Statistical Testing of Common Ancestry: Something To Be Embarrassed About?
David Baum, a biologist at the University of Wisconsin-Madison, corresponded with NCSE staff about a challenge he and his colleagues faced. He shared this account of his experience trying to publish research which, in part, attempted to put certain creationist claims to the test.
Why are there so few published statistical tests of common ancestry? When two colleagues (Cécile Ané and Bret Larget) and I arranged a graduate seminar course at UW-Madison on statistical evidence for evolution in spring 2014, we didn't expect to tangle with that ultimately sociological question. However, in the course of the seminar, and during the process of publishing the resulting papers, we not only learned more about the nature of evidence for evolution but also gained some insight into the surprising reluctance of scientists to discuss this question.
When we planned the seminar course we had already concluded that the most direct evidence of evolution is evidence of common ancestry. Who could doubt that evolution had happened if living organisms as different as, say, humans and spider monkeys could be shown to trace to common ancestry? There is no way to get from a single ancestor to two such different descendants without one or both descendant lineages evolving considerably. Given this insight and our desire to have student do something concrete, Ané, Larget, and I asked students to implement statistical tests of common ancestry using data for members of the primate clade. We chose primates partly because of the abundance of high quality data that are available. But also, we couldn’t help noticing that the common ancestry of humans and non-human primates remains one of the biggest sticking points for religious opponents of evolution. We reasoned, perhaps naively, that establishing the common ancestry of humans and spider monkeys, or even just humans and chimpanzees, would go a long way towards contradicting the frequent creationist claim that scientific data do not support evolution.
The class collected diverse kinds of primate data from the literature with the goal of applying statistical tests for comparing the hypothesis of common ancestry (CA) with four alternative separate ancestry (SA) models: 1) species SA (each species of primate originated separately), (2) family SA (each primate family originated separately and then underwent evolutionary diversification to give rise to its current species), (3) dual SA of haplorhines and strepsirrhines (each major primate clade arose separately but then diversified into its component families and species), (4) dual SA of humans and other primates (primates have a shared phylogenetic history, but humans are a disconnected lineage). We soon discovered, though, that surprisingly few statistical tests of CA have been described in the literature. Indeed, prior to our seminar, only three published papers had proposed new statistical tests to compare CA and SA (Penny et al. 1982; Theobald 2010; White et al. 2013). This puzzling fact is the main motivation for this blog post: Why are there so few published tests of CA? Should the participants in our course be embarrassed that we wanted to formally test evolutionary theory?
One possible explanation for the few published statistical tests is that there are actually only a few ways to compare predictions of CA and SA, and these have already been described in those three papers. However, the team of talented graduate students who took the seminar class, from programs as diverse as botany, genetics, geoscience, statistics, and zoology, quickly ruled out this possibility. As a group we sketched out many possible tests of common ancestry beyond the three published ones and then developed a few new tests for diverse kinds of data (gene sequence, morphology, geography, chromosome numbers, and fossils). This goes to show that the paucity of work on statistical tests of CA is not indicative of it being a dead-end topic. Quite the contrary, I am sure there remains a rich seam of interesting theoretical work to be done to develop new, yet more powerful tests of CA versus SA.
So, if we can rule out the idea that there are no new tests of CA to be developed, we need a different explanation for the lack of publications in this area. One possibility: scientists have not published more formal tests of CA vs. SA because such tests yield results that are unfavorable to evolutionary theory. This explanation might appeal to a creationist, especially one with a paranoid streak, who thinks that scientists would hide data that supported SA. However, this could not be further from the truth. I say this because every test we developed yielded an incredibly strong rejection of SA. And when I say “strong,” I really mean it. The p-values we obtained were often much smaller than 10-80, a noteworthy number since this is the probability of picking, at random, one atom in the entire known universe. Sorry, Mr. or Ms. Creationist, there is no attempt to hide data that are embarrassing to us scientists. If anything, scientists are hiding data that ought to be embarrassing to those who doubt evolution!
During the process of trying to publish our paper, two other sociological factors came into focus that better explain the paucity of work in this field. First, the current scientific publishing ethos stresses the sharing of novel discoveries. In that context, “discovering” that primates do indeed trace back to CA is of limited significance given that this is already universally accepted within science. The fact that some non-scientists still have doubts does not carry weight when the audience for scholarly journals is composed almost entirely of professionals in the field. Second, and perhaps more important, there is a fear that by publishing papers that focus on testing SA, we risk giving undue credit to creationism by implying that it is science. That is, to say, the argument that creationism is non-scientific, where “science” is understood to entail the formulation of testable hypotheses, would be undermined by studies that test claims of creationism. And, the argument would go, this is true even if the tests in question royally reject SA and favor CA.
To explain how I came to appreciate the latter argument, and why I think it is the primary reason why there has not been more work on tests of CA, consider the editorial review process that our submission to Evolution experienced.
The manuscript we submitted is largely about phylogenetic data and statistical tests, but there were three mentions of creationism per se. When introducing the species SA model we justified it as “corresponding loosely to the doctrine of Special Creation, against which Darwin pitted his theory.” Then, after introducing the family SA model we noted: “This model loosely resembles the creationist doctrine called baraminology, which assumes creation of separate “kinds,” typically equated with the family rank (Gishlick, 2006).” Finally, in the discussion we included these sentences:
Given the overwhelming statistical support for CA, and specifically the CA of humans and other primates, it might be tempting to view this work as a refutation of creationism. However, although it is hard to see how anybody could objectively look at these data and still favor SA, we must acknowledge that deeply held beliefs, such as those that underlie rejection of evolutionary theory, are rarely if ever overturned by “evidence” (Lord et al. 1979; Mooney, 2011). For this reason, defending evolution against its critics is not a sufficient reason to conduct statistical tests of CA.
These references to creationism seemed pretty benign, yet the editor was quite adamant that all be expunged from the paper. I was disappointed. To be sure Evolution has a mission to publish research that advances our “understanding of evolutionary phenomena and processes,” and does not claim a political or educational agenda. Nonetheless, since creationism is a political and religious factor that negatively affects our attempts to educate the public on “evolutionary phenomena and processes,” it seemed to be going too far to disallow all uses of the word “creationism.” Creationism should not be, like “Voldemort,” something we can refer to only obliquely: “the-ideology-which-must-not-be-named!”
By way of reinforcements, I contacted Glenn Branch at NCSE for his reaction to the journal’s censoring of “creationism.” Branch provided the following, insightful commentary, as well as some suggested rewording for our sections, which I have removed for reasons of brevity. [To clarify, his comments only relate to two of the three references to creationism, because it was not until the next submission that the editor noticed, and asked us to cut, the sentence mentioning baraminology.]
I certainly sympathize with the idea that it would be neither accurate nor helpful for Evolution to publish a paper that suggests, in the absence of convincing evidence, that creationism is scientifically credible or even that it is worth taking seriously. But the passages that the editor wants to omit don't make any such suggestion, and there are, I think, good reasons to preserve them, if perhaps with some minor revisions.
The first passage (presumably just from "corresponding" to "theory") implies that SA is scientific (at least in the sense that it is scientifically testable), but since "Special Creation" is said only to "loosely" correspond to it and is described as a "doctrine," the passage can't naturally be read to imply that Special Creation is scientific. I understood this as a useful historical gloss, intended to remind the reader of the historical background of Darwin's development of CA. (And in fact I was reminded of a textbook discussion that I don't like precisely because it represents Darwin as reacting against SA without any mention of Special Creation.)…
The second passage doesn't even hint that creationism is scientifically credible -- in fact, it presupposes (given the results of the paper) that it isn't! But it does imply that creationism is, or might be, scientifically testable ("it might be tempting to view this work as a refutation of creationism"), and perhaps that's what's giving the editor pause.
Without conducting a whole philosophy seminar here, I will suggest that creationism generally involves a mix of testable and untestable claims. Take, for example, young-earth creationism, which has as a central principle the idea that God created the universe about 10,000 years ago. Prima facie, this can be decomposed into a testable component -- that the universe is about 10,000 years old -- and an untestable component -- that God created the universe. (This is only prima facie, because the testable component can be made untestable by using auxiliary assumptions, e.g., that God created the universe with the appearance of great age; it would take us too far afield to pursue the ramifications.)
By the same token, Special Creation can be decomposed into the testable SA (or versions thereof) and the untestable claim that God created the species separately.…
I should add that creationism is not a problem that goes away if you throw enough science at it, so your paper's warning to readers is quite appropriate and useful.
With Branch’s permission, I shared his arguments with the editor of Evolution, but she was unmoved. When, in a later round of correspondences, I shared my frustration about the removal of what seemed to me reasonable mentions of creationism, the editor replied thus:
It has been my understanding throughout that your group's development of tests for common ancestry is motivated by the need for such tests to address hypotheses of current scientific interest, of which special creation is not one. This response from you indicates otherwise, and this makes me think I misunderstood. If refuting the doctrine of special creation is the sole or even a primary motivation for this development, as your email suggests, then I seriously question whether it should appear in Evolution.
I can only respond to your last paragraph by noting that the outcomes of scientific tests of hypotheses that arise from miraculous scenarios do not seem likely to be persuasive to those inclined to accept miracles. After all, when fossils, clearly of extinct marine organisms, were found on mountaintops, wasn't their distance from appropriate habitat attributed to deceit of the devil?
Although she insisted we remove all mentions of creationism, she seemed to understand that we might want to raise these issues in less formal settings such as this: “You are, needless to say, free to make this point in talks or in publications in other venues, but please leave it out of this paper.” (In granting permission to quote those emails, the editor noted that the journal’s mission statement describes it as, “the premier publication devoted to the study of organic evolution and the integration of the various fields of science concerned with evolution. The journal presents significant and original results that extend our understanding of evolutionary phenomena and processes.”)
So, what can we learn from this? I have come to believe that the editor’s concern about validating creationism by referring to it in an academic journal is revealing of a well-reasoned perspective that is widespread in evolutionary biology. The argument takes the view that if we just get on with the science, this pesky religious/political movement will eventually go away and leave us alone. A relatively direct symptom of this thinking is the reluctance to statistically test CA since such tests could be viewed as addressing claims made by creationists. However, while I understand this argument, I don't agree with it.
It is a good practice for any scientific theory to ask the question, How might one quantify its evidential support relative to competing theories? I would argue that the need for such quantification applies even if the alternative theory is associated with a non-scientific worldview. Consequently, we should not let fear of implying that there is a debate within science about the truth of evolution (which there is not), prevent us from conducting formal quantitative tests of separate ancestry theories that ultimately trace to anti-evolutionary religion. As Branch succinctly argued, the creationist ideology as a whole may not be science, but its proponents often do make scientific claims. Rather than ignoring claims of the ideology-which-must-not-be-named, we should bring the power of science to bear. And within such an agenda, it is critical to use statistical tests to rigorously evaluate (and in our experience refute) alternative separate ancestry models.
It is also worth bearing in mind that by conducting unbiased tests of common ancestry versus separate ancestry we go some way towards undermining the dubious claim that scientists promote evolution based on a secular humanist agenda rather than data per se. I certainly found it very helpful last summer, when I spoke to a group of evangelical Christians, to be able to say that we had taken the claim of human SA seriously, done rigorous tests that were in no way biased towards the evolutionary perspective, and had come out with very strong statistical support for CA and against SA. This made it much easier to argue that if they want their religious views to be compatible with science in general then they really ought not to subscribe to creationism, or at least not to SA.
Scientists overwhelmingly agree that evolution is the best explanation for the diversity of life, and that the evidence supporting the fact of evolution is extremely strong. However, even among scientists who can provide a convincing qualitative argument for why we should accept evolution, these arguments all too rarely use a formal, statistical framework, complete with p-values, posterior probabilities, and the like. In an uncontested area of science, like atomic theory, it might suffice that scientists are aware of the flavor of the evidence without needing to cite its statistical support. But for a topic like evolution, which faces anti-science attacks, any scientist who deals directly with the public would be served by being able to point to quantitative statistical tests of common ancestry.
Thus, even ignoring its intrinsic scientific interest, as scientists concerned by anti-scientific views we should engage more broadly in mathematically analyzing the different kinds of evidence that support evolutionary theory. Statistically testing common ancestry is not something we should be embarrassed about.