Stratigraphic data and phylogenetic hypotheses
The polarization seen in this debate arises, says Dr Dan Fisher, from a desire for both sides to argue over different questions. He cautions that the zeal of cladists to discard stratigraphic data as irrelevant is in danger of throwing out the baby with the bath water.
It is always difficult to distil a complex issue into one pithy question. The organisers of this debate have attempted to do so by asking, "Is the fossil record adequate?" - a query that recalls Chris Paul's continuing contributions to this subject1-3. However, another question is pressed by critics of the use of stratigraphic data in phylogenetic analysis: "Is the fossil record relevant?" Addressing adequacy implies acceptance of some degree of relevance. Only then is it meaningful to ask whether data have a signal-to-noise ratio that lets their meaning be discerned. If instead, doubts are raised regarding relevance, adequacy becomes moot.
Adequacy involves matters of practicality, details of implementation, and questions of appropriate scale and resolution - hard questions that will take some time to address in their entirety. Relevance involves matters of principle, less subject to such mundane concerns. Like armies picking the battles they would prefer to wage, supporters of the use of temporal data in phylogenetic analysis tend to focus on adequacy, while detractors focus on relevance. Small wonder that consensus eludes us.
I speak from the perspective of stratocladistics, one of the methods proposed for linking stratigraphic and character data in phylogenetic inference4-6. I am thus a "supporter", but to help break the pattern of disengagement, I choose to grapple on the field of relevance.
Slippery reference frames
In his first-round contribution to this debate, Siddall challenges use of stratigraphic data, asking, "Is it the order of appearance or is it the amount of missing time that is important?" - as if Huelsenbeck and Rannala7, and possibly the rest of us, were unclear on the difference. Of course there is a difference between these two types of data, but at this point in their paper, Huelsenbeck and Rannala were referring to a range of approaches, some of which use one, and some of which use the other. The two are never used interchangeably. Siddall's sally may confuse the incautious reader, but the rest of us are not distracted.
Less benign is an unannounced default to the reference frame of a detractor, while ostensibly testing the methodology of supporters. This lies at the heart of the old claim that, because apomorphy distributions do not favour ancestor-descendant hypotheses over sister-group hypotheses, ancestors cannot be recognized. Several forms of the argument have been circulated8-10, but they ignore or reject the role of stratigraphic (and biogeographic) data in evaluating hypotheses of ancestry11. Although character data and cladistic arguments can be used to sort among alternative cladograms, more is required to recognize ancestors. Cladists may decide they cannot recognize ancestors using their own methods, but turning this into a prohibition against such a result from any method is like complaining that because we cannot fly by flapping our arms, all birds should drop from the sky.
Are data indentured servants or free agents?
Although less extreme than prohibiting results outright, cladists also prohibit use of certain data for certain results. Notably, many of them bridle at letting temporal data participate in choosing among cladograms. Yet argument is the bridge between data and interpretation. Declaring classes of data irrelevant to a given problem presumes that every argument within which those data might be used is invalid. The history of science is replete with examples of data once judged irrelevant to a given problem that were later recognized to shed new light, once incorporated in a new argument.
Norell and Novacek12 argued that stratigraphic data cannot be used in phylogenetic inference because temporal position does not incorporate a hypothesis of homology. Yet homology is not the only expression of order in nature. Stratigraphy provides another. Faulting stratigraphy for its lack of a homology relation judges one argument by a criterion native to another. Homology may be critical to use of characters, but not to use of stratigraphy.
Smith13 presents a similar argument: because "there is no inherent hierarchical pattern to stratigraphic order", "stratigraphic data cannot by itself generate a phylogenetic hypothesis nor even overturn a hypothesis established on character distribution." I agree that stratigraphic order is not inherently hierarchical, but it can be used to recognize nested sets of taxa5, and its fit to competing hierarchical arrangements can be quantified4, 6. That is enough. Both stratigraphic and character data suffice to distinguish hypotheses that are more, and less, compatible with observation.
As for inherent hierarchy, character data also lack this property. A single binary character divides taxa into two groups, but only after polarization, or rooting of the unrooted network (by extrinsic information, such as outgroup data), are there nested sets5, 14. A single multi-state character (a better analogue for stratigraphic data) has similar strengths and limitations. Hierarchy is a relation that emerges through the context in which characters are placed, individually and collectively, not something inherent to characters themselves.
Interestingly, Smith (and Forey, in his contribution to this debate) acknowledges the force of stratigraphic data to select among equally parsimonious hypotheses13,15, but stops there. Yet the same arguments involving fit of observations to hypotheses apply more broadly. Once the relevance of stratigraphic data has been recognized, ignoring it whenever it conflicts with character data entails ad hoc hypotheses that must somehow be accounted for in the calculus of hypothesis choice. That is what generations of cladists have taught us16, and what stratocladistics asks us to remember5.
"Negative evidence" is not entirely negative
A perennial favourite of cladists criticising use of stratigraphic data is the issue of negative evidence. It surfaces in Siddall's version of why ancestors cannot be recognized (he sees no evidence favouring them beyond "lack of an autapomorphy"). More generally, it arises because stratigraphic data reflect in part where certain things are not found. Yet equating "evidence of absence" with "absence of evidence" is better word-play than science.
Absence of evidence, in this context, is not finding any ingroup fossils, and stratocladistics ignores such intervals in coding stratigraphic data. Absences that are coded and charged against phylogenetic hypotheses occur when members of some lineages are present, but other lineage segments that would have been expected, are not. Any independently tested (for example taphonomic) explanation for an absence suffices to neutralize it. It is not simply absence of data that is meaningful, but presence of certain data conjoined with absence of other data.
Biogeography and much of ecology depend on similar reasoning. Some lower-level sampling theory is required, but it is not (contrary to Siddall's view) that preservation occurs at random relative to environment or organismal attributes. Interpretation of such data, as always, is fallible, and yet this modus operandi is both rational and highly effective in the long run17, 18.
Discovery is our friend, not our enemy
Stratigraphic data summarize today's observations, but what about tomorrow's? In week two of this debate, Hedges warned that "ruling out the possibility of earlier fossils...is not a winning strategy," but most users of stratigraphic data would not so rule. Stratocladistics only asks, "Which hypothesis is best supported by data now in hand?" Adamant truth claims are not part of the package (contra Rieppel19). New data call for new analyses - as with characters and sequences - and if fit of current data to hypotheses is any measure of what to expect, stratigraphy is no worse off than morphology6.
The golden calf is a red herring
With zeal befitting a latter-day Moses, Siddall admonishes us to renounce the "cult" of "ancestor worship" 20, and abandon the "golden calf" of stratigraphic data, but this misses the point11. If ancestors emerge in our hypotheses, so be it; they do not drive us. Stratigraphic data have potential effects beyond their relevance to ancestors.
The point of including stratigraphic data in phylogenetic analysis is to discover hypotheses that better fit available data, including a wider range of data types. The point is to recover more explicit histories of change and clarify process5, 11. The point is to avoid the ad hoc covering assumption16 that stratigraphic data incongruent with phylogenetic hypotheses are of no account. The point is to provide an alternative to the infinite regress of outgroup-based rooting and the unsupported assumptions of ontogenetic rooting5. The point is to enjoy the fruits of a more broadly based view of the history of life.
Daniel C. Fisher
References
Museum of Paleontology and Department of Geological Sciences, University of Michigan, USA
