2003 Seattle Annual Meeting (November 2–5, 2003)

Paper No. 3
Presentation Time: 8:30 AM

DEVELOPING BIOLOGICALLY REALISTIC NULL MODELS TO EVALUATE BIAS IN DEATH ASSEMBLAGES


KIDWELL, Susan M., Geophysical Sciences, Univ of Chicago, 5734 S. Ellis Avenue, Chicago, IL 60637, skidwell@uchicago.edu

Taphonomy concerns post-mortem modification of biological patterns, and thus, whenever possible, realistic "biological patterns" should constitute the null model or expectation. Here are two mea culpa examples that standard null hypotheses -- which assume no or random variation in the biological system that feeds death assemblages – can be "too null", leading to both erroneous over-estimates and under-estimates of bias. (1) Even among groups with well-mineralized skeletons, post-mortem durability is experimentally correlated with a range of species characteristics including body size, shell mineralogy, etc. Thus, if durability determines the composition of the fossil record, we should observe a positive correlation between species’ body sizes etc and their frequency of occurrence. The PBDB Taphonomy group (2002) instead found zero correlation and concluded no strong bias. However, in living communities, body size and abundance are stereotypically negatively correlated (large-bodied taxa usually represented by few individuals; small-bodied species range from rare to numerically dominant). Thus zero correlation in the fossil record might argue for considerable bias rather than none. (2) How closely must the composition, rank-ordering of species, evenness etc of a death assemblage agree with that of the local living community for us to conclude "good" capture of biological information? The biological null model should be the average agreement of one live census to another, i.e. the picture of the community that would be generated if only live samples were available. I find that, in datasets of adult mollusks, average live-dead agreement in species’ rank-order is 0.55 but live-live agreement also averages ~0.6. Thus, although dead data are less than perfect (r < 1), their agreement with any live census is no worse than agreement among live censuses. Studies that include larvae and early juveniles actually show higher live-dead than live-live agreement(!). Biologically realistic null models for paleoecology will be hard-won via macroecologic analysis of modern systems, but are essential for evaluating taphonomic bias in death and fossil assemblages, given that simple null assumptions regarding biological input can lead to erroneous conclusions about the existence and magnitude of bias.