[Taxacom] methodological plurality

Kipling (Kip) Will kipwill at berkeley.edu
Thu Nov 22 11:47:03 CST 2012


I was doing other stuff and couldn't contribute to the thread for a 
couple days. I do appreciate everyone’s comments. I am going to reply to 
a few in several messages (ready your delete finger!), in hopes of 
poking the broomstick into the hornet’s nest a bit. I will attempt to 
polarize things a bit more, a behavior rare among systematist [I need a 
sarcastic tone app for email].

>> David Campbell:
>
> I think I see where I misunderstood you.  What is a "method"?  Testing
> across a varied range of parameter space could be classified as a
> "method" of analysis, specifically analyzing for sensitivity.  If
> "method" is thought of very broadly, then mutliple methods is a useful
> approach, in contrast to the "here's the tree produced by my favorite
> phylogenetic black box method, therefore all other classifications are
> wrong" approach.  Multiple methods for assessing phylogenetic trees
> are hopefully fairly redundant, assuming that the method in used is
> appropriate for the data set.  However, arbitrary selection of a
> single method would be even less informative.  What is needed is a
> consideration of what methods of analysis, testing for support, etc.
> are likely to be most appropriate for a given data set, why different
> data sets might give different results, why different methods might
> give different results, etc.
>
> So "I ran parsimony and Bayesian analyses under a single set of
> parameters apiece and got similar/different trees" is not all that
> informative, whereas "I ran phylogenetic analyses using a wide range
> of parameter space and consistently obtain high support for clades A,
> B, and C" is more informative.  Yet both could be described as using
> multiple methods-the difference is "methods of what?"

It is a very good point that we need to be clear about what we mean by 
methods.  I would take the stance that a method is a means to end, i.e. 
I want to discover the sister-group relationships between entities that 
I believe exist(ed) (my ontology) and so I need to use a method that is 
consistent with or able to do this, and is expected under most 
circumstances to actually find what I am after (my epistemology). To 
fully justify (and by justify I do not mean call it my “favorite black 
box”, which would just be bad science) a methods I need to explicitly 
accept assumptions about the method that are directly counter to 
assumption for other methods, e.g. Pars no common process contra a need 
for an explicit model in ML.  There is a long literature contrasting 
methods that includes many aspects that put them at odds, but you get 
the point. If this sort of dissonance does not bother you, then deciding 
which methods of the plethora available you to include becomes an 
arbitrary choice (some will see glimmers of an old argument against 
phenetics in this). Most people show a distaste for UPGM, some can even 
say why, but few are as able to express why Bayes is good (I do love 
those huge support numbers).

If we were to just “arbitrarily” chose our “favorite” method (which I 
certainly hope no one is doing) it would indeed be bad science. But if 
“favorite” is actually proper justification of why that method is likely 
to answer the questions posed, with no, little, or at least detectable 
bias and artifact and we state this explicitly, it’s still an arbitrary 
choice, but it’s not random. All choices are ultimately arbitrary at 
some level. Surely choosing to use all the current methods that are 
popular is both arbitrary and, I continue to maintain, less informative 
unless one purposely sets out to compare how they differ and what that 
means.

It looks like in the end of your statement you slip back into the usual 
mode of equating methods by treating them as variants of parameter space 
and using them for a measure of support. I have already commented on why 
this seems problematic.


-- 

Contact info:

Kipling W. Will
Associate Professor/Insect Systematist
Associate Director,Essig Museum of Entomology

send specimens to:
Essig Museum of Entomology
1101 Valley Life Sciences Building, #4780
University of California, Berkeley
Berkeley, CA 94720-4780

letter mail to:
130 Mulford Hall
ESPM Dept.- Organisms & Environment Div.
University of California
Berkeley, California 94720

fax 510-643-5438




More information about the Taxacom mailing list