Florian Hartig

← Publications

Studying speciation and extinction dynamics from phylogenies: addressing identifiability issues

Hélène Morlon, Stéphane Robin & Florian Hartig

Trends in Ecology & Evolution, 37(6), 497–506 (2022)
Cite this
@article{morlon2022studying,
  author = {Hélène Morlon and Stéphane Robin and Florian Hartig},
  title = {Studying speciation and extinction dynamics from phylogenies: addressing identifiability issues},
  journal = {Trends in Ecology \& Evolution},
  volume = {37},
  number = {6},
  pages = {497--506},
  year = {2022},
  doi = {10.1016/j.tree.2022.02.004},
}

DOI: 10.1016/j.tree.2022.02.004
Citation statistics: 91 (Google Scholar) · 82 (OpenAlex), as of 29 October 2026

Google Scholar citations per year, as of 23 October 2026 (yearly counts are fetched separately from the total)

View article (DOI)

Abstract

A lot of what we know about past speciation and extinction dynamics is based on statistically fitting birth–death processes to phylogenies of extant species. Despite their wide use, the reliability of these tools is regularly questioned. It was recently demonstrated that vast ‘congruent’ sets of alternative diversification histories cannot be distinguished (i.e., are not identifiable) using extant phylogenies alone, reanimating the debate about the limits of phylogenetic diversification analysis. Here, we summarize what we know about the identifiability of the birth–death process and how identifiability issues can be addressed. We conclude that extant phylogenies, when combined with appropriate prior hypotheses and regularization techniques, can still tell us a lot about past diversification dynamics.

What the paper shows and why it matters

(AI-generated)

Past speciation and extinction are mostly inferred by fitting birth–death models, which describe lineage splitting and extinction, to dated family trees (phylogenies) of living species. Louca and Pennell (2020) proved that if speciation and extinction rates may vary freely, infinitely many different diversification histories (called congruent) give exactly the same likelihood, so the data cannot tell them apart. They proposed switching to so-called “pulled” rates, which are identifiable but hard to interpret. The paper accepts the result but challenges the assumption behind the conclusion, namely that the aim is to recover completely unconstrained rate functions. Only under that assumption would non-identifiability mean that phylogenies cannot reveal speciation and extinction dynamics.

The paper reviews which birth–death models can be identified from phylogenies and argues that non-identifiability is a familiar problem in science, usually handled by constraining the model. It separates problems that persist with unlimited data (asymptotic) from those caused by realistic tree sizes (practical). Remedies are reparameterising with identifiable quantities (at the cost of interpretability), adding fossils, testing a few biologically motivated hypotheses set in advance, and preferring simple solutions through parsimony or regularisation (penalties or Bayesian priors that encode a preference for simplicity or prior biological knowledge). May et al. (2016) showed that a prior on the share of species surviving a mass extinction makes it possible to tell mass extinctions from shifts in the underlying speciation and extinction rates, in an otherwise unidentifiable model. This is a counter-example to the claim that regularisation cannot resolve non-identifiability. A simplified analysis of Madagascan vangas illustrates the point: the two families of congruent models examined for the best-supported “early burst” model (fast speciation at the origin of the group, then a slowdown) are either biologically implausible, with negative extinction rates, or keep its declining speciation trend.

The paper concludes that non-identifiability of the fully flexible model limits unconstrained inference, not phylogenetic diversification analysis as such. Estimates of speciation and extinction rates can remain informative when interpreted in light of the hypotheses or regularising assumptions behind them, and the paper advocates regularised models that act directly on these rates. Whether such constraints steer inference towards the true history among many congruent ones is left open. The paper suggests testing this mathematically and by simulating eco-evolutionary models.

References

  • Louca & Pennell (2020). Extant timetrees are consistent with a myriad of diversification histories. Nature 580, 502–505. doi:10.1038/s41586-020-2176-1
  • May et al. (2016). A Bayesian approach for detecting the impact of mass‐extinction events on molecular phylogenies when rates of lineage diversification may vary. Methods in Ecology and Evolution 7, 947–959. doi:10.1111/2041-210x.12563