GENEALOGY-DNA-L Archives

Archiver > GENEALOGY-DNA > 2003-11 > 1067733267


From:
Subject: Re: [DNA] DNAPrint majority % accurate but not minority %
Date: Sat, 1 Nov 2003 19:34:36 -0500 (EST)
References: <18635-3FA3F6FF-3230@storefull-2176.public.lawson.webtv.net> <000501c3a0a7$432c38a0$0a0110ac@mminshel>
In-Reply-To: <000501c3a0a7$432c38a0$0a0110ac@mminshel>(marys2cats@earthlink.net)


Mary wrote:
> I have been reading the posts about DNAPrint with interest. I was going to
> try to convince my sole living relative from my father's line to be tested,
> but...

I'm wondering why you would seek to test some relative before testing
yourself. I suspect you have pigeonholed the DNAprint test in the
wrong category. It doesn't focus on the paternal line, and it doesn't
make any use of the Y chromosome. Furthermore, if this relative you
mentioned isn't a VERY CLOSE relative, then his DNAprint results will
be mostly independent of yours.

> given the controversy over BGA results,

Actually, the controversy has died down. There isn't anyone who now
asserts that the existing test supplies meaningful "high-precision"
results. Even the DNAprint web site has enough weasel-words (and has
always had them) that the implied claims of accuracy are not backed up
in the least. What they have always claimed rightfully is that the
test is a good indicator of majority ethnicity. Their striking
success with the Louisiana serial killer is a good example of what
they can do. The only remaining controversy is over what multiplier
to use for the inherent statistical uncertainty in their results --
should it be 1X, 2X, or 3X? The company itself supplies a plot of the
results with confidence contours corresponding to the 1X, 1.5X, and 2X
multipliers.

If, in the future, they succeed in calibrating their results (e.g., by
comparison among enough test subject all over the world), then they will
be justified in pointing to the 1X level of uncertainty. Until then,
the risk of systematic errors cannot be dismissed. As for the expected
near release of version 2.5, that may indeed reduce the statistical
uncertainty, but not by anything close to the ratio of markers used
(nor even close to the square-root of that ratio).

John Chandler

This thread: