Walter W. Piegorsch
University of South Carolina
Hierarchical Statistical Modeling in Environmental Toxicology
Hierarchical modeling is a growing area of interest in the environmental
sciences. In modern environmental toxicology, elaborate experiment designs
and complex assay systems are in place, and others are under development,
where hierarchical effects play important roles. Endpoints observed in
these systems increasingly exhibit the kind of complexity that calls for
hierarchical perspectives. Any statistical modeling and data analytic
efforts involving such data must build these hierarchical effects into their
base structure. This presentation considers a series of different
toxicological examples where a particular hierarchical model seems
pertinent. Emphasis is on development of relevant models; specific methods
for performing the actual statistical analysis and associated inferences
will be of secondary interest. The examples include (i) hierarchical
Poisson regression for clustered count data, (ii) normal and non-normal
hierarchical models for use in meta-analyses/data combination of
carcinogenic potency estimates, and (ii) a regression model for proportion
data based on hierarchical exponential assumptions that lead to dual
dose-response and (nuisance) correlation functions. In the latter example,
a question for discussion will be how to assess a dose effect due to the
environmental exposure using one or both of the two functions.
Back to Colloquium Series