Viewpoint: using gene-environment interactions to dissect the effects of complex mixtures.

Teasing out the health effects of constituents of complex mixtures poses formidable statistical challenges owing to the problem of multicollinearity. While statistical devices such as regression on principal components, model selection, and model averaging offer some approaches to this problem, incorporation of external information is likely to be more helpful. I explore a general hierarchical modeling framework that would allow such information as sources, genetic interactions, and toxicology to be included in the higher levels of the model.