GENEALOGY-DNA-L Archives

Archiver > GENEALOGY-DNA > 2005-03 > 1110402365


From: (Raymond Whritenour)
Subject: Re: [DNA] Re ; DNA Print test
Date: Wed, 9 Mar 2005 16:06:05 -0500
In-Reply-To: john.chandler@alum.mit.edu (John Chandler)'s message of Wed, 9Mar 2005 15:15:56 -0500 (EST)


Okay. Here is my response to John Chandler's comments on my post about
the DNAPrint's minority ancestry probablilties, in capital letters.

Ray wrote:

[[No. I'm afraid it is you who are confused, John. The 91% probability I
refer to has absolutely nothing to do with B.J.'s 91% "IE" result. The
two identical percentages are just a coincidence.]]

Ok, you succeeded in confusing me by making no effort to distinguish
between the two 91s. Don't you think it would be all the more confusing
to BJ?

JOHN: YOU'RE A PARADIGM OF GRACE, IN DEFEAT!

[[DNAPrint Genomics
Inc. states that a majority ancestry "European" with an "East Asian"
result of 13% (on the 2.0 test), has a 95% probability of having "some
level of real East Asian ancestry"; and further..."]]

As I pointed out elsewhere, this statement is based on faulty
interpretation of the simulations and is simply false.

AS I STATED, THIS IS THE COMPANY'S LINE - NOT MINE.

[[By DNAPrint genomics, Inc.'s definition of "deep ancestry," Hun and
Mongol forbears do represent "deep ancestry." It's their test, so I'm
using their terminology.]]

Well, it's not their ancestry. The point is that there is absolutely no
basis whatsoever for taking a test result and estimating the age of the
admixture from the result. The age can only be guessed at, based on
knowledge of the ancestors. The company does not have such knowledge and
should therefore not speculate about things beyond their ken. To put it
bluntly, they can't even tell whether a minority contribution came from
the mother, from the father, or from both, unless they test the mother
and the father, too.

IN MY OPINION, IF THERE IS "EAST ASIAN" ANCESTRY IN EUROPEANS, THE
LIKELIHOOD THAT THE VAST MAJORITY OF IT IS FROM DEEP ANCESTRY, RATHER
THAN FROM RECENT ADMIXTURE, IS A GIVEN. WHETHER OR NOT THAT ANCESTRY IS
FROM HUNNISH OR MONGOL FORBEARS, OR OTHER, MORE DISTANT "EAST ASIAN"
ANCESTORS, IS IMMATERIAL TO THE POINT. I USED THEM ONLY AS
POSSIBILITIES, AS WAS CLEAR FROM MY LANGUAGE.
[[NotĀ if previous statements by John Chandler, to this list, are
correct. I got that 88% figure directly from you.]]

But you misinterpreted it. I'll answer your other message separately.

OKAY.

[[Does anybody on this list, but you, understand the parenthetical
remark you made above? If so, could that person please explain it to the
rest of us.]]

What I wrote was:

No. The black ring is the 50% confidence contour. (More precisely, it is
the max/2 probability density contour, but it just so happens that these
are the same thing for a two-dimensional normal distribution.)

The parenthetic remark was intended for those who know the difference
between a confidence contour and a probability density contour and might
be puzzled by my assertion before the remark. I'm not sure that anyone
who doesn't already know these things would want to find out, but here
is a brief discourse:

Probability density is the probability per unit area on the plot. The
DNAprint algorithm computes the likelihood of regularly spaced outcomes
(at intervals of 1% in each ethnic fraction), and these likelihoods are
therefore proportional to the probability density. When the company says
the black ring is where the outcomes are twice as unlikely as the MLE,
they mean that the likelihood (i.e., the probability density) is one
half of the maximum (i.e., "max/2"). In principle, there need be no
connection between this concept and the 50% confidence interval, but the
two-dimentional normal, or Gaussian, distribution has some very nice
mathematical properties. The Gaussian probability density is exp(-0.5
x^2), where "x" is the distance in "natural" units from the MLE. Given
that distribution, it turns out that the 50% confidence contour is
exactly the same as the 1/2-max probability density contour. I could run
through the proof of this theorem, but I suspect that most readers
wouldn't be interested.

I CAN ASSURE YOU THAT ANYBODY WHO DIDN'T ALREADY KNOW THESE THINGS STILL
DOESN'T!!!

RAY WHRITENOUR



This thread: