Statistics Seminars: Modelling species effect data for chemical hazard assessment
28 February 2011 15:15 in CM221
Chemical risk assessment is an important tool for restricting the potential ecological damage from chemical substances while still permitting industry and agriculture to use them to advantage. We will first overview a particular aspect, namely "intermediate tier hazard assessment", which reduces to the problem of estimating the environmental concentration of concern for the toxicant assessed. The standard approach is to first measure the tolerance (in terms of concentration) of a small number of species to the toxicant and consider them realizations from a normal distribution, called the SSD hereafter. The sought-after concentration is then taken to be the 5th percentile of the SSD -- a risk management decision. Whilst the risk assessment framework may appear overly simple, it is deliberately designed to be that way in order to be transparent to all stakeholders and economical to implement, focusing resources only where the potential for adverse effects is not within acceptable limits.
The small sample sizes involved in estimating the 5th percentile introduce large uncertainty, and so Bayesian methods have recently become of interest to the risk assessment arena. We make an empirical assessment of a proposal by the United States Environmental Protection Agency which uses estimation of unmeasured tolerances using "historical" relationships to augment the data available for estimation of the SSD 5th percentile. Finally, we consider a simple extension of the SSD model which includes "species effects" with the same intention as the EPA approach but which we hope will be more coherent.
Work done in collaboration with: Peter Craig (Durham University), Andy Hart (The Food and Environment Research Agency), Stuart Marshall (Unilever), Oliver Price (Unilever), Mathijs Smit (Statoil ASA), Robert Luttik (RIVM), Peter Chapman (formerly Unilever).
Contact firstname.lastname@example.org for more information