[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Subject Index][Author Index]

Re: Wukongopterus and Darwinopterus



On Nov 29, 2009, at 5:12 PM, David Peters wrote:

Correlated? Perhaps. But now let's nest stegosaurs in Dinosauria. Are they still correlated? Doesn't this get to be a matter of opinion at one point or another? Isn't opinion something to be avoided?

Correlation coefficients are something that one can mathematically calculate; they are not a matter of opinion. Their relative importance may be a more subjective matter, but testing for correlation is quantitative.



We might have nailed down the problem you're having with your trees. The obvious in your estimation becomes "inapplicable".

I was not aware that he had any problem with his trees [he=David M]. Incidentally, I don't see anything inaccurate about the eye placement scoring example the David M. provided - it would appear to be an inapplicable scoring to me, too.


We're only trying to build a model here. So we don't wait until all the animals that ever lived are available for their analysis, we jump right in with a dozen to a dozen hundred taxa and build our model. We're not trying to ignore a problem. As in calculus, we're trying to simplify it so it can be understood.

Yes, but some simplifications are reasonable, while others produce errors. Correlated characters (the issue that started this particular subargument) are important to tree topology. This is a mathematically demonstrable feature. Therefore, it is not something that can be ignored.


That doesn't even enter the question. Ordering is something that's determined before the analysis. The states of a potentially continuous or meristic character must be ordered (Wiens 2001, Syst. Biol.) because you already used that assumption for partitioning the character into states in the first place -- the assumption that it's easier to go from a state to a similar state than to a morphologically distant one.

Let me ask you, once a fenestra appears, can it then disappear? If you ordered that character, you'd be making a mistake with all the best intentions.

Yes, it can disappear (see: turtles). I don't see why that character should be unordered. Furthermore, David M.'s argument was not that every character needs to be explicitly ordered, but rather that ordering choices need to be considered for each character, and that a potentially continuous character are still implicitly ordered by partitioning. This does not imply that explicit ordering should always be used, nor does it imply that such ordering should be avoidable. The importance and impact of character ordering is something with a relatively rich literature now - the impact of ordering and the mathematical underpinnings are something that can be studied quantitatively. There is no point in arguing about it philosophically.


If you have 3 characters in your matrix that are correlated to each other, that's the same as having 1 character in it that shouts 3 times as loud as the others. Because you gloss over this problem, the say-so you get is distorted.

Not in my experience. My trees have included legless taxa, crestless taxa, toothless taxa and in all cases the overall suite of characters separates convergent appearances of all these traits.

How do you know that your topology is accurate? You seem to be saying that despite having correlated characters, you get a topology that you think is reasonable, therefore they must not matter - but you can't eyeball a tree as a test. Correlated characters do produce disproportionate signal from some cells: that is a mathematical fact. Whether or not this produces a wrong answer will depend on the particular dataset, but given that it *could* give a wrong answer, and given that it is fiendishly hard to determine if a tree is "good" in a post-hoc manner, the best approach is to remove correlations where possible. To be honest, I'm not sure why you're fighting this: the quantitative background for this problem in the literature is very robust, and I can't see why it should hurt to take correlations into account. Your primary argument against it so far seems to be that you consider it too difficult (something about complexity and madness if I recall, which made me chuckle). While I agree that it is a pain in the butt, that does not mean correlation calculations should not be done.


So I don't buy your dutiful consideration. Simple "is-is not" dichotomy works very well. Avoids biases.

Several robust studies indicate that this dichotomy does not work very well in many circumstances. I can provide specific citations if need be, but a quick browse of Systematic Biology or Cladistics will provide several examples.




By definition. Apomorphies are rare. By definition evolution is a slow process with minor changes between generations. I'm not making this up.

Actually, evolution is a slow, incremental process only by observation, not by definition. In any case, that has nothing to do with apomorphies being rare, especially at the scale of large phylogenies. Furthermore, there is no clear threshold for what "rare" would be, anyway. So, regardless, we cannot count apomorphies and use it as a test of tree robustness. Apomorphy count is taken into account *within* a given tree search if parsimony is used (to determine an semi-optimization within a particular tree space search): but that's for different reasons (and means different things) than some kind of assumption that apomorphies have to be rare because evolution isn't fast.


This seems logical, but in fact it's backwards. The denser the taxon sampling, the shorter every internode will be -- and the fewer changes to the matrix will be required to overturn any of them.

David, your scale is out of whack. We're talking about large clades, big branches. You're talking about uncles and cousins. The tens of million-years lineage that led to the redheads, blondes and brunettes will be unaffected.

I think you missed his point. Numerically, internode distance shortens as taxon sampling increases - that's just a fact of tree shape.

Cheers,

--Mike H.