GENEALOGY-DNA-L Archives

Archiver > GENEALOGY-DNA > 2008-07 > 1216253741


From: "Ken Nordtvedt" <>
Subject: Re: [DNA] calibrating for common ancestor - QUESTION?
Date: Wed, 16 Jul 2008 18:15:41 -0600
References: <mailman.89372.1215791041.8264.genealogy-dna@rootsweb.com><16EF8B32-C14A-46BE-B2E8-98BBF49A1BBE@vizachero.com><00b401c8e751$19198810$6400a8c0@Ken1><E7842801-7B3F-4C69-8F5C-41608A763CE7@vizachero.com><f3f05ce80807160844t1751d9f7gd25e0e53d2185982@mail.gmail.com><8E2E4B64-AD30-44B2-B549-03E688B5F7A2@vizachero.com><f3f05ce80807161429vb8ff25dwad26faf73879c352@mail.gmail.com><4F99FAA7-AF81-4104-A7ED-6B0ECEA5B0D0@vizachero.com><f3f05ce80807161547v26445037rd75d06c8dee96a04@mail.gmail.com><00d801c8e799$3861df90$6400a8c0@Ken1><f3f05ce80807161626v3ce0cfbaq4939c50a0664e784@mail.gmail.com>


----- Original Message -----
From: "Dienekes Pontikos" <>
To: <>
Sent: Wednesday, July 16, 2008 5:26 PM
Subject: Re: [DNA] calibrating for common ancestor - QUESTION?


> That is precisely true, that is why it doesn't make sense to average
> over the 5 estimates, since for example the I1 and the J estimate have
> a long common component (CT to IJ). They aren't _independent_
> estimates of the quantity in question (CT to now).

Each of those interclade MRCA estimates is an estimate of the same thing,
however. And a way, indeed the optimum way, to combine those multiple
estimates of the same thing needs to be formulated and has been formulated,
although a simple averaging is not the optimum way to combine all that
information in order to get the best confidence interval.

When the SNP clocks come along, they will face the identical issue: they
will have multiple estimates, each subject to statistical noise, for the
same ages. In both the cases here we have the advantage of basically
knowing the tree node structure. So the optimum estimation procedure should
reformulate the problem as estimating the branch lengths between each pair
of nodes subject to constraints such as the present is the present by the
various routes. The times of occurrences for all the nodes are the
independent entities. Then with all those node times estimated one
subtracts the times between appropriate nodes to get the ages of interest.


This thread: