Technical Comments

Phylogenies, Temporal Data, and Negative Evidence

See allHide authors and affiliations

Science  20 Aug 1999:
Vol. 285, Issue 5431, pp. 1179
DOI: 10.1126/science.285.5431.1179a

In their report (1), D. Fox et al.attempt to demonstrate that the use of temporal data contributes directly in the reconstruction of phylogenies. They tested the proficiency of stratocladistics, a method of constructing phylogenies in which temporal data is included, by modeling evolutionary change in hypothetical taxa and then attempting to reconstruct this known phylogeny. They then compared the result with that obtained when they used the more traditional method of cladistics. But, however provocative the modeling and statistics may appear, the underlying science in this report is incorrect, because the data are unsuitable and the analysis is inappropriate.

The use of temporal data inherently invokes the use of negative evidence. In stratocladistics (1), the lack of a fossil taxon within some stratigraphic horizon is used as evidence that that taxon did not occur during that time interval. Thus, first and last occurrence become pivotal characters. However, the absence of a fossil trace of a taxon within a stratigraphic horizon does not preclude the possibility that the taxon did in fact exist during that time interval (with an arguable exception involving certain plankton fossil deposits). The chances that organisms are preserved by fossilization and then documented by paleontologists are stochastic events governed by random and nonrandom processes. The tremendous gaps in the fossil records of both monotremes and coelacanths illustrate how profoundly inadequate the results can be. Fox et al. state that, in stratocladistics, there can be discordance in a phylogeny between the parsimony analysis of the character data and the stratigraphic record. A phylogeny's character debt (which is based on observed distribution of characters states among taxa; that is, the positive evidence) and its stratigraphic debt (which is the stratigraphic lack of a fossil record for a taxon; that is, the negative evidence), are incomparable lines of evidence and therefore should not be combined in an analysis.

The analyses in the report are problematic. Fox et al.begin with the set of most parsimonious trees generated from the character data, then add their stratigraphic characters to these trees. They then swap branches and force terminals to be ancestors, and search manually for the hypothesis with the least “stratigraphic character debt.” The problems with this approach are twofold: characters cannot be added post hoc into a cladistic analysis without requiring total reanalysis, and searching for most-parsimonious trees by manually swapping branches is only a crude approximation of a heuristic search. Fox et al. state that their searches are not exhaustive. Consequently, they cannot justify their sweeping conclusions regarding the putative superiority of stratocladistics over cladistics.

The incorporation of fossil taxa without stratigraphy in a phylogenetic analysis can significantly alter tree topology (2). Fossil taxa contribute to richer taxon sampling and may “fill in” the gaps along long diverging branches. The temporal component of fossils can help to elucidate changes in character state, which assists in the confirmation of homology. The general correlation between cladistic branching patterns and the first occurrence of taxa does not provide support for the use of stratocladistics, but merely implies that there are insignificant gaps in the fossil record for the taxon under study (3), minimizing the need to construct ghost lineages.


Response: Heyning and Thacker state that the use of “negative evidence” in stratocladistics renders that method invalid. Data about the order of occurrence of organisms within a temporal sequence are distributional. As with any distributional data, the signal used to evaluate competing hypotheses is a complex mix of “positive” and “negative” observations; that is, where things are, and where they are not (1). Distributional data are used in many fields where practitioners neither presume them to be infallible nor deny the possibility for new observations to change current patterns; we claim no more for stratigraphic data. Intensity of sampling is important, but whether distributional data are compelling in any given case is a different matter from Heyning and Thacker's implication that they are inadmissable in principle.

We agree that “the absence of a fossil trace of a taxon within a stratigraphic horizon does not preclude the possibility that the taxon did in fact exist during that time interval.” Heyning and Thacker describe our position more accurately in a previous sentence, in which they state, “the lack of a fossil taxon within some stratigraphic horizon is used as evidence [emphasis added] that that taxon did not occur during that time interval.” Our study's focus on incomplete fossil records addressed precisely this point. With less than 100% completeness, some of our recorded absences were “false,” in that a lineage had been present, but incomplete preservation excluded it from our data set. Nevertheless, even with levels of completeness below those documented for many groups of organisms (2), the complex pattern of presences and absences in the available data often helped achieve a better reconstruction of the modeled phylogeny than was possible through cladistic analysis of the morphologic data alone. Morphologic and stratigraphic data are different, and within cladistics they may appear “incomparable,” but stratocladistics provides its own analytical framework within which both can be evaluated for their fit to phylogenetic hypotheses.

With respect to stratocladistics' criterion of fit, morphologic and stratigraphic data are comparable. On an operational level, stratocladistics does not “force terminals to be ancestors”; it only allows them to be so if the data warrant this interpretation. Further, we did not search “for the hypothesis with the least ‘stratigraphic character debt’”; instead, we sought hypotheses that minimized overall (that is, morphologic + stratigraphic) debt. We agree that “characters cannot be added post hoc into a cladistic analysis without requiring total reanalysis,” but when we add stratigraphic data to a problem, we are operating outside the framework of cladistics. Our manual branch-swapping, “crude” or not, demonstrates our agreement that new data call for new analysis. During such reanalysis, stratocladistics applies its own criterion for choosing a hypothesis, not that of cladistics, and in our study, this method frequently led to the actual phylogeny or a hypothesis closer to it than any of the hypotheses selected by cladistics. This result, focusing on the best that each method has achieved [CFImax(3)], is not tarnished by any lack of exhaustiveness of stratocladistic searches. We would welcome a more efficient search algorithm, but the result of a search ultimately reflects the criteria being applied. As for the “superiority of stratocladistics over cladistics,” our results were explicitly conditioned on the models of evolution and preservation that we studied. This field of investigation is open: if anyone were to discover model conditions under which the outcome were different, there could follow a discussion about which conditions better reflected reality.


Navigate This Article