- Copyright © 2015 The Paleontological Society. All rights reserved.
Morphospace occupation through time provides a view of diversification distinct from the more familiar taxonomic tabulations. However, this view is subject to the same geological biases long recognized in studies of taxonomic diversification, where techniques for correcting secular bias in sampling have become standard practice. In this study, we apply sampling standardization techniques to a morphospace investigation to test whether observed stratigraphic trends in morphospace occupation are artifacts of trends in sampling. When sampling bias is corrected by randomized subsampling, all disparity metrics show stationary patterns, or at most directional changes of small magnitude. Metrics describing the average dispersion of taxa in morphospace are less subject to sampling bias than those describing the total extent of morphospace occupied. We also investigate a measure of disparity that is insensitive to sampling intensity, introducing a geographic component of morphological disparity. By analogy to α and β components of taxonomic diversity, we suggest the notions of α and β disparity, and find that α disparity remains roughly constant through time. Our analysis also allows us to present the first taxonomic diversity curve of diatoms under shareholder quorum subsampling (SQS), showing similar results to previously published subsampling methods: a roughly twofold rise over the Cenozoic, with peak diversity around the Eocene/Oligocene boundary. Tests for methodological bias from choices in ordination method and data culling during morphospace construction indicate that our results are relatively insensitive to both factors: Cenozoic occupation of planktonic diatom morphospace is largely unchanging. We find a similarly stationary pattern when we directly analyze the morphological data, seeing no change in the prevalence of taxa with different sets of morphological characters. More broadly, our results make clear that a complete view of morphological disparity must consider sampling biases, which can be addressed with wellestablished, quantitative methods in morphospaces populated using occurrence-level data.