In recent years, the lab has been working on how to account for undescribed species when estimating extinction rates (see previous posts here, here, and here). The issue at stake is that some species may go extinct before they become known to science, and that failure to account for these statistically can lead to underestimates of extinction rates. In a new paper led by Deon, we compare our lab’s SEUX model with an earlier model by Tedesco et al. (2014). For the new paper we collaborated with Tedesco and tested the two models against his original global data sets as well as simulated data.
Reassuringly, the two models produced fairly similar estimates of the proportion of extinct species, accounting for undescribed species, when applied to the global data sets (see figure below). For example, the SEUX and Tedesco models estimated, respectively, that 12% and 11% of Australian mammals have gone extinct over the last 200 years, compared to the naïve estimate of 7% (which ignores undescribed extinctions). However, a few caveats emerged. Firstly, the SEUX model assumes that there are no extant undescribed species in the present day, and as a result of this assumption may underestimate the absolute number of extinctions (as opposed to the proportion). Secondly, both models assume that the probability of a species going extinct is independent of its probability of being described. Applications to our simulated data showed that violation of this latter assumption can lead to large biases in the extinction estimates. We also found evidence that the assumption may be violated in a few of our real-world data sets. More work is needed to investigate possible correlations between extinction and detection rates. Despite these caveats, our two models reinforce the notion that undescribed extinctions can be large and need to be accounted for in holistic assessments of human impacts on the environment.