The Problems of Very Small n
Murray Aitkin and David Fox - University of Melbourne
Department of Mathematics and Statistics/Australian Centre for Environmetrics
Monday, October 20th
This talk describes a model-based framework for assessing response to toxins in small animal laboratory studies with very small numbers of
animals, to correctly assess the variability of estimation of the standard measures of toxicity. Such studies are the basis for establishing water-quality limits on toxic pollutants in ecotoxicology.
With very small n we cannot rely on asymptotic arguments or results, and there are no general frequentist procedures with exact sampling distributions. Bayesian methods are essential.
We give a well-known example of a small animal dose-response study and compare the available frequentist procedures with the Bayesian analysis.