Dose-response. Each species follows a log-logistic curve, P(C) = 1/(1 + (LC50/C)h), the fraction responding at concentration C. A test puts n organisms at each of k concentrations and counts responders (binomial sampling with a fixed random seed). The curve is fitted by maximum likelihood for a binomial GLM with logit link in x = log10C using iteratively reweighted least squares; the LC50 is −a/b and its 95 % interval comes from the delta method on the inverse information matrix.
Species sensitivity distribution. The fitted log10 LC50 values of the 15 species are modelled by a normal distribution (μ, σ). The hazardous concentration for 5 % of species is HC5 = 10μ + z0.05σ with z0.05 = −1.645; its 95 % interval comes from 1000 bootstrap resamples of the species. The fraction of species affected at concentration C is PAF = Φ((log C − μ)/σ). Dividing the HC5 by an assessment factor (here an illustrative 5) gives the protective limit.
Honest simplifications. The substance and the 15 organisms are invented, generic, and the true sensitivities are drawn from a bell curve plus small group offsets (algae more sensitive, fish less). Real assessments use measured data, check the distribution fit, handle chronic endpoints and mixtures, and choose assessment factors by guidance. The lab never depicts harm to animals; “responding” is just a count in a statistical test. Beaker, organism and bottle sprites were rendered in Blender.
Independent checks (run on load in your browser):
| running… |