|
Abstract:
|
In environmental epidemiology studies, both exposures and outcomes are often inherently high-dimensional and true effects likely small, with limited a priori information on relationships potentially linking exposure to outcome. We discuss these issues using an algorithm that (i) estimates population causal parameters, (ii) uses marginal inference on estimates, accounting for high-dimensional confounding, and (iii) allows for multiple testing. In short, we identify the relative importance of effects while simultaneously annotating those likely to be "real." Interaction between variables and longitudinal effects add complexity. The ideas are illustrated with data from the CHAMACOS study, a longitudinal birth cohort examining the effect of pesticide exposure on birth outcomes and infant neurological development. This is joint work with Jessica Young, Alan Hubbard and Brenda Eskenazi.
|
