Measures of the Environmental Impact of a Pollutant from Aquatic Chronic Ecotoxicity Tests

The use of no-observed effect concentrations (NOECs) as summary statistics for low toxic effects has been severely criticized in the ecotoxicological literature. Concentration-response models represent an appealing alternative. They consider the whole information onthe toxic action of the pollutant that can be gained from a chronic ecotoxicity study. Point measures of the toxic action, such as the p%-effective concentration (ECp) and the p%-lethal concentration (LCp) can easily be obtained from the fitted response curve. This paper illustrates the attractiveness of the nonlinear regression approach by means of a study on the impact of the pesticide dinoseb on the survival of the micro-crustacean 'Daphnia magna'. We point out some little known pitfalls, such as the infringement of the regularity conditions that guarantee the validity of the asymptotic approximations to the distribution of the maximum likelihood estimator and classical test statistics. This occurs for instance if threshold models are used, where the threshold parameters are estimated from the data. A way to overcome this difficulty is to resort to non-parametric resampling techniques to simulate the distribution of the statistic of interest. Attention has to be paid to the definition of an appropriate sampling plan.