GENEALOGY-DNA-L Archives

Archiver > GENEALOGY-DNA > 2006-11 > 1162671199


From: "Sasson Margaliot" <>
Subject: Re: [DNA] Neolithic J2 and E3b in Britain? Maybe not.
Date: Sat, 4 Nov 2006 22:13:19 +0200
References: <042401c6fe97$5ce8b830$6401a8c0@Precision360><000001c6ff28$2aaaf720$4001a8c0@BigMem2>
In-Reply-To: <000001c6ff28$2aaaf720$4001a8c0@BigMem2>


John

Thank you for such a detailed discussion.

> However, seeing it was my table that was being (mis)quoted I thought I
> had better step up and take some heat.

> 1) I used Zhiv et al's 2004 interpolated estimate of effective mutation
> rate with ages calibrated from archaeological sources (0.0007 is the
> value I used). There was a reason for using the value, while the figures
> are interesting in themselves, I was EXPLICITLY using exactly the same
> method as described by Moore et al 2006 to age what is now known as the
> M222+ R1b subclade and demonstrate that it was markedly older then their
> estimate. To date I know no other attempt to use a consistent
> methodology (even if flawed and subject to caveats) across all the
> haplogroups.

You have said that you use the Zhivotovsky 2004 rate, I therefore
understood that you are using their low rate of 0.0007

> 2) Zhiv et als estimate was then explicitly adjusted for differences in
> the mutation rate of the markers used. For the 37 markers used by FTDNA
> the conversion rate was 2.09 so the "effective" mutation rate used for
> the 37 markers used was 0.00146. This is detailed on the page in
> question.

This is 2.09 faster than the rate of 0.0007

> 3) ...

> 4) Moving on to Sasson's message who considered the values were fudged
> and divided them by 3. Well Sasson please be aware these are not the
> rates that Zhiv et al estimated. By my calculation 3/2.09 = 1.44. So we
> have a fudge factor of 1.44 in your terms.

Dividing the estimate by 1.44 and applying the double SEM to get 95%
confidence level, still keeps the age I suggest for J2 and J1 within
reasonable range.

> As an aside Zhiv et al used
> 25 years as a unit of generation, if he had used 32.5 (based on SMGF
> figures) which is the value on which the "accepted" father son rates are
> implicitly calculated...

The father son rates were calculated based on observations in the
recent generations (32.5 years), but in the pre-historic times 25
years is more appropriate.

In fact switching from 25 to 20 "improves" the fudge factor from 1.44 to 1.8

> Now, I am the first to admit there is a real difference
> between quoted father son rates and those derived from archaeological
> calibration,

I'm trying to figure out how the age estimates without fudge factor
could correspond to popularion replacements and relocations known from
history, rather than blaming almost everything on imaginary
"paleolithic" and "neolithic" population

> but if you venture down this road I suggest you use values
> appropriate to the case at hand.

Thanks for pointing out that the actual fudge factor was 1.44 and not 3.0

> At the VERY LEAST you have to adjust
> for the differences in generation length.

The generation length issue works both ways: not only there is no
basis for extrapolation of the average generation length of 30 to
pre-historic times, but, even more importantly, the *variance* in the
generation length makes the 95% confidence interval even larger (to
allow for possible deviations in generation length in particular
lines)

> 5) Sasson the SEM are provided in the same table, so you can calculate
> the 95% confidence interval of the TMRCA if you wish (and as an aside
> the way I calculated them is very conservative).

I did now, but I want to point out that this is the statictical SEM,
valid if all the background assumptions are correct, so there is an
additional layer of variation resulting from uncertainty of these
background assumptions

> 6) Sasson then moves on to say that between F versus the rest is the
> true branch split of the tree from apes.

I do not accept the concept of "split from apes". In any case, the
word "GENEALOGY", which is the subject of this List, applies only to
humans.

What I'm saying is that the MRCA of modern humans (tested so far) can
very well have the genetic state of F, with AE derived by M89-, then
AB derived by M168-

> Now I have not double checked
> the original published calculations and sequence alignments that placed
> the split between A and B but I will accept the published values until a
> detailed alternative case is made. Gareth specifically reviewed more
> recent literature.

The starting point was that if finding seven non-unique SNPs in D2 is
not as strange as I think it is, then the argument against alternative
rooting is also not very strong.

> More important to me though, is the AB age estimate
> from my table is 29500 yrs (SEM 5499) and F is 73856 (SEM 26040) and was
> based on very small sample of individuals in each case.

Doesn't SEM already reflect the smallness of the sample?

The 65% confidence intervals are not even overlapping...

> Using the values produces a t value for a test of a difference in the branch ages of
> ~1.666 which is ...ahem... not significant.

How is this t value calculated and how is the difference (F 2.5 times
AB) not significant?

Sasson


> Now these calculations could
> be improved, but Sasson's statement if solely based on values listed in
> my table, should be treated with some caution.

This thread: