Comparing diversity under collection bias

Two assemblages, A and B, are recorded from synthetic populations under different collection biases. Pick a bias scenario and a standardization strategy, and see which one still recovers the true richness ratio.

The two populations share one abundance shape, so a difference in diversity is real, not an artefact of structure. Sample sizes capture population-size and productivity (“rich-get-richer”) biases — some assemblages simply attract more records. The rarity-bias slider captures Stromer’s Riddle: records are drawn with weight ∝ count^(1−γ), so larger γ over-records rare, novel variants. Coverage-based standardization corrects all three; size- and rarefaction-based comparisons do not.

Richness accumulation and sample coverage use the Chao / iNEXT estimators — the same estimate_coverage and rarefaction_extrapolation as copia — and the standardization follows the reference code (coverage at Ĉ_max = the smaller of the two sample coverages; rarefaction at n_min), averaged over many resamples. For clarity the two populations share one log-normal abundance shape, where the paper’s full study draws populations from a neutral (Wright–Fisher) model. Its accuracy is governed primarily by the effective standardized sample size — the number of records retained at the comparison point — not by the coverage level reached (n ≥ 30 sufficed in the paper’s simulations).