Empirical test of sampling bias
Unlike conventional lower-bound richness extrapolators such as Chao 1
(Chao, 1984), the species richness estimators proposed here (eqns. 10
and 16) are free of substantial sample size bias. This fact is
demonstrated with a subsample-and-extrapolate protocol. First, richness
estimates are obtained for each of the species inventories. Second, each
one is randomly degraded by removing half of the individuals, and the
estimates are recomputed. Consistency is assessed by computing the
median offset on a log scale between the two sets of estimates.