessarily true with a wide enough and deep enough database
to work on, as the academic papers (and some non-academic) have shown.
For a given dataset, there is a "right" number of categories. If you
use too many, the mathematics of finding them simply can't
do the job ... there is not enough data. If you try too few, the
mathematics is forced to blend categories that are in fact different.

Science, 298, p2381 (2002) addresses this.



Doug McDonald


This thread: