Nested Hierarchies (Tree of life)

Moderator’s remark: this post is long enough to need a “more” tag.  But the wordpress editor will only allow me to add that at the very beginning or the very end.  So here it is at the very beginning.

Do you want to be my cousin?
Sure. If not me, then who?

  1. “Nested hierarchies” or “cladistic analysis” or “consilience of independent phylogenies” is often offered as support for Darwinist evolution. This is the idea that the “tree of life” classification of organisms is somehow objective despite being a creation of very zealous “evolution” advocates. The three basic assumptions of cladistics models are: a) Any group of organisms are related by descent from a common ancestor (UCD – universal common descent); b) There is a bifurcating pattern of cladogenesis; c) Change in characteristics occurs in lineages over time. Although not explicit, UCD (“descent from a common ancestor”) here means by a Darwinian “natural selection mechanism” and not by a process generated by a designer that also happens to make use of biologic reproduction.
  2. No assumption can be tested by the model that uses them. That is why they’re called ‘assumptions’ and not ‘conclusions’. Instead, assumptions have to be tested independently through an entirely separated method or be accepted as axioms. An UCD “mechanism” has never been observed or proved elsewhere and is not “self-evidently true”, therefore not a valid axiom. Because UCD is an assumption in “cladistic analysis”, it cannot be logically also a conclusion of any such analysis. Furthermore the conclusions of any “cladistic analysis” will always and trivially be compatible with the UCD assumption in that model.
  3. Hypothesis testing requires an alternative (null) hypothesis and a procedure that demonstrates how the data available is compatible with the successful hypothesis and at the same time is statistically incompatible with the alternative hypothesis. In the “cladistic analysis” case, the alternative hypothesis to UCD is “common design”, and of course UCD cannot be an assumption of such an analysis. However this rule is violated twice, first by the use of an assumption also presented as conclusion, and second by the prejudiced rejection of the alternative “common design” hypothesis before analysis. This clearly demonstrates that “cladistic analysis” can never be logically used as proof of UCD. What “cladistic analysis” is instead is ‘curve fitting’ where the cladistics model is best fitted to certain (conveniently selected!) morphologic/biochemical/genetic biologic data points.
  4. The ‘designer’ hypothesis cannot fail against the ‘no designer’ (Darwinist evolution) alternative in a biologic comparative analysis as designers have maximum flexibility. This is not surprising as designers are free to incorporate whatever mechanism they want, including intelligent “selection” (human breeders do!) and “common descent” (human breeders do!) if they so desire.
  5. The claim that cars and other entities cannot be uniquely and objectively classified (“nested hierarchy”), while organisms can, is false. On one hand, we do know the history of the automobile, so a proper classification must be able to reconstruct their unique “evolution”. Yes, vehicle share parts, so to get to the actual development tree, we must group them differently than organisms since mass production works differently than biologic reproduction. On the other hand, organisms may not be uniquely classified as demonstrated by the numerous revisions and exceptions to the “tree of life”, and in any case, “uniquely classified” is an absolute claim that can never be proven since it is impossible to compare the infinity of possible organism classifications.
  6. The claim that the “tree of life” based on anatomy is validated by the match with the tree based on biochemistry fails. Anatomy is not independent of biochemistry. Also, the oldest DNA ever found was 700k years old therefore any match between the independent trees is limited. This is not to say that the fossil record is complete, or that fossils can be positively linked to one another and the living without – once again – presupposing UCD. The claim that “there is no known biological reason, besides common descent, to suppose that similar morphologies must have similar biochemistry” is false as the ‘designer’ hypothesis produces the same result when one designer creates all morphologies, and furthermore “I cannot think of an alternative reason why…” is not a valid argument.
  7. A “tree of life” is an artificial human construct as organisms do not come labeled with their position in a cladistics hierarchical structure. To decide the position of a certain organism, the human creators of the “tree” have to decide which morphologic/biochemical/genetic characteristics to include and what weight to attach to each of those measures. This further supports the claim that “cladistic analysis” is ‘curve fitting’ rather than ‘hypothesis testing’ – if a tree must be built, a tree will be built as in this example: “The close relationship between animals and fungi was suggested by Thomas Cavalier-Smith in 1987, […] and was supported by later genetic studies. Early phylogenies placed fungi near the plants and other groups that have mitochondria with flat cristae, but this character varies. More recently, it has been said that holozoa (animals) and holomycota (fungi) are much more closely related to each other than either is to plants […].”

 

1,059 thoughts on “Nested Hierarchies (Tree of life)

  1. Nonlin.org:
    No, you cannot.

    Yes, yes we can. I do that in my very own job. We ask, what if we assume such to be the case? Then this should happen. Let’s try it. Then we discover that either the assumption was wrong, or it was right. Nonlin, our words are our tools, not our enslavers. Stop “dictionarying” and start actually thinking. For someone who presumes of nonlinear thinking you’re not doing much thinking, let alone nonlinear. So, either think or change your monicker to Straightjacketed nonthinking.

    Nonlin.org:
    Since this is not going anywhere, go ahead and ask someone else.

    I don’t need to ask somebody else. I’m explaining this to you because you seem very ignorant about it, but for many other people this is everyday stuff to the point that I feel ashamed of having to explain something so basic. Things can start as assumptions and be shown either wrong or right, thus being updated to something other than an assumption. So fucking basic. I cannot believe that you cannot get it. Why? What’s wrong with you?

    Nonlin.org:
    This is 100% retard: “what was initially an assumption becomes a conclusion”.

    Sure. 100% retard on your part Straightjacketed nonthinking. I told you, your words backfire. I hope some creationist has the guts to explain this to you, because you’re really ridiculing yourself to extraordinary levels.

  2. Nonlin.org: As explained many times, what you do is curve fitting, not hypothesis testing.

    You mean to say, as ignorantly claimed by Nonlin too many times. Also, as explained too many times, curve fitting can be a way to test hypotheses. If only you learned reading for comprehension, and then some logic …

  3. Joe Felsenstein: 2. Yes, I have heard of the “independent and identically distributed” (i.i.d.) condition, but it is not strictly required for inferring phylogenies. nonlin.org, being apparently well-acquainted with all these methods, will know all about “mixing conditions” as they affect the Law of Large Numbers, and also such methods as the “block bootstrap” (Künsch, 1989).

    So you never got back on these:
    “Nonlin:
    1. Whatever you do, your model is fine as long as you’re not trying to use it to prove its underlying assumptions (circular logic).
    2. You should know that artifices can only do so much. Can you prove they can turn non-iid data into iid pseudo-samples?
    3. See 1. No doubt, once UCD is the main assumption, your curve fitting model can be optimized to that assumption. But the model will never validate its own assumption. This is elementary and quite frankly I am very surprised you have a hard time accepting it.
    4. Here we go again: “changes in different genes are very strongly nonindependent”… You’re just asking me to accept UCD without proof and to demonstrate “non-independence” when in fact the burden to demonstrate IID is on you. And regardless, see 1. and 3. Again.”

    Does it mean you admit your failures and now repent? Or more likely that you have no good answers. I should keep track of you guys going silent in lieu of admitting failure.

    Now, do you understand why your re-fried beans re-sampling doesn’t take care of the IID problem? Here is how it works: take a distributions and duplicate x times. Now you have x+1 distributions that are 100% correlated. Now re-fry re-sample however you want from this set and your samples will remain 100% correlated, so no IID.

  4. Entropy: I don’t need to ask somebody else.

    I instructed you to go fetch a counterexample:

    “Even better, look up a situation outside of the Darwinism pseudoscience where someone builds a model based on an assumption and then tries to justify that assumption based on the model’s output. And stick with the hard sciences.”

    Your impotence in finding a counterexample is all we need to know. Keep looking…

  5. Nonlin.org:
    “Even better, look up a situation outside of the Darwinism pseudoscience where someone builds a model based on an assumption and then tries to justify that assumption based on the model’s output. And stick with the hard sciences.”

    This shows that you’re not reading. You’re skipping the explanations and answering in haste. Otherwise you’d know that I exemplified with curve fitting in generic terms, not in tree building terms. So this was already covered. People might assume a linear relationship between two variables, fit it to a linear model, test the fit, and discover that it doesn’t fit, or that it does. This is everyday physics, for example, which qualify as hard science as far as we all know.

    Nonlin.org:
    Your impotence in finding a counterexample is all we need to know. Keep looking…

    More like your impotence at thinking and at reading for comprehension.

    Since you cannot understand things at the kindergarten level, I’ll stop here. Adios Straightjacketed nonthinking.

  6. Joe Felsenstein,

    And of course the “modification” that occurs on lineages as they diverge can be anything, even Design (as John Harshman has just noted, and as he and others have pointed out to you many times).

    I understand that John has discussed this many times but for me it makes common descent a claim that lacks clarity.

    It explains the pattern but only partially. Also the hierarchical pattern could be the result of intentional genetic change (design) versus inheritance.

    The evidence that is claimed to infer common descent may not be caused by common descent.

  7. Nonlin.org: As explained many times, what you do is curve fitting, not hypothesis testing.

    You mean “as asserted many times without making any sort of actual argument that the assertion is true”.

  8. colewd:
    Joe Felsenstein,

    I understand that John has discussed this many times but for me it makes common descent a claim that lacks clarity.

    I don’t know what more could be done to clarify the difference between common descent and the causes of mutations and fixation.

    It explains the pattern but only partially.Also the hierarchical pattern could be the result of intentional genetic change (design) versus inheritance.

    Now that’s a problem of clarity: how does intentional genetic change conflict with inheritance? What do you mean?

    Also, a tree entirely explains the pattern, while design (by which you mean separate creation) does not.

    The evidence that is claimed to infer common descent may not be caused by common descent.

    May not, but that’s the way to bet, and that’s how science works. You have shown no other explanation that provides the expectation of nested hierarchy.

  9. colewd: When you study the eukaryotic cell to try and understand such hard things as human disease, and if it is indeed designed, you will get much further if you make design you’re working assumption.

    OMagain: Then why are you not heading up a premier research department curing disease with ease?

    J-Mac: Why would he?

    Yes, why would colewd use his superior knowledge and “get much further” curing the diseases that plague humanity?

    I can’t think of a single reason now that you mention it.

    J-Mac: Evolution explains everything…even the evolutionary “advantage” of cancer… as it should…

    So your justification for colewd not curing disease is that evolution explains cancer?

    Out of interest, do you think that telling a cancer sufferer that they could be cured but until evolution stops claiming to explain everything they won’t be would be a good thing? From what you say it’s clear that you are happy that diseases are not being cured as a “punishment” for Evolution being understood as the source of biological diversity.

    So, just to be absolutely clear, colewd could help people “get much further” curing disease but he chooses not to, and J-Mac supports him in that.

    The germ theory was proposed by Girolamo Fracastoro in 1546, and expanded upon by Marcus von Plenciz in 1762. Such views were held in disdain, however, and Galen’s miasma theory remained dominant among scientists and doctors. The nature of this doctrine prevented them from understanding how diseases actually progressed, with predictable consequences.

    https://en.wikipedia.org/wiki/Germ_theory_of_disease
    Help us understand disease colewd? What possible reason do you have for denying humanity your wisdom?

    But seriously, it’s almost as if they know that they cannot do what they are claiming to do but for some reason cannot admit it, and instead make up spurious reasons for not doing it. But that can’t possibly be the case. The integrity and honor of J-Mac and colewd has been demonstrated over and over in these very pages.

    Curiouser and curiouser.

  10. Nonlin.org: Now, do you understand why your re-fried beans re-sampling doesn’t take care of the IID problem? Here is how it works: take a distributions and duplicate x times. Now you have x+1 distributions that are 100% correlated. Now re-fry re-sample however you want from this set and your samples will remain 100% correlated, so no IID.

    It is not necessary to have i.i.d. in order for a method to be consistent — that is, for it to converge on the correct answer as data is accumulated. There are “mixing conditions” that, when they apply, guarantee consistency.

    Here is how it works: Toss a coin independently, but suppose that every other time, you fail to toss before writing down the result. So after tossing a Heads, you mistakenly write down two Hs. After tossing a Tail you write down two Ts.

    The sequence of outcomes written down is thus not i.i.d. Will it converge to the correct probability of heads? Yes, it will.

    Similarly with trees. Now it is incumbent on you to argue that any nonindependence of changes at different genes is more like your example than it is like my example. I raised this issue before and asked for you to show this.

  11. Nonlin.org: 1. Whatever you do, your model is fine as long as you’re not trying to use it to prove its underlying assumptions (circular logic).

    3. See 1. No doubt, once UCD is the main assumption, your curve fitting model can be optimized to that assumption. But the model will never validate its own assumption. This is elementary and quite frankly I am very surprised you have a hard time accepting it.

    I addressed this issue before, and it is you, not me, who have avoided commenting on it. To make things easy, let me re-post that comment here:

    Joe Felsenstein:
    nonlin.org keeps saying that we are assuming Universal Common Descent and therefore cannot test it.

    The consilience of phylogenies from different parts of the genome (say, different genes) shows that there is a common phylogeny affecting all these parts of the genome. In doing that, we do not assume that all these phylogenies are exactly the same — we check to what extent they are similar.

    Is having a common phylogeny “Universal Common Descent”? If so, did we assume a common phylogeny when we went to test it? There is nothing in our inferences of phylogenies that forces them to be the same for different loci.
    So are we assuming UCD? No! So nonlin.org is simply wrong about that.

    More simply: we are confirming that there is a common underlying phylogeny by looking at the phylogenies inferred from different parts of the genome (or from different parts of the phenotype). Do our inferences assume that the phylogeny from different genes is the same? No!

  12. John Harshman,

    Also, a tree entirely explains the pattern, while design (by which you mean separate creation) does not.

    What do you mean a tree explains the pattern? Are you claiming a “tree” is the cause?

  13. colewd,

    Bill, seriously, what do you think creationism gains by their members showing off as impossible to reach intellectually? Do you think that by playing the fool creationists make their case compelling?

  14. colewd:
    John Harshman,

    What do you mean a tree explains the pattern?Are you claiming a “tree” is the cause?

    Yes. The tree, by which I mean the descent of populations from prior populations, combined with the splitting of speciation, is 100% of the explanation for the nested hierarchy of life. Changes (mutation followed by fixation) happen at various times on the tree and are inherited by descendant populations. Isn’t it a bit late for that to become clear to you?

  15. dazz: You give Bill too much credit.

    That might be so, but Nonlin is really something else. I never encountered anybody else who thought that dictionaries are so much more important than thinking. Anybody else who thought that assumptions cannot be tested because that’s not mentioned in the dictionary entry for assumption, or that evolution is false because a definition for evolution, found via google, didn’t mention all the phenomena involved. Nonlin gives the word ineptitude a whole new dimension.

  16. Entropy: That might be so, but Nonlin is really something else. I never encountered anybody else who thought that dictionaries are so much more important than thinking. Anybody else who thought that assumptions cannot be tested because that’s not mentioned in the dictionary entry for assumption, or that evolution is false because a definition for evolution, found via google, didn’t mention all the phenomena involved. Nonlin gives the word ineptitude a whole new dimension.

    I just hope that nonlin has all the energy in the world to keep tormenting you for another few years…. lol

  17. John Harshman,

    the descent of populations from prior populations, combined with the splitting of speciation, is 100% of the explanation for the nested hierarchy of life.

    You stand by the above claim?

    The descent of populations along with populations being isolated cause the hierarchal pattern?

    Do you claim that evolutionary innovations like the placenta are not part of the hierarchal pattern?

  18. J-Mac: I just hope that nonlin has all the energy in the world to keep tormenting you for another few years…. lol

    So do I. If I could, I would make you, Nonlin, and colewd the primary spokesmen for IDcreationism.

  19. colewd:
    John Harshman,

    You stand by the above claim?

    The descent of populations along with populations being isolated cause the hierarchal pattern?

    Do claim that evolutionary innovations like the placenta are not part of the hierarchal pattern?

    The innovation isn’t the cause or explanation of the pattern. The pattern is the distribution of innovations, so yes, each innovation is part of the pattern. But pattern itself is not the explanation of the pattern, and the parts of the pattern are not the explanation of the pattern, and the explanations of the innovations that make up the parts of the pattern are not the explanations of the pattern.

    This is indeed the central fact that you have never been able to understand.

  20. Joe Felsenstein,
    Mung,

    The innovation isn’t the cause or explanation of the pattern. The pattern is the distribution of innovations, so yes, each innovation is part of the pattern. But pattern itself is not the explanation of the pattern, and the parts of the pattern are not the explanation of the pattern, and the explanations of the innovations that make up the parts of the pattern are not the explanations of the pattern.

    Do you agree with John’s explanation above?

  21. Entropy: People might assume a linear relationship between two variables, fit it to a linear model, test the fit, and discover that it doesn’t fit, or that it does.

    You’re describing nonsense. Where is the hard science example? Link to it.

  22. John Harshman: You mean “as asserted many times without making any sort of actual argument that the assertion is true”.

    You’re wrong:

    3. Hypothesis testing requires an alternative (null) hypothesis and a procedure that demonstrates how the data available is compatible with the successful hypothesis and at the same time is statistically incompatible with the alternative hypothesis. Cladistic analysis is not really hypothesis testing since statistics are not essential to the methodology, but if it were it would need to compare against the alternative hypothesis to UCD which is “common design”. And of course UCD cannot be an assumption of such an analysis. However this rule is violated twice, first by the use of an assumption also presented as conclusion, and second by the prejudiced rejection of the alternative “common design” hypothesis before analysis. This clearly demonstrates that “cladistic analysis” can never be logically used as proof of UCD. Instead of hypothesis testing, “cladistic analysis” is ‘curve fitting’ where the cladistics model is best fitted to certain (conveniently selected!) morphologic/biochemical/genetic biologic data points.

  23. Nonlin.org: You’re describing nonsense.

    Are you seriously not understanding how one tests a linear assumption by comparing your assumption of how the data should be if there is a linear relationship, to how data the data actually is?

  24. Joe Felsenstein: It is not necessary to have i.i.d. in order for a method to be consistent — that is, for it to converge on the correct answer as data is accumulated. There are “mixing conditions” that, when they apply, guarantee consistency.

    Here is how it works: Toss a coin independently, but suppose that every other time, you fail to toss before writing down the result. So after tossing a Heads, you mistakenly write down two Hs. After tossing a Tail you write down two Ts.

    The sequence of outcomes written down is thus not i.i.d. Will it converge to the correct probability of heads? Yes, it will.

    Similarly with trees. Now it is incumbent on you to argue that any nonindependence of changes at different genes is more like your example than it is like my example. I raised this issue before and asked for you to show this.

    So that takes care of the “identically distributed” only. Of course when you duplicate the data by my method or yours you end up with the same distribution. But what does that prove? To go back to DNA, this tells you that re-sampling gets you the same ratio of A, C, G, T. So what?

    I am not clear exactly how you use statistics in the first place and why you need to resample. And since you’re doing curve fitting, I already said that, yes of course, your output tree will be compatible with your assumptions including UCD. Say you got the Pinniped DNA and now compare with Ursidae, Felidae, Canidae. By definition one of these three will be more correlated with the Pinniped than the other two. So what? Just as well you can correlate shoe sizes with cloud shapes and automobiles. Does this tell you UCD applies to those three? No way!

  25. Joe Felsenstein: Is having a common phylogeny “Universal Common Descent”? If so, did we assume a common phylogeny when we went to test it? There is nothing in our inferences of phylogenies that forces them to be the same for different loci.
    So are we assuming UCD? No! So nonlin.org is simply wrong about that.

    More simply: we are confirming that there is a common underlying phylogeny by looking at the phylogenies inferred from different parts of the genome (or from different parts of the phenotype). Do our inferences assume that the phylogeny from different genes is the same? No!

    As shown, Berkeley thinks you’re wrong:
    http://www.ucmp.berkeley.edu/clad/clad1.html assumptions
    http://www.ucmp.berkeley.edu/clad/clad2.html method

    Why don’t you argue against them?

    And UCD or not, phylogenesis (evolution) is also an assumption, hence not cannot be a conclusion.

  26. Rumraket: Are you seriously not understanding how one tests a linear assumption by comparing your assumption of how the data should be if there is a linear relationship, to how data the data actually is?

    “Linear” is not an assumption needed to test if a curve is linear within a certain error margin.

    Look up the difference between Hypothesis and Assumption.

  27. Nonlin.org: You’re wrong:

    3.Hypothesis testing requires an alternative (null) hypothesis and a procedure that demonstrates how the data available is compatible with the successful hypothesis and at the same time is statistically incompatible with the alternative hypothesis. Cladistic analysis is not really hypothesis testing since statistics are not essential to the methodology, but if it were it would need to compare against the alternative hypothesis to UCD which is “common design”. And of course UCD cannot be an assumption of such an analysis. However this rule is violated twice, first by the use of an assumption also presented as conclusion, and second by the prejudiced rejection of the alternative “common design” hypothesis before analysis. This clearly demonstrates that “cladistic analysis” can never be logically used as proof of UCD. Instead of hypothesis testing, “cladistic analysis” is ‘curve fitting’ where the cladistics model is best fitted to certain (conveniently selected!) morphologic/biochemical/genetic biologic data points.

    While it’s true that common descent is seldom explicitly tested against an alternative of no common descent (but see Theobald D.L. A formal test of the theory of universal common ancestry. Nature 2010; 465:219-223), there is nevertheless an implicit test every time a tree or node is tested against the expectation of randomized data, which is that there will be no significant support for one topology over another. Likelihood ratio tests, gene jackknifing, bootstrappiing, and a host of others are all such tests.

    Now of course “common design” can’t be a null model because it has no expectations and can’t be rejected, even in principle, by any conceivable data.

  28. Nonlin.org: By definition one of these three will be more correlated with the Pinniped than the other two. So what?

    But why would it? Why should such a relationship hold for different portions of the genome that don’t constrain each other in sequence? Why do we nevertheless discover that they ARE correlated? What, other than a common genealogical history, explains that?

  29. Rumraket: Why should such a relationship hold for different portions of the genome that don’t constrain each other in sequence?

    Who told you the “different portions of the genome… don’t constrain each other in sequence”? Just to be clear, what do you mean “don’t constrain each other in sequence”? This is the discussion with Joe F. – read that.

  30. Rumraket: Are you seriously not understanding how one tests a linear assumption by comparing your assumption of how the data should be if there is a linear relationship, to how data the data actually is?

    Very seriously. With a straight face to boot. Evan after I explained it many times to her/him. I just didn’t use puppets. I’m really too embarrassed of having to explain something that basic, so many times. Yet Nonlin just won’t get it.

  31. Nonlin.org: “Linear” is not an assumption needed to test if a curve is linear.

    You understand the difference between the model and the data, right?

    You have a model that, if true, says the data should lie ideally on a straight line.
    That’s your model, your assumption.

    Then you have the data you get by doing measurements.

    You do your measurements and plot them into a coordinate system and see how the data points lie. Draw a line through them if you can.

    You draw a line where you expectation is (given the linear model), and then you compare the data from your measurements to the model.

    Suppose this is what you get:

  32. John Harshman: there is nevertheless an implicit test every time a tree or node is tested against the expectation of randomized data, which is that there will be no significant support for one topology over another.

    Now of course “common design” can’t be a null model because it has no expectations and can’t be rejected, even in principle, by any conceivable data.

    Testing one topology versus the other based on a set of assumptions including the assumption of “one topology” doesn’t prove “topology”

    So how would we reject the UCD when it is an assumption of the topology? Why not start from a point where UCD and “evolution” in general is not an assumption? That would be proper science!

  33. Entropy: Very seriously. With a straight face to boot. Evan after I explained it many times to her/him. I just didn’t use puppets. I’m really too embarrassed of having to explain something that basic, so many times. Yet Nonlin just won’t get it.

    I have a made a drawing now even the thickest person on the planet should be able to get. I cheated a bit by drawing a red line through the data.

    There’s some chance a chimp would get it.

  34. Nonlin.org: So how would we reject the UCD when it is an assumption of the topology?

    There could be an absense of one topology. Look at the pretty picture, think with your brain. The data can simply fail to conform to expectations. That’s how test your assumptions. There’s nothing mysterious about it.

    [What should the data look like if the assumption is right] is compared to [what does the data actually look like?]

    The model says the individual phylogenetic trees should agree. But they are not constructed in a way that forces them to(you can look up how the actual algorithms work, or ask the nice experts here like Joe and John), they are constructed without bias towards common descent.
    They are then compared to see if they really do agree on a common topology. Understand that they could simply fail to do that, and there are trees known that deviate.

  35. John Harshman,

    Now of course “common design” can’t be a null model because it has no expectations and can’t be rejected, even in principle, by any conceivable data.

    Functional innovation would be an expectation of common design.

    Use of common parts would also be an expectation of common design.

  36. Nonlin.org: “Linear” is not an assumption needed to test if a curve is linear within a certain error margin.

    Of course it is. The data has to be fitted to a line in order to test if it fits. The mere act of fitting the data to a line implies the assumption that the data follow a line.

    Nonlin.org:Look up the difference between Hypothesis and Assumption.

    I know the difference, and I know that we can test an assumption by thinking of it as a hypothesis. We can also test a hypothesis by inferring what we should see if we assumed the hypothesis to be true, and then checking. Again, language is our tool, not our enslaver.

    From some statistics place (by way of example):

    As you prepare to conduct your statistics, it is important to consider testing the assumptions that go with your analysis. Assumption testing of your chosen analysis allows you to determine if you can correctly draw conclusions from the results of your analysis.

    Hum. Seems like assumptions can be tested despite Nonlin’s command that this shouldn’t be so! How dare them!

    What? Testing assumptions in business? No way!

    When you’re developing a strategy for a new business, testing assumptions in a logical order gives you the best chance to make course corrections early — and not waste time and money.

    Testing linearity? No, of course not:

    Regression diagnostics: testing the assumptions of linear regression

    It must be all lies because testing is not mentioned in the dictionary entry for assumption quoted by Nonlin.

    Once something is called an assumption we cannot make it into a hypothesis because Nonlin forbids it. All praise Nonlin The Lord. His Words Are His Commands. Though shall not contradict Nonlin’s Word.

  37. Rumraket: … think with your brain …

    And that’s where you lose Nonlin. You have to give her/him a dictionary that includes testing assumptions, and that Nonlin likes (good luck with that). If it’s not in Nonlin’s dictionary it’s immovable, like the Earth.

  38. Nonlin.org: So how would we reject the UCD when it is an assumption of the topology? Why not start from a point where UCD and “evolution” in general is not an assumption? That would be proper science!

    UCD is not an assumption of “the topology”; common descent within the taxa being studied is an assumption of a tree, and once again that assumption is tested by comparing that model with another model. If there is no common descent, no tree should be significantly preferred by the data over any other tree. Thus a strong preference for one tree rejects the no common descent model.

    Also, read Theobald, for chrissake.

  39. colewd:
    John Harshman,

    Functional innovation would be an expectation of common design.

    If this means anything, and I’m not sure it does, it refers to your main confusion, the one between an explanation of nested hierarchy and an explanation of the various differences in features that happen within the hierarchy. Have you in fact learned absolutely nothing during all our conversations over a period of years?

    Use of common parts would also be an expectation of common design.

    That might actually be a testable and relevant claim, but it’s a useless statement until you flesh it out with some kind of clear prediction of what the actual data should look like. I could try to flesh it out but it wouldn’t go well for you. For example, I would expect reused common parts to be identical among species. Certainly one wouldn’t expect them to vary in a way conforming to a nested hierarchy. We would also need to figure out what entities are being created from these common parts. Are they species, families, phyla, or something else? What level of common descent are you trying to deny?

  40. colewd,

    One analogy to the discussion as to whether common descent brings about the changes: The shape of the roof of our house largely determines how the rain will run off it. But it does not cause the rainfall (or at least, I don’t think it does). A discussion of how the rain runs off the roof can go forward without any discussion of what caused it to rain.

  41. Nonlin.org: So that takes care of the “identically distributed” only. Of course when you duplicate the data by my method or yours you end up with the same distribution. But what does that prove?

    In both my example and yours, we add new sets of tosses of the coin. In your case, they do not change the fraction of Heads, as they are exact duplicates of all of the previous sets of tosses. My case is very different: adding new pairs of tosses we get new outcomes. The issue is whether the fraction of Heads will converge on the true probability of Heads. In your case it does not, in my case it does. So there is some issue of what sort of nonindependence occurs. I asked you to present an argument that the nonindependence between different genes is like your case instead of like my case. But you seemed to have missed the distinction between them. This calls into question your confident and dramatic assertions that nonindependence of outcomes in different genes shows that they do not help us check whether there is an underlying phylogeny.

    To go back to DNA, this tells you that re-sampling gets you the same ratio of A, C, G, T. So what?

    You did not understand my example.

    I am not clear exactly how you use statistics in the first place and why you need to resample.

    I mentioned the block-bootstrap, but it is not essential to my argument. Set it aside, no problem, just provide the argument that nonindependence is sufficiently strong that we routinely expect new genes to show the same phylogeny just because of the nonindependence.

    And since you’re doing curve fitting, I already said that, yes of course, your output tree will be compatible with your assumptions including UCD. Say you got the Pinniped DNA and now compare with Ursidae, Felidae, Canidae. By definition one of these three will be more correlated with the Pinniped than the other two. So what? Just as well you can correlate shoe sizes with cloud shapes and automobiles. Does this tell you UCD applies to those three? No way!

    First, “curve fitting” is a vague analogy here. Inferring phylogenies is not “curve fitting”. Phylogenies are not curves.

    Your argument about Pinnepedia etc. compares apples to oranges. The issue is, when we take the same set of species (say carnivores plus pinnipeds) and use different genes (or regions of the genome) to infer phylogenies, what do we see, and is this evidence for there being an underlying phylogeny. Spoiler: consilience, and yes, it is.

  42. Nonlin.org: And UCD or not, phylogenesis (evolution) is also an assumption, hence not cannot be a conclusion.

    Totally wrong: there can be a comparison of phylogenies derived from different genes, to see whether they are similar.

    John Harshman has explained this beautifully simply, here. I refer you there.

Leave a Reply