Meeting Abstract

67.3  Saturday, Jan. 5  A conundrum of covariation: The effects of missing data on disparity analysis SMITH, AJ*; ROSARIO, MV; EITING, TP; DUMONT, ER; UMass, Amherst; UMass, Amherst; UMass, Amherst; UMass, Amherst ajsmi1@cns.umass.edu

Disparity, or morphological diversity, is an important metric of biodiversity used to analyze evolutionary trends in form over geological timescales. Although missing data are common in fossil datasets, we do not fully understand how different disparity metrics respond to increasing levels of missing data. Past research investigated this by randomly removing morphological characters from simulated taxa. However, the loss of anatomical characters is not a random process; characters in close physical proximity to one another are likely to be correlated in presence or absence. First we calculated covariation in character loss from 12 extinct taxa coded for 196 characters, then used that covariance structure to remove characters from a data-rich matrix of 49 extant taxa coded for the same characters. Starting from a maximum of all characters present, we sequentially removed characters in every taxon from the extant matrix such that the average character loss across taxa represented 0% to 75% loss. At each character loss step, we calculated morphospace range and variance (average spread and dissimilarity among taxa respectively). We then repeated this process without character covariation (i.e., randomly removing characters). With covariation, our range metrics exhibited inverse exponential declines whereby the slope changes at ~40% missing characters before declining rapidly. Our variance metrics declined linearly with confidence intervals narrowing as loss increased. Without covariation, range metrics displayed linear declines, while variance metrics exhibited exponential declines. Our results show that character covariation has important consequences for disparity metrics, and should be taken into consideration in future disparity studies.