GENEALOGY-DNA-L Archives

Archiver > GENEALOGY-DNA > 2004-01 > 1072969526


From: (Raymond Whritenour)
Subject: [DNA] New DNAPrint format and NA results
Date: Thu, 1 Jan 2004 10:05:26 -0500 (EST)


Folks:

I suppose this new format is "new," in that SA possibilities are now
given for all, on the triangle plot and graphic chart. However, it
appears pretty clear that no improvement in the statistical accuracy of
the test has been gained by use of the new algorithm.

Thank you, Elizabeth, for making your results available for all to see.
The triangle chart is the best indicator of the real test results, as
always, and Elizabeth has a first confidence interval which will allow
her to say that she has a high probability of EA ancestry (whatever "EA"
really means, which is another matter). But, the margin of error is
about + or - 12.5% from her MLE (4%-29%). She has "possible" NA and SA
minor ancestry; but, it's not so possible as the graphic chart shows,
because that chart shows no possibilities below 0%. This is misleading.
The triangle plot clearly shows that her NA and SA possibilities plot
out just as far below 0% as they do above 0%. Therefore, there are
margins of error of about + or - 7-8% from her MLE for these ancestries.
The graph would have us believe that her possiblities of NA and SA
ancestry is 0-7% and 0-8%, when they are really about -7 to +7% and -8
to + 8%. In other words, there are as many possibilities "at 0%" as
there are "above 0%.

As for Ernie's remarks: I have a family tradition of Cherokee ancestry.
My father's paternal grandmother was said to be Cherokee. My DNAPrint
test results showed 84% IE, 13% NA and 3% EA. I don't have any Mexican
ancestry, so you can't make DNAPrint's "EA" out as "Eastern Woodlands
American Indian." Things just don't sort out that easily. My mother
has NO American Indian ancestry, but her test came out 90% IE and 10%
NA. There's a lot more wrong with this test than re-formatting the
charts is going to fix! I was hoping the new algorithm would show some
improvement in the test's accuracy. It doesn't seem that it has.
There's obviously some systematic errors, as John Chandler tried to show
them. I don't think their NA sample is wrong, in itself, other than
being too small, but their IE sample population is very suspect, in my
mind. Anyway, I remain unimpressed by this "new" stuff.

Ray Whritenour

This thread: