Archiver > GENEALOGY-DNA > 2009-12 > 1260752965

From: "Anatole Klyosov" <>
Subject: Re: [DNA] R-U152 and R-L21 on the European Continent
Date: Sun, 13 Dec 2009 20:09:25 -0500
References: <>

> From: "Tim Janzen" <>
> I certainly do take confidence intervals seriously. The confidence
> levels aren't particularly large for the 24 slow marker option, even for
> old
> nodes. ... Note that in one of my recent > messages I gave a TMRCA
> estimate for the node for haplogroups A and B using 50 markers.
> The 95% confidence interval was relatively narrow (3070 years),
> but the TRMCA (35,245 years) was clearly wrong and we all know that even
> after adding 3070 years to the estimate, that you are still far lower than
> the true TMRCA. In contrast, the 95% confidence interval for the 24 slow
> markers was 22,000 years. As you may recall, the TMRCA estimate was
> 77,576
> years.
>For the calculations I did I only had 9 67-marker haplogroup A haplotypes
and 2 67-marker haplogroup B haplotypes.

>The average mutation rate for the remaining 24 markers is .00040.

Dear Tim,

First, the 95% confidence interval could not possibly be less that 20%
with your 35,245+/-3,070 years) in that situation, unless you have used some
unrealistically "fine" assumptions.
77,576+/-22,000 seems to be O.K. for 95% confidence interval
for your system only if the error margin for the mutation rate constant
(0.00040) is
only 10%.
Last, but not least, you are not serious with 77,576 (!) years, aren't you?
I mean, those five digits... with such a precision...

Let's see. In order to obtain TMRCA 35,000 or 78,000 ybp with two of
haplotypes using the slowest 24 markers (that is, two of 24-marker
haplotypes) with the
given mutation rate (0.00040) you should see 12 and 19 mutations,
respectively in those
two haplotypes. This would result in (two sigma, or 95% confidence interval)
of plus-minus
35% and 30%, respectively, provided, again, that your mutation rate constant
has the margin of error of only 10%.

For 9 of 50-marker haplotypes (again, relatively slow markers, I assume,
with a mutation
rate constant, say, 0.001), you will see about 280 and 440 mutations to get
35,000 and 78,000
ybp, respectively. This will give you - at 95% confidence interval - a
margin of error of 20.87%
and 20.56%, respectively, that is approximately 35,000+/-7,300 and
78,000+/-16,000 ybp.
Again, with only 10% inaccuracy in the mutation rate constant.


Anatole Klyosov

This thread: