Barry seems to have noticed TSZ again, and so I will take this opportunity of inviting him over here, where he can post freely, and will not be banned unless he posts porn or malware or outs someone, which I expect he can manage not to do.
And he responds to my post, Lawyers and Scientists. He does so in two parts, so I will devote two posts to them. Here is my response to his first part. Barry writes:
PART 1
First Liddle writes that I have
. . . confused the assumption of common descent with the conclusion of common descent, and thus detected circular reasoning where there is none.
Where did I do such a thing? Boiling that paragraph down I made the following claims:
- Common descent is not necessarily false.
- But Cladistics does not establish common descent one way or the other.
- Instead, cladograms are constructed ASSUMING common descent.
- It is circular reasoning to conclude that a technique establishes that which it assumes in the first place.
- Therefore, anyone who says that cladistics establishes the fact of common descent has used faulty reasoning and is mistaken.
- There are in fact people who make that mistake.
To establish beyond doubt point 6, Glen Davidson kindly jumps into Liddle’s own combox with this:
Barry: “This is not to say that common descent is necessarily false; only cladistics does not establish the matter one way or the other.”
Glen: “Of course it does. What a ridiculously ignorant dweeb.”
All six assertions seem to me to be on solid ground. Not only are they true, they are not even controversial. But for Liddle’s charge to be correct, at least one of the points I made must be false. OK Liddle, which of the six totally non-controversial points I have made do you disagree with? If the answer is “none,” then the only gracious thing to do is to withdraw your claim.
The short answer is that I disagree with 2-6, for the reasons I gave in my first post: the answer lies in null hypothesis testing. Far from “assuming a tree”, both linear correlations and tree distributions are tested by FITTING a slope/tree, and testing whether the best fit is a better fit than would be expected under the NULL of no linear relationship/no underlying tree structure. If, having fitted the slope/tree, the fit is no better than would be expected under the null of no linear relationship/no underlying nested hierarchy, then you RETAIN THE NULL. If it is better, i.e. if a fit as good as that observed is UNLIKELY under the null, you reject the null and consider your hypothesis (linear fit; common descent pattern) supported. Of course there could reasons other than common descent that could explain the tree – but the tree can be established as an OBSERVATION to be EXPLAINED. Which Linnaeus did before Darwin. And it was that clear tree that Darwin sought to explain by, firstly, Common Descent, and, secondly, by a mechanism that would explain adaptive change-over-time.
If Barry cannot understand that testing a NULL HYPOTHESIS is the OPPOSITE of assuming that your model is true, then perhaps he could shoot an email to the former owner of his site.
It is of course true that null hypothesis testing is counter-intuitive and doesn’t do what many of its practitioners think it does, but it’s still an excellent workhorse, and what’s more, is the beating heart of ID’s very own eleP(T|H)ant.
[My response to the second part will have to wait – I have some null hypotheses to test first….]
ETA: Looks like this response deals with Pt II at as well.
What exactly are you “fitting”?
Lizzie,
The problem is that he can’t moderate or ban people here. Perhaps if we give him a fake ban button to click on?
keiths,
Be gentle, he’s clearly having angst / not in a good space right now.
And this differs from other times how?
Glen Davidson
This is pertinent, so I’ll repost from the other thread:
What’s especially funny about that thread is that it shows without doubt that
BarryBanny never understood the “Bomb” argument (aka the objective nested hierarchy argument).No wonder he ran whenever I asked him to explain, in his own words, what was wrong with it.
Vincent,
Perhaps you could step in and explain Barry’s mistake to him?
with the new understanding that HGT is widespread throughout nature, even widespread amongst unrelated species, I’m not sure how any relationships inferred by phylogenetic trees or cladograms could possibly hold any water. Not only that but the fact that all mammals, for instance, share pretty much the same compliment of genes, and that they’re simply expressed differently is another knock against the notion that relationships can be inferred by looking only at genes.
I just posted this over at UD:
How about reading my actual blogpost, instead of the Berlinski quote? I actually explain how cladistics *does* test common ancestry, and supports common ancestry when the tests are positive (which is virtually always, with biological datasets). Read the parts about CI, null distributions, etc.
I *also* say that cladistics successfully tests common ancestry, *despite* the fact that cladistic methods cannot detect direct ancestors, only collateral ancestors (sister-group relationships). The reason cladistics cannot detect direct ancestors is primarily because the computer methods propose trees where every taxon is a tip. Thus every tip can only be sister to other tips, you can never have a tip directly below another tip.
As it happens, this limitation of cladistics has recently been overcome in modern Bayesian statistical phylogenetics. (Note carefully: cladistics is just one, fairly simple, form of phylogenetics. In fact, cladistics is considered old-fashioned at this point, even though it is useful for public debates with creationists since it is easier to explain than statistical likelihood calculations and Bayesian MCMC searches.) I predicted that direct-ancestor detection was coming in that blogpost in 2013, and lo and behold, it’s been published in 2014 and 2015. These new methods propose and test trees where some of the “tips” are moved into direct ancestor positions. You have to have a model of fossil sampling through time and some other computational machinery to make this work, which is why it’s taken so long.
These new methods are cool, but they don’t change anything fundamentally. The terms “common ancestry” and “direct ancestry” mean different things. You can infer the first without having the second. Just like a DNA test can tell you if someone is your cousin, even if neither of you know who your grandparents were. Getting this confused is one of Barry’s mistakes, amongst numerous other failings (high arrogance, low knowledge, high propensity to blab without doing any research in the primary literature, reliance on quotes from popularizations rather than absorbing the consensus in the primary literature.)
On testing common ancestry: the Berkeley page is only correct for a one-dataset, one-cladogram analysis. In such a situation, what you are doing is saying common ancestry is so well-supported generally in biology, that it is safe to assume it for for this particular analysis, rather than wasting time re-doing already-proven science. This is perfectly reasonable. But it’s not the full story. Introductory science sources often don’t have the full story, and plus, intro-science websites are often written by beginning grad students who haven’t learned everything yet.
If for whatever reason, one wants to waste time arguing with ignorant creationists, cladistics *can* be used to test common ancestry. The simplest test is…run *two* different datasets! For example, two different genes, or a gene and a collection of many morphological characters, or whatever. Common ancestry predicts that the trees will be much more similar than expected by chance, because the characters being studied will have shared history for the part of a tree in which they were inside the same ancestral lineages (Note: *not* freaking identical.There are many different, known, observable, and unavoidable and thus expected stochastic processes in mutation, population genetics, etc., that can produce some disagreements).
Of course, it’s even a simplification to say that two datasets is, by itself, a test of common ancestry. A comparison of two datasets is really just one datapoint. The *real* test is comparing dozens, hundreds or thousands of datasets. The conclusion is: the statistical signal of vertical inheritance is amazingly, fantastically good for eukaryotes, especially multicellular eukaryotes with protected germlines. (Even the claimed exceptions in prokaryotes etc. are mostly overblown — as are hybridization events in eukaryotes, which are typically between what are close relatives anyway.)
A more complex test (non-cladistic) involves constructing a null distribution from the data and then seeing how many sigma above the null your actual observed data. Often data are 10+ sigmas above the null distribution of no tree pattern.
Yet more complex methods involve fitting probabilistic models and using standard statistics like Likelihood Ratio Tests and Bayes Factors to compare the hypotheses of no common ancestry and common ancestry.
The authors of this work are people like David Penny, and Doug Theobald (Nature, 2010). The work is *extremely* well-known to anyone serious who has followed the creationism issue. Much of the earlier work is laid out, with explanations and citations, in Theobald’s “29+ Evidences for Common Ancestry” FAQ at talk.origins, which has been online for 10+ years now.
Anyone serious would know about all of this, and review and rebut it in detail, if they wanted to say anything serious about how common ancestry is not supported. But, Berlinski doesn’t do this, and neither does Barry Arrington. Why? Are ignorant, lazy, and arrogant words that are too strong for that kind of behavior?
I see Banny Arrogant has responded to Nick’s post with his usual blustering, insulting, content-free style. Being the blustering bully seems to be all the shyster knows.
Oh, the persecutions Barry has suffered:
Somehow a familiar whine, most IDiots who come here feel piteously maligned, and all because they write utter bilge about evolution and extremely prejudicial and evidence-free assertions about all who disagree with them. But, oh, how can one fault such sensitive souls?
Glen Davidson
The “non-cladistic” was supposed to go here in my post:
Yet more complex methods involve fitting probabilistic models
–>
Yet more complex (non-cladistic) methods involve fitting probabilistic models
Using KF’s bilge, it is very telling that evolution proponents are frequently willing to post at UD, even though they know they are likely to be banned, but very few IDists are willing to engage here, where they won’t be banned.
I have respect for Mung and Sal, even though i disagree with them. They are willing to participate here even though they know they are the minority.
Sorry Frankie, no matter what you do, you are still Joe. None of us can change that.
The amount of HGT in multi-cellular species is minuscule compared to the genetic signature of common descent. There’s nowhere near enough HGT “noise” to make an impact on the common descent powerful signal.
Wow:
Barry:
What a shameless hypocrite Barry is. He knows full well that KN’s screwup was promptly corrected and that the deleted comments were restored.
Barry’s fear is not that his comments will be deleted. It’s exactly the opposite. He’s afraid that his comments won’t be deleted and that everyone, for months and years to come, will see his humiliation at the hands of his opponents.
P.S. Barry, I’m still awaiting your excuses for deleting an entire thread that made you look like an idiot.
Barry gets caught red-handed
nickmatzke,
Welcome, Nick!
keiths,
I posted this on his thread:
Mike1962: ““Too many emotional retards over there.”
Me: “I agree. People at UD are much more mature than at TSZ. I have seen very few people here call opposing commenters liars, stupid, willfully dishonest, pathetic snivelling cowards, Strawman Humpers, dirt worshippers, and other immature epithets that the people at TSZ use all the time in describing us.”
And then I was banned, I was shunned, nobody loved me anymore….(read that in the voice of Bill Murray channelling Dr. Venckman, it will bring a tear to your eyes).
What is strange is that I was banned for simply repeating a few of the insults that Barry himself has used, other than ‘dirt worshipper’, that was Mapou’s. it is almost as if he is ashamed of what he has said in the past. As he should be.
Elizabeth, I realize that this borders on guano. Feel free to move it if you deem it worthy.
keiths,
Where is the TSZ thread? We might as well link to it on UD so people can see it.
Yes, but look at how many threads that make him look like an idiot that he has never deleted. You don’t think he’s just that clueless, do you?
Oh, I see…
Glen Davidson
Here’s some background: http://www.uncommondescent.com/intelligent-design/arrington-effectively-banned-at-the-skeptical-zone/
Nick,
Here are Barry’s comments, along with links:
November 27, 2013 at 1:15 am:
November 27, 2013 at 2:13 am:
November 27, 2013 at 2:15 am:
November 27, 2013 at 2:15 am:
In the case of a linear fit, a line. The fitting process, typically, finds the line that minimises the summed squares of the differences between the actual points and the line.
Nick may explain the procedure for tree fitting. It’s more complicated, but the principle is the same.
You may be fitting something to a line, or a tree, or some other distributive shape, but that doesn’t tell me what you are fitting to that distributive shape. What are you fitting?
William J. Murray,
You fit the shape to the data.
in other words you postulate a CLASS of shape (line, tree) and you fit that shape to the data.
If the best specific shape (line, tree) you can fit still fits poorly, you retain the null.
What is the data, then?
William,
Do you think Barry has made an error or are you supportive of his position?
Nick seems to do a good job of explaining that. Did you read his comment?
The data is a DNA alignment, typically. Or sometimes amino acids or morphology. The model is a substitution model — a rate matrix describing the rate of change from A to C, A to G, etc. The tree, and its branch lengths, are another part of the model.
Barry is very busy:
Plus he has to wash his hair.
nickmatzke said:
Does “DNA Alignment” mean DNA sequences, as in finding the same DNA sequences in different what, species? Families? What? Does it mean finding the same DNA sequences in the same position? Are the sequences supposed to correspond to similar or identical morphological characters, like ears or fingers or claws? As far as the reference to amino acids, what does that mean? Finding identical proteins in disparate species?
What I’m trying to understand here is what items of data are, how that “data” is compiled or interpreted to form any “shape” at all.
William J. Murray,
DNA alignment does indeed mean DNA sequences, but it’s not about finding the same sequences in different species. Each site in the sequence is compared individually among species. You can use a variety of models to do so. The conceptually simplest model, parsimony, tries to minimize the number of inferred changes (say, A to G at some particular site in the sequence) over the tree. That is, you pick the tree that requires the fewest base substitutions, summed over the entire sequence and the entire tree. Other models are more complicated, but most of them also look at sites one at a time, though they also estimate various parameters of the model using all the data.
But all of the models can be used to test many trees and to test whether one tree (or small number thereof) is a significantly better fit to the data — i.e. explains the data better — than others. I think the conceptually simplest method for this is the non-parametric bootstrap, in which you sample sites randomly, with replacement, to build a data set of the same size as the original. Do this many times, analyze each in the same way as for the original, and see how closely the resulting trees resemble each other. If there’s no consistency to the data, you will get no resemblance. If the data all agree on what the tree ought to be, you get the same tree every time. And the latter is good evidence that the there is indeed a true phylogeny. Otherwise, why the agreement?
One can do the same thing with amino acids. Again, the sequences don’t have to be identical; in fact they can’t be, as it’s the differences that provide information. Sequences don’t correspond to fingers and toes; the relationship between genome and morphology is much more complicated than that. For many purposes, it’s better to use junk DNA for sequence analyses, and junk corresponds to nothing at all in morphology.
Here is a very simplified discussion of one tiny bit of DNA sequence and one tiny bit of a tree that I wrote several years ago. Try it and see what you think.
In the case of a linear correlation, any pair of continuous variables. IQ versus academic success; height vs shoe size; cigarettes smoked per day vs chest expansion.
In the case of a phylogenetic analysis, morphological characters. Or DNA sequences.
Well, let’s take a linear correlation as it is relatively simple. One way to think about it is to visualise a scatter plot where you take each subject in whom you have measures of two variables, and plot each of them on a diagram in which the horizontal axis represents the value of one variable, and the vertical axis represents the other.
Then you simply draw a line through the dots, and measure the vertical distance from each dot to the line and sum them (add them up). Usually we square them first because dots below the lines will have a “negative” distance to the line and dots above the lines will have a “positive” distance, but you can just think of them as distances and ignore the sign.
Of the infinite number of lines you can draw there will be ONE line that produces a smaller sum of squared distances than any other. This is called the “Ordinary Least Squares” best fit.
It will almost always have a slope, but if the slope is very frequently found under the null hypothesis of “no linear relationship between the variables” we “retain the null” of “no slope”. If it is so steep as to be very rarely found under the null hypothesis we “reject the null” and claim support for the hypothesis that there is a real linear relationship between the variables.
For trees it obviously more complicated, but the approach is identical. The computer will test trees until it finds the best-fitting tree. If that tree is highly likely under the null hypothesis of “no tree relationhship in the data”, then you retain the null. If it is highly UNLIKELY under that null, your hypothesis of an underlying tree structure is supported.
Well, let’s take a linear correlation as it is relatively simple. One way to think about it is to visualise a scatter plot where you take each subject in whom you have measures of two variables, and plot each of them on a diagram in which the horizontal axis represents the value of one variable, and the vertical axis represents the other.
Then you simply draw a line through the dots, and measure the vertical distance from each dot to the line and sum them (add them up). Usually we square them first because dots below the lines will have a “negative” distance to the line and dots above the lines will have a “positive” distance, but you can just think of them as distances and ignore the sign.
Of the infinite number of lines you can draw there will be ONE line that produces a smaller sum of squared distances than any other. This is called the “Ordinary Least Squares” best fit.
It will almost always have a slope, but if the slope is very frequently found under the null hypothesis of “no linear relationship between the variables” we “retain the null” of “no slope”. If it is so steep as to be very rarely found under the null hypothesis we “reject the null” and claim support for the hypothesis that there is a real linear relationship between the variables.
For trees it obviously more complicated, but the approach is identical. The computer will test trees until it finds the best-fitting tree. If that tree is highly likely under the null hypothesis of “no tree relationship in the data”, then you retain the null. If it is highly UNLIKELY under that null, your hypothesis of an underlying tree structure is supported.
If you want more info on tree fitting, ask Joe Felsenstein, as I think he developed the method for statistical testing of fits.
Heh, or John Harshman.
Golly, I should read the thread in full before responding!
Elizabeth,
It’s amazing (to me) that they want to advance statistical design arguments without understanding the basics. For the majority of them, I think ID is an emotional / intuitive thing: They talk about concepts, create acronyms (that are placeholders for design) but do no math.
Cladistics does not test nor confirm Common Ancestry. It assumes it and tries to find relationships based on that assumption. Also evolution is too complex to produce orderly trees.
This is false, frankie. Read my post to see why.
Well, they write equations sometimes. My beef is that they (not even Dembski) understands what the terms in the equations mean.
Your post has nothing but more of your bald declarations. And evolution is too complex to produce orderly trees/ Each population can give rise to numerous branches and defining traits can be lost as well as gained.
BTW the eleP(T|H)ant in the room is your inability to support your position. Yours is the chance hypothesis position and yet you have nothing. There isn’t any way to test the claim that natural selection and drift produced the diversity of life. Cladistics is absent a mechanism.
The lost of a trait is defining, e.g. snakes lost their legs. It’s not unusual that the original state is more complex than the subsequent state.
And exactly why evolution doesn’t produce orderly trees. Thank you for agreeing with me. Elizabeth won’t be pleased though.
Elizabeth,
Joe is the first person to apply non-parametric bootstrapping to phylogenetics, among other things.
Except that it does.
Excuse me if I’m repeating something that’s come up in another thread. What Errington is saying lately about his superior lawyerly skills of discernment (“I’m better able to see what’s wrong with the form of the argument than are scientists who’ve devoted their lives to the substance”) is a shameless rehash of what Phillip Johnson said about his own, a quarter century ago.
Which do you have in mind?
To be fair, Barry did credit (blame?) Phillip for that line of “thinking.”
Funny how both lawyers are counter-examples to that claim. Clearly, other attorneys do rather better with this matter, however.
Glen Davidson