GENEALOGY-DNA-L Archives

Archiver > GENEALOGY-DNA > 2003-07 > 1057084380


From: "John F. Chandler" <>
Subject: Re: [DNA] Ancestry test - a must-read article URL
Date: Tue, 1 Jul 2003 14:33 EDT
In-Reply-To: w_s_arthur@yahoo.com message <20030701031523.68524.qmail@web13309.mail.yahoo.com> of Mon, 30 Jun 2003 21:15:11 -0600


Steve wrote:
>Exactly what does that exercise you describe prove? Is that a standard way of
> defining errors for all types of triangle plots? If so, how is it that the DNA
> Print folks don't see that?

First of all, it is not simply an "exercise" -- it is fundamental to the
whole process. Indeed, it is something that needs to be done BEFORE
doing the very first test (using simulated data). In science, the
value of a parameter is meaningless without a statement of its
uncertainty, and you can't even write down the value properly until you
know the uncertainty as well. To give an example, suppose the
uncertainty on a particular value is +/- 25, and suppose the result of
the measurement/calculation is 0.448 (i.e., what you see on your
calculator as the final answer). When you report the result, you do
NOT say 0.448; you say 0 +/- 25. If, instead, the uncertainty were 0.25,
you could report 0.45 +/- 0.25, or you could be more realistic and
dispense with the nearly meaningless extra digit and simply report
0.4 +/- 0.2.

Second, the triangle plot is a clever trick for putting three quantities
into a 2-dimensional graph, but it is not a standard as such, and the
exact form of the graphic representation is completely irrelevant to
the fundamental question you are required to ask (i.e., what's the
uncertainty of this value?). As it happens, the uncertainty can indeed
be read off quite easily from the format presented by DNAprint. That's
all that matters. Obviously, the folks at DNAprint can see the
uncertainties for themselves, too, when they choose to look.

> If you think they're intentionally disregarding
> something obvious to any mathematician, and misleading their customers, do you
> think you could prove it in court?

You're joking! Why would I want to prove such a thing in court? And who
would care? Are you planning on suing the company for $100,000,000 for
the pain and suffering caused by your result? (I didn't think so.) From
Ray's quotaion of the user's guide, it appears that the company actually
makes a full disclosure of the truth to anybody who takes the time to
read the fine print. Also, if you look closely at the really misleading
statements they make, both in public and in private, you can always see
an "out". When they say the uncertainty of their measurements is 2-3%,
they can claim with perfect justification that they are really talking
about the uncertainty due to their possible measurement errors (and I
hope they are being generous in that 2-3% figure, because that's not an
impressive achievement). If you look at the really disgusting blurb
that Malcolm quoted today (where they say that they can give you the
"precise" admixtures), you can see that they say "precise" instead of
"accurate". To a scientist, that is a dead giveaway, and it leaves them
with a perfect excuse. They are selling precision instead of accuracy.

> class-action lawsuit and bring down the company, which is apparently making a
> lot of money off of false claims (if you're right).

I think you misjudge the scale of this operation. Let's suppose they
snag a few thousand customers per year -- that's only on the order of
$1,000,000 revenue, and they have to meet a payroll and pay off the
capital costs of expensive lab equipment and rent. I suppose a lawsuit
could damage them severely, but I can't imagine any lawyer jumping into
the case. Besides, they're only trying to make a buck by selling what
people want to buy.

>If you're so sure....well, let me put it this way, if I were you, I would feel
> morally obligated to set up a web page showing my calculations, and warning
> people against wasting their money.What do you think?

No, that's not my job. What's more, as I said earlier, they really do
make a full disclosure on their own. Well, not quite full. As far as
I can tell, they have never admitted that the percentages they quote
are just a mathematical abstraction which logically includes the
possibility of negative values. Indeed, from what I hear, the software
they use for producing the results and plots is a package they got from
outside, so it's quite possible they've never given it any thought.

Now that you mention it, though, I have in effect "set up a web site"
within the archives of this list. What more could you ask?

John Chandler

This thread: