MODELING THE EFFECTS OF TIME-AVERAGING ON RELATIVE ABUNDANCE AND EVENNESS

Measures of relative abundance and evenness in the fossil record have been viewed skeptically due to concerns about the fidelity of abundance data preserved in time-averaged assemblages. Computer modeling provides an approach for assessing the existence, directionality, and magnitude of post-mortem biases that potentially affect abundance data in the fossil record. A model was developed to compare the agreement between rank order and evenness in simulated live and time-averaged death assemblages. Simulations included four community types that varied in the evenness of their abundance distributions. From these, live assemblages were drawn at random. Live inputs were then subjected to different decay scenarios. The resulting time-averaged assemblage was comprised of the summation of all individuals still present after decay. Time-averaged assemblages were then sub-sampled, and evenness metrics calculated and compared for all sub-samples and live assemblages. Spearman rank order correlation of live and time-averaged assemblages was conducted for each of the four community types under each decay scenario. Randomly sampled pairs of live assemblages from a single community type were correlated in order to place the correlation results of live and time-averaged samples in context. Principal Components Analysis was used to determine whether the magnitude of bias in time-averaged abundance data varied among the four community types. Euclidean distance measures between community-type centroids were calculated to quantify the deviation between live and time-averaged assemblages under each decay regime.

The majority of decay scenarios resulted in varying degrees of deviation between time-averaged assemblages and the live assemblages from which they were derived. The magnitude and directionality of this bias varied between community types and under different decay regimes. Strength of Spearman rank order correlation was a function of the evenness of the distributions examined. As such, poor rank order agreement may not always result from bias associated with time-averaging processes but instead may be an accurate reflection of the evenness of the abundance distributions being compared.