Evaluating Geographically Weighted Regression Models for Environmental Chemical Risk Analysis

• Activity Number: 364
• Type: Contributed
• Date/Time: Tuesday, August 11, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Risk Analysis
• Author(s): Jenna Czarnota* and Chris Gennings and David Wheeler
• Companies: and Icahn School of Medicine at Mount Sinai and Virginia Commonwealth University
• Keywords: Environment ; GWR ; GWL ; lasso ; cancer

Abstract

In the evaluation of cancer risk related to environmental chemical exposures, the effect of many correlated chemicals on disease is often of interest. The relationship between correlated environmental chemicals and health effects may not always be constant across a study area, as exposure levels may change spatially due to various environmental factors. Geographically weighted regression (GWR) has been proposed to model spatially varying effects. However, concerns about collinearity effects, including regression coefficient sign reversal (i.e., reversal paradox), may limit the applicability of GWR for environmental chemical risk analysis. A penalized version of GWR, the geographically weighted lasso, has been proposed to remediate the collinearity effects in GWR models. The ability of GWR and GWL to correctly identify spatially varying chemical effects for a mixture of correlated chemicals within a study area was assessed through a simulation study. Results showed that GWR suffered from the reversal paradox while GWL overpenalized the effects for the chemical most strongly related to the outcome.