THE EFFECTS OF SAMPLING ON EXTINCTION SELECTIVITY IN DEEP TIME

The fossil record is an unparalleled source of information for studying the dynamics of diversity of life and its determinants. One of the most pressing questions in paleobiology is understanding why certain taxonomic groups go extinct under environmental perturbations. Assessing the selectivity of marine extinction over time has been made possible by compilations of fossil occurrences in time and space in databases such as the Paleobiology Database. A common method to assess extinction selectivity is logistic regression linking particular ecologic or taxonomic attributes with extinction probability. However, the potential impact of sampling bias on this approach has not previously been explored. Taking observed last appearances at face value ignores potential sampling biases that distort observed and true times of extinctions. The “true” probability of observing an extinction event depends on sampling and fossil preservation potential. We investigated, using phylogenetic simulation-based methods, how sampling biases in the fossil record may influence studies that look at the relationship between traits and extinction risk. We generated three categories of complete trees, with traits evolving under a Brownian motion model. In our control scenario, extinctions were completely random; in the second, taxa with “larger” trait values preferentially went extinct; and in the third, “smaller” trait values preferentially became extinct. We generated fossil occurrences, where we control for preservation and sampling potential. We then assessed diversity dynamics using subsampling and capture-mark-recapture (CMR) methods to analyze our recovered fossil occurrences. Diversity dynamics were compared to our predetermined preservation and sampling parameters. Preliminary results from our control scenario indicate a mismatch between pre-allocated sampling parameters and recovered rates. Our results show the necessity of developing a robust framework for assessing extinction selectivity that better characterizes sampling biases.