EXAMINING POTENTIAL SOURCES OF BIAS IN PALEOCOMMUNITY ANALYSES, FROM DATA COLLECTION TO PATTERN INFERENCE

As fossils represent the only direct evidence of life on this planet across geological time scales, we rely on paleontological data to reconstruct major ecological changes in Earth’s history. However, inferences of this magnitude require large paleoecological databases that have been carefully constructed, vetted, and analyzed. It is therefore of paramount importance to ensure that these databases: 1) are built as objectively as possible; 2) tailored to appropriately answer the biological questions of interest; 3) account for potential sources of systematic error, and present those biases honestly. Here we present a series of re-analyses of two published and publicly available paleocommunity datasets to investigate how common aspects of taphonomic bias, collection bias and analytical choice affect conclusions regarding spatiotemporal diversity patterns. These datasets, from the middle Cambrian Burgess Shale and the Late Cretaceous Belly River Group, each incorporate tens of thousands of fossil occurrences with corresponding stratigraphic data, allowing for robust analyses of spatiotemporal changes in diversity. Broadly, both datasets demonstrate that in order to make meaningful comparisons between the collected fossils, other similar localities, and the hypothetical original community composition, significant taphonomic bias needs to be considered. In the Burgess Shale, more than three quarters of all species are unlikely to preserve under typical conditions. In the Belly River Group, the taphonomic and morphological features of each taxonomic group must be considered before meaningful ecological comparisons can be made with either coeval or modern systems. Both datasets also demonstrate that the interpretation of patterns in the data can be significantly altered based on seemingly small analytical choices, such as the chosen diversity metric or sampling regime. These results highlight the need to include experts on focal datasets when attempting meta-analyses, as simply incorporating published data without a full understanding of the systematic choices made when building them can lead to misinterpretations. They also highlight the value of comparing and contrasting multiple types of complementary analyses to arrive at quantitatively robust and philosophically sound conclusions.