Mo' parsimony, mo' problems

Leonard Finkelman writes...

1. Introduction

Submitted for your approval: the strange case of Henry Levine, a lawyer living in Bethesda, Maryland. On Monday, 8 February, Mr. Levine received a strange text message from an unfamiliar phone number. The message contained only a single word: “Raincheck.” That message was followed by another, and then another, all within moments of one another. Each contained only that same word, each from a different, unfamiliar number. Two things would happen to Mr. Levine before the sun set. First, he would receive several hundred more text messages. Second, he would provide the means to understanding one way that paleontology is unique among the life sciences.

As a result of the events that I’ll call RaincheckGate [1], Mr. Levine started a Twitter account. There was an unintentional irony in this choice: you see, Twitter had just played host to an academic dispute whose outcome would determine the philosophical framework through which Mr. Levine’s story should be viewed. This debate, known as ParsimonyGate, offers a similar opportunity to paleontologists: ParsimonyGate cuts to the very core of how paleontology should be conceived.

I’ll admit that the connection between RaincheckGate and ParsimonyGate may seem tenuous at first. The latter debate started when editors of the journal Cladistics announced that research submitted to the journal should either conform to specific theoretical assumptions or “philosophically” justify any divergence from those assumptions. One doesn’t find any explicit mention of phone numbers or rainchecks when looking through the history of the resulting dispute. But if you’ll bear with me, I’ll show how Mr. Levine’s text messages raise the issues inherent in ParsimonyGate and how those issues distinguish paleontology from other life sciences.

For the benefit of the different audiences that might read this essay I've broken it up into more easily digestible chunks. Section twoexplains what parsimony is and how the application of that concept differs between philosophy and biology. For readers already familiar with principles of parsimony, section three lays out both sides of the ParsimonyGate debate. And readers who have already lived through that battle can skip to section four, where I explain how ParsimonyGate gives paleontologists an opportunity to distinguish themselves among life scientists.

2. Principles of Parsimony

Why would hundreds of people all send the same message to the same unassuming lawyer? People mistakenly send messages to wrong numbers all the time; perhaps it was just coincidence that Mr. Levine received an unusual number of these messages in a relatively short timespan. But if each of these messages was a genuinely independent event, then coincidences would start piling up: each person would have to make the myriad mistakes necessary to enter Mr. Levine’s phone number; each person would have to be struck by the desire to claim a raincheck at roughly the same time; each person would have to phrase their claim in essentially the same way; and so on. All of these independent events could happen, but intuition cries out for a simpler explanation.

If the cry sounds familiar it’s probably because empiricists have so often preached the principle known as Ockham’s Razor. While Ockham's Razor is popularly summarized as “the simplest explanation is the best,” this phrasing is unhelpful. After all, how do we measure simplicity? Philosophers suggest that conservativeness, or parsimony, is one such measure: when given a choice between multiple explanations, we should always prefer the one that requires the fewest number of theoretical commitments unless we have a compelling reason to prefer otherwise [2]. Explanations that are “parsimonious” by this philosophical standard are simpler than alternatives not only because they require fewer assumptions, but also because they are more likely to be consistent with future data.

The particular theoretical commitments that should be minimized vary from one discipline to another. In biology, some researchers trying to reconstruct evolutionary history (technically known as phylogeny) suggest a specific way of measuring parsimony. When Darwin first proposed the theory of natural selection, he argued that characteristics inherited from common ancestors (synapomorphies, in more modern terms) give us more information about evolution than do characteristics that appear similar but have different ancestries (homoplasies). For example: the fact that humans and whales both have lungs tells us that whales shared a terrestrial ancestor with humans, but the fact that whales and fish both have tail flukes doesn’t tell us much about the evolution of either group. The editors of Cladistics hold that homoplasies—evolutionary coincidences—are an onerous burden on the theory of natural selection, and so when we reconstruct phylogenies we should minimize our commitments to homoplasies ex hypothesi. This is the standard of cladistic parsimony.

The explanation of RaincheckGate certainly conforms to the philosophical standard of parsimony. Each of the hundreds of people who sent Mr. Levine these messages was trying to claim a free burrito from the fast food chain Chipotle through a phone number that differed from Mr. Levine’s by a single digit. Instead of hundreds of people independently making hundreds of mistakes that all yielded the same outcome, hundreds of people jointly made similar mistakes.

If we try to view RaincheckGate through the lens of cladistic parsimony, however, we can start to see why so many biologists have so vocally criticized the editors of Cladistics. It is debatable whether or not the actual explanation of RaincheckGate is the most parsimonious by cladistic standards. The RaincheckGate problem wasn’t just that the two phone numbers differed by a single digit; the problem was that Mr. Levine’s phone number was one digit longer than Chipotle’s, and so anyone in Mr. Levine’s area code who entered the last digit of Chipotle’s number twice would send their message to Mr. Levine instead. Which mistake requires fewer coincidences: to enter one digit of a phone number incorrectly or to accidentally add a digit to the end of an otherwise correct phone number? It’s a difficult question to answer: not all mistakes are created equal, but it’s difficult to measure inequalities.

3. Opening the Parsimonygate

Cladistic parsimony has difficulty answering questions like the one just raised. Some homoplasies seem easier to evolve than others, just as some mistakes, like dialing a phone number incorrectly, are easier to make than others. But cladistic parsimony treats all homoplasies as equally unfavorable. There are obvious qualitative differences between the two aforementioned different ways to dial a wrong number, but it almost seems disingenuous to stipulate that the two are quantitatively the same. Nevertheless, cladistic parsimony makes the analogous demand on homoplasies. Biologists who study more easily evolvable systems therefore have good reason to use methods of phylogenetic reconstruction that accommodate other models of evolution.

This criticism of cladistic parsimony opens the door—or, more appropriately, a gate—to others, hence ParsimonyGate [3]. Some biologists argue that evolution is demonstrably not parsimonious. The standard model of natural selection holds that evolution is slow, steady, and incremental—per Darwin, selection is ‘daily and hour scrutinizing, throughout the world, every variation, even the slightest’ (1859, 84)—but any divergence from this model may produce results that are not parsimonious (by the cladistic standard), resulting in statistical inconsistency (Felsenstein 1978). Another objection is that cladistic parsimony is subject to the error of long branch attraction: as divergent lineages evolve they become more and more likely to acquire similar traits independently, but parsimony measures do not account for temporal data and so may overestimate the relatedness of older lineages (Schuh & Brower 2000, 139-140). Given these potential failings, those siding against the editors of Cladistics argue that biologists wouldn’t be justified in assuming cladistic parsimony.

Should biologists assume cladistic parsimony or not? It’s not my goal to decide that question—and even less to cast my lot in the particulars of ParsimonyGate [4]—but it’s worth noting, as Sober (2015) does, that many biologists do productively assume cladistic parsimony without obvious error (199-200). In an earlier essay, Sober  (1983) explained why this should be so: the objections raised against cladistic parsimony show only that this method of determining relatedness might be worse than others, and so the theoretical value of parsimony can only be determined by a posteriori comparison. When comparing one theoretical method against another, choosing which method should be assumed and which should be compared is arbitrary at best. Without having justification to assume otherwise, then, biologists do no harm in assuming cladistic parsimony—which seems very well in line with the philosophical standard of parsimony.

The real question at hand in ParsimonyGate is therefore whether or not biologists have a priori justification in preferring phylogenetic methods other than cladistic parsimony. One group of life scientists does have that justification. The gate that Henry Levine opened is one through which paleontologists should walk.

4. How to be a paleontologist

Whether or not cladistic parsimony is an appropriate assumption for paleontological research depends very much on how the field is conceived. Turner (2011) recounts how this question has been engaged over the years: whereas Darwin (1859) conceived paleontological data as incomplete and therefore misleading, Eldredge & Gould (1972) argued that the apparent incompleteness of the fossil record should be considered as data per se. Darwin’s view followed from his consideration of paleontology as continuous with other evolutionary disciplines; for their part, Eldredge & Gould argued that paleontology has foundations distinct from other life sciences and should be considered an independent discipline with a unique research program.

What do paleontologists do and how do they do it differently from other scientists? Introductory texts (e.g. Raup & Stanley 1971 and Benton 2005) emphasize two primary targets of paleontological study: functional morphology, aimed at determining how extinct organisms actually made use of their biological traits, and phylogenetic reconstruction. What can make paleontological study in these fields unique, however, is precisely what makes cladistic parsimony an unlikely theoretical assumption in paleontology.

Consider functional morphology. This is an important antecedent to phylogenetic reconstruction since the traits that we compare in order to infer evolutionary history may be functionally defined. Where zoologists and botanists have the luxury of using field and lab observations to determine the functions of traits, paleontologists must make inferences about biological functions from clues in the fossil and geological records. To recall an example from my last essay: Dimetrodon and Spinosaurus sails share structural similarities, but are they the same kind of thing--that is, the same trait? Is this trait also shared with, say, Edaphosaurus? Paleontologists may collect morphological, environmental, and bio- or geo-chemical data to determine which of these sails is more or less like the others. The question of whether or not Dimetrodon, Spinosaurus, and Edaphosaurus have a notable trait in common might not have a simple paleontological answer. By contrast, cladistic parsimony treats traits as binary: either an organism has a sail or it doesn't, just as you've either dialed a correct phone number or you haven't. Cladistic parsimony therefore ignores much of the important information gleaned from the paleontological study of functional morphology (cf. Turner 2011, 194-196).

Consider now phylogenetic reconstruction in paleontology. Phylogenetic reconstruction is an important part of other fields ranging from molecular biology to ecology. Each of these disciplines engages in phylogenetic reconstruction for different purposes. In paleontology, the goal is not simply to understand how taxa are related to one another, but also to determine the particular historical events that have influenced evolution. Information such as the amount of time that a lineage remains morphologically stable, when collated with geological data, may help to explain causal relations between (say) extinctions and environmental events. For example: sharks, turtles, and crocodiles—all groups with remarkable morphological stability over long periods of time—may have weathered the K-Pg Event better than other groups did. This is valuable information that may help to answer an important paleontological question—why did so many taxa disappear during this mass extinction?—but it would be lost through the assumption of cladistic parsimony because that method has no measure of trait stability or lineage duration (Schuh & Brower 2000, 140).

Following Eldredge & Gould, we would recognize paleontology as a distinct and matured discipline (rather than as a handmaiden to other life sciences) if we could find some kinds of information uniquely valued in paleontological research. Qualitative similarity and temporal placement seem to be two such kinds of information. Inability to account for these kinds of information also happens to be the greatest weakness with cladistic parsimony.

This, then, is one more way to differentiate paleontology from other life sciences: paleontologists always have an a priori reason for favoring methods other than cladistic parsimony whereas other life scientists do not. This is not to say that paleontologists cannot productively employ cladistic parsimony in their research; they can and many do. Rather, it is to turn the tables on the editors of Cladistics and say that paleontologists should only use the method in light of some a posteriori justification. A paleontologist should have a good reason for using a method of phylogenetic reconstruction that is weakest in accounting for information that paleontologists strongly demand.

5. Conclusion

Why did Henry Levine receive so many strange text messages on that fateful February day? We do have the answer, but is it the simplest one that we could have? Like so many questions, the satisfaction you'll feel with different answers will depend on why you asked. So it should be for paleontologists: if they are asking questions for specifically paleontological reasons, then the means by which they find the answers will only be satisfying if the answers serve those reasons. Since paleontologists want information that answers implied by cladistic parsimony are unlikely to provide, their lot in ParsimonyGate seems cast against the cladists.

As for other life scientists--it's an interesting question, but for now we'll have to take a raincheck.

 

Notes

  1. Alas: “BurritoGate” was already taken, and several times over.

  2. Formulations of Ockham’s Razor vary. My former dissertation advisor, Massimo Pigliucci, wrote an insightful post about the subject back on his old blog; for a longer treatment, I recommend Elliott Sober’s recent book (also cited below).

  3. To be fair, ParsimonyGate isn’t only a theoretical debate. Much of the dispute has centered on the behavior of the parties involved and disciplinary standards of decorum. I would recommend that the biologists look to philosophers, who figured out the best means of conflict resolution decades ago.

  4. If I were pushed, though, I’d agree with just about everything expressed in this post from the PhyloBotanist blog.

References

  1. Benton, M. J. (2005). Vertebrate Palaeontology (Third Edition). Blackwell Publishers.

  2. Darwin, C. (1859). On the Origin of Species. John Murray.

  3. Eldredge, N. & Gould, S. J. (1972). Punctuated equilibria: an alternative to phyletic gradualism. In Schopf, T. J. M. (ed.), Models in Paleobiology. Freeman, Cooper and Company, San Francisco, 82-115.

  4. Felsenstein, J. (1978). Cases in which parsimony or compatibility methods will be positively misleading. Systematic Biology, 27(4), 401-410.

  5. Raup, D. M. & Stanley, S. M. (1971). Principles of Paleontology. W.H. Freeman and Company.

  6. Schuh, R.T. & Brower, A.V.Z. (2000). Biological systematics: principles and applications. Cornell University Press.

  7. Sober, E. (1983). Parsimony in systematics: philosophical issues. Annual Review of Ecology and Systematics, 14, 335-357.

  8. Sober, E. (2015). Ockham's Razors. Cambridge University Press.

  9. Turner, D. (2011). Paleontology: A philosophical introduction. Cambridge University Press.