Trees and Star Diagrams

That’s false. Its demonstrably false. Why would you keep repeating nonsense like this? What would it take to convince you? What if we made some random sequences and put it into a phylogeny program and showed that it didn’t produce a structured tree?


If they were truly random, then the odds would be astronomically low of generating a nested hierarchy.

Almost as low as the odds of your understanding why that is the case.


You could, but it would be arbitrary. You can make anything a nested hierarchy by arranging the pieces that way. What you can’t do is find hierarchical structure in the data and use that to build the hierarchy of pieces. Witness your repeated failures at doing so.

No, it’s true. You can put random numbers into a tree you choose yourself and that would be a nested hierarchy, just not a well-supported one. Stick random data into a phylogenetic algorithm and you will probably get a single “best” tree, though it won’t be significantly better than a huge number of other trees. Not sure how you define “a structured tree”, but if you refer to highly dichotomous resolution, that’s probably going to happen, especially under a likelihood model. But the support for that tree won’t be much better than the support for a host of other trees. Real data, of course, would be different.

Short answer: an arbitrary nested hierarchy and a hierarchy arising from hierarchical structure in the data are two quite different things. It’s the former that @scd is proposing and the latter that we find in biology.


Yes, that’s my point. You’d get a “tree” but it wouldn’t have deep structure with good support.


OK. In that case you expressed it in a problematic way, as did @Faizal_Ali.

A star diagram is technically a tree, but it isn’t a well supported tree with deep structure. When we discuss “nested clades”, we mean a “well supported tree with deep structure”, not a “star diagram.” Shorthand is to equate “well supported tree with deep structure” with “tree,” even though a star diagram is technically a restricted type of tree too.

The key point is that we can actually show that random sequences do not produce a well-supported tree with deep structure. Biological sequences also produce a well-supported tree with deep structure.

@Joe_Felsenstein is there a more precise or technical way to express “well-supported tree with deep structure”?

I’m afraid this is confused. A fully bifurcating tree is not necessarily well supported. As I have mentioned, most phylogenetic algorithms will give you a fully bifurcating tree from random data. Not sure what “deep structure” means to you.

Very confusing. No, random sequences do not produce a well-supported tree. Still not sure what “deep structure” means.

One would first have to know what you intend to convey by “deep structure”.

A tree with well supported bifurcations at different time points so we can definitively rule out a star diagram.

So you just mean to repeat “well-supported”? The term for that is “well-supported”. Your language is consistently blurring the difference between “well-resolved” and “well-supported”. Random data will probably give you a well-resolved tree but is very unlikely to give you a well-supported tree. You have claimed otherwise, but I suspect it might have been a typing error. When you said:

…did you actually mean to say that?

1 Like

Nope. :slight_smile:

You can imagine my confusion.

Well, it’s fixed now. And ideally you’ll delete this diversion.

I’ll just add that they major groupings of placental mammals (Afrotheria, Xenarthra, Laurasiotheria, and Euarchontoglires) are separated by rather short branches in the mammalian tree. It took a long time and a lot of data, but they are now pretty well-confirmed. So, as John has emphasized, well-supported need not mean well-resolved. Obviously being well-resolved helps achieve well-supportedness with less data.


Oh, dear. We seem to mean different things by “well-resolved”. I just use it to refer to a tree with few polytomies. A well-supported tree, on the other hand, offers strong support for that resolution. Short branches may be resolved but not well-supported, but if they’re well-supported they must certainly be resolved, at a minimum.


I sense there is some terminology that needs to be worked out here. This is important because it seems that there is a legitimate point being made. Put random sequences into a program, and you will in fact get a tree. Case in point, a star-diagram is a type of tree. So merely the fact that a tree was output can’t be evidence for evolution.

It must be something about the tree we observe from DNA, some qualities of this tree, that make it evidence of common descent. Some qualities the DNA tree has that random sequences do not. What precisely are those qualities?

Likewise, YEC model’s of origins are predicted to produce universal trees too. Within each kind, it should be indistinguishable from the evolutionary model (except the time estimates), but what of the tree between kinds? What precisely is it that common descent predicts about the total tree that is not predicted by the YEC orchard models?

Perhaps someone else has worked it out already. I think we all know what it is on an intuitive level. But what is it exactly?

1 Like

Yes, but that isn’t what I was talking about. Put random sequences into a program, and you will very probably get a fully bifurcating tree, not a star tree. But you are correct that this isn’t evidence for evolution.

Those qualities are the consistency of support for that tree rather than some other tree. We might assess that support by bootstrapping, for example, or by a likelihood ratio test, or by various other means.

Very simply, that the tree will have strong, consistent support from different partitions of the data. We would expect a tree of separate kinds to be resolved, given enough data, but we would not expect it to have strong support.

I will point out that there is so far no actual YEC orchard model. No YEC can actually tell so far where a kind ends and another begins. The most common substitute for such a model is just to say “family” without providing any justification; and of course Homo sapiens is an exception even to this. I would claim, on the other hand, that if there were separate kinds, the breaks should be obvious.

One other feature of an orchard model, if one were to posit one, would probably be that the branches connecting kinds on a tree of separate kinds ought to be very long compared to those within kinds, since they represent original, created separations. In other words, what we see should approximate a star tree at its base, with long stems leading to the ancestral created kinds, followed by diversification within each kind. The connections at the base of the star tree should not be well-supported and should collapse into a real star tree upon any test of support, while the long stems and resolution within kinds would remain.


I did indeed. So thanks to the pros here for putting it correctly.

By the way, PS alum @EricMH has been arguing on Biologos that phylogenetic signal provides no evidence for common descent (here and here).


He is welcome to discuss it here if he likes. If there is anything that would benefit from a more technical response, feel free to clarify the scope and question in a new thread.

I did suggest that there was more expertise in the subject here. I haven’t even had time to look at what he’s written – been busy with covid-related stuff. Two grants and a paper submitted within a few weeks.