[Taxacom] Taxacom Digest, Vol 68, Issue 8

Michael Heads michael.heads at yahoo.com
Wed Nov 9 18:55:54 CST 2011

Hi Sergio,That's a good summary of the problem. No-one is
actually testing anything and it's an interesting question why reviewers and
editors let this stuff through. Here's a typical example: based on a Miocene
fossil seed of Wahlenbergia from sediments of Upper Karpathian age
(16.5-17.5 Ma), Prebble et al. (2011. Mol Phylogen Evol 59: 636) assigned
priors for its clade age (a normal distribution, a mean of 17 Ma, and a
standard deviation of 0.4 Myr) that gave a prior 95% HPD interval of
16.34–17.66 Ma for the clade. Thus the clade age is made to correspond to the
age range of the Upper Karpathian stage. Why would you do this? There is no
logical argument here, just assertion. Anyway, based on these priors and two
other similar sets, Prebble et al. were able to confirm a Cenozoic age for this
intercontinental genus (ruling out vicariance). The modern synthesis relies on
the 'default setting' that clade age = fossil age + ~5 m.y. (e.g. see Ken's
last post) and this is being slipped into the Bayesian work in the assignment
of priors.
Wellington, New Zealand.

My papers on biogeography are at: http://tiny.cc/RiUE0
Information on my new book, 'Molecular panbiogeography of the tropics', is at: http://www.ucpress.edu/book.php?isbn=9780520271968

From: Sergio Vargas <sevragorgia at gmail.com>
To: Michael Heads <michael.heads at yahoo.com>; taxacom at mailman.nhm.ku.edu
Sent: Thursday, 10 November 2011 11:49 AM
Subject: Re: [Taxacom] Taxacom Digest, Vol 68, Issue 8

Yes, I agree, but that's how you end up obtaining maximum ages from minimum ages. That's the price you pay. The problem in most cases is not been able to clearly state that your conclusions are valid only GIVEN all the modelling you place in between the minimum age and the estimated region were the maximum age could be. That means of course that the next time a new (older) fossil is found your ages will change and so on until no older fossil is found which doesn't mean you found the older. So in any case you will need to resort to some kind of modelling if you really want to date a node using molecular clock like analysis. You can use uniform priors on your nodes. Most biologists feel better using flat priors because most of us recognized that there is no good way to choose a prior for the age of a node but then you tend to get wide error bars. In my experience, biologists and palaeontologists working to molecular dating get very nervous when they see
 huge error bars after the analysis. The uncertainty is so high that you cannot really say anything. The usual answer is to play around with the priors and the root age until they feel comfortable with the results, i.e. the error bars are not too large. Also at some point the user becomes very pragmatic and recognizes that the priors are so because... well because... and since no one knows which one to pick is justifiable to pick just one that "works good". My problem is what "good" generally means in this context: I like what I see because it is what I though. So I am not testing anything, I am biasing the analysis using my favorite priors and I am finding that what I thought (my priors) were very good! This might be self-fulfilling for some, but reviewers should filter such inconsistencies very quickly, ask for more robust analyses: cross-validation, effect of different priors and root ages, and prevent that authors sale their date as The Dates. This
 is simply reminding authors that the dates are only approximations given the model.



On 11/9/11 10:58 PM, Michael Heads wrote: 
Hi Sergio, 
>You wrote: 'Most authors agree that fossils provide only minimal ages, I don't think they ignore this but they base their conclusions on the credible region obtained from the analysis, and treat this credible region as an approximation to the maximal age of the clade'.  
>   In other words, they transform a minimal age into a maximum age (the maximum age of the 95% credibility interval). But how can you actually do this? They do it simply by decree: they stipulate an exponential or log normal prior for the clade age, with rapidly decreasing probabilities for clade ages older than the oldest fossil. An exponential prior even attributes the most likely clade age to the age of the oldest fossil! It's all smoke and mirrors.   
>Wellington, New Zealand.
>My papers on biogeography are at: http://tiny.cc/RiUE0
>Information on my new book, 'Molecular panbiogeography of the
          tropics', is at: http://www.ucpress.edu/book.php?isbn=9780520271968
>From: Sergio Vargas <sevragorgia at gmail.com>
>To: taxacom at mailman.nhm.ku.edu; jgrehan at sciencebuff.org
>Sent: Thursday, 10 November 2011 10:08 AM
>Subject: Re: [Taxacom] Taxacom Digest, Vol 68, Issue 8
>I have a couple of questions:
>>Don't just take the word of 'experts' (experts in what
              anyway). You should be able to determine for yourself what
              the empirical reality is and what is
>by theoretically claimed you mean hypothesized?
>>With molecular clocks you will often see the dates
              represented as actual or maximal, but the empirical proof
              of that is lacking (that you can read for yourself, no
              need to>take anyone's word for it either way). You will
              also see so-called mathematical modeling or statistic
              estimations claiming some sort of average and probability
              spread>which is ok as it stands as a theoretical claim
              that may or may not have an empirical foundation. You will
              also see authors contradict themselves within a paper,
              saying on>the one hand that the dates are minimal and
              then later ignoring that understanding. Anyway, glad you
              are aware of the pitfalls.
>well... if you had empirical evidence for the
              maximal/actual age of a clade you would not need molecular
              dating. I would like to know what the empirical evidence
              for the maximal age of a clade could be? There is no way
              to demonstrate that a fossil is of maximal age, you can
              only approximate the maximal age of a clade hypothesizing
              that there is no fossil of taxon A that is older that the
              oldest known fossil of taxon A, or using molecular dating
              to provide a credible interval for the maximal age of the
              clade given the calibrations (minimal ages) and prior
              distributions for the node ages (to model the maximal
              age). Most authors agree that fossils provide only minimal
              ages, I don't think they ignore this but they base their
              conclusions on the credible region obtained from the
              analysis, and treat this credible region as an
              approximation to the maximal age of the clade. Of course
              some authors really want to have The Date of cladogenesis,
              but with bayesian methods this is getting less common.
>Sergio Vargas R., M.Sc.
>Dept. of Earth&  Environmental Sciences
>Palaeontology&  Geobiology
>Ludwig-Maximilians-Universität München
>Richard-Wagner-Str. 10
>80333 München
>tel. +49 89 2180 17929
>s.vargas at lrz.uni-muenchen.de
>sevra at marinemolecularevolution.org
>check my webpage:
>check my research ID:
>Taxacom Mailing List
>Taxacom at mailman.nhm.ku.edu
>The Taxacom archive going back to 1992 may be searched
              with either of these methods:
>(1) by visiting http://taxacom.markmail.org
>(2) a Google search specified as: 
              site:mailman.nhm.ku.edu/pipermail/taxacom  your search
              terms here

Sergio Vargas R., M.Sc.
Dept. of Earth & Environmental Sciences
Palaeontology & Geobiology
Ludwig-Maximilians-Universität München
Richard-Wagner-Str. 10
80333 München
tel. +49 89 2180 17929 s.vargas at lrz.uni-muenchen.de sevra at marinemolecularevolution.org check my webpage: http://www.marinemolecularevolution.org check my research ID: http://www.researcherid.com/rid/A-5678-2011  

More information about the Taxacom mailing list