GENEALOGY-DNA-L Archives

Archiver > GENEALOGY-DNA > 2010-02 > 1266078432


From: "Ken Nordtvedt" <>
Subject: Re: [DNA] Y Tree STR Mutations can not be counted
Date: Sat, 13 Feb 2010 09:27:12 -0700
References: <mailman.4572.1266045490.2099.genealogy-dna@rootsweb.com><6A16FA701F2F4350A90C379D3AA5DC21@anatoldesktop>


You are lucky this morning to get a reply from a lister who you now have
repeatedly labeled dishonest. It probably won't happen too often in the
future.
But some comments are inserted below within [[[....]]] symbols.

----- Original Message -----
From: "Anatole Klyosov" <>

>>[[[ Since what you are counting is something akin to GDs between
>>haplotypes, it is not bad for making age estimates, but it is not a
>>basic for making the statistical confidence intervals you quote
>>for those estimates. (earlier message)]]]
>
> First, "something akin" is not a scientific term. [[[ In your description
> of what you do below you divide the number of STR changes you count from
> your defined base haplotype by the number of haplotypes showing changes
> from the base haplotype. Leaving aside how you or others count changes,
> those who count average GDs from the base haplotype would probably divide
> the 1788 by 500. Afterall, those 14 haplotypes have a branch history from
> the founder equally long as the 486, so their outcomes are just as much
> evidence of the the mutational potency and elapsed time of mutational
> opportunties as the 486 haplotypes which ended up with change. That's
> what was meant by "something akin to GDs". But I use variances, not GDs,
> so I don't know if your variation on GDs is improvement or regression.
> <GD> eventually has a non-linear dependence on elapsed generations, so
> your variation may or may not be incorporating some of that
> non-linearity?? I suspect your method is akin enough not to matter much in
> most cases if the TMRCA is not so large. ]]

> Second, I know that "it is not bad for making age estimate". That is what
> I
> do.
> Third, would you care to explain why it is "not a basic... for those
> estimates"? [[As I said above, your variation on the GD method works for
> the age estimates pretty well; it is the statistical confidence intervals
> for those age estimates which do not rest on a solid basis. See below for
> further elaboration. ]]]
>
> I have tried to help you out and gave a specific example. I have asked you
> for an honest a direct answer, and even thank you in advance for it.
> Unfortunately, I do not see in your reply either one.

[[ As I mentioned in my earlier detailed message, age estimates themselves
for GDs or variances from assumed founding haplotype (your base haplotype)
do not depend on the demographic details of the tree which emerges from the
founder --- other than the total elapsed generations. Each generation of
the tree makes an identical contribution to total variance of the final
haplotypes, although the early generations make their contribution in few
big hunks, while the later generations make their contributions in many
small hunks. So the age estimate can be made without knowing the
demographic history.

But the statistical confidence interval of the age estimate does depend on
the demographic history of the tree. This is known by both analytics and
simulations. Given the same founding (base) haplotype and the same final
500 haplotypes, the statistical confidence interval could be quite different
under these two scenarios: either the tree population grew at slow rate in
its early generations and then a more rapid rate in the late generations, or
vice versa. The early, less populated, generations of the tree contribute
more heavily to the total statistical confidence interval than do the later,
more populated generations. The more the tree lingers in its low population
state, the larger the final statistical confidence interval. Since you
don't know that demographic history, you can't evaluate the statistical
confidence interval. You ask me to do so: I can't as well because I don't
know the demographic history, either. All I can do or you can do is assume
a demographic history and then evaluate a statistical confidence interval
for the assumed demographic history. ]]



This thread: