Assessing Variable Importance in an Environmental Observational Study

• Activity Number: 546
• Type: Invited
• Date/Time: Wednesday, August 1, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Risk Analysis
• Author(s): S. Stanley Young*+ and Jesse Q. Xia
• Companies: National Institute of Statistical Sciences and National Institute of Statistical Sciences
• Keywords: variable importance ; PM 2.5 ; geographic variation ; effect estimates ; air pollution

Abstract

In environmental observational studies, often authors do not address the relative importance of variables under consideration, choosing instead to concentrate on specific claims of significance. Yet good policy decisions require knowledge of the magnitude of relevant effects. In this paper we examine data on the relationship between air quality and mortality in the United States. The analysis uses two methods for determining variable importance, regression analysis and recursive partitioning, showing how this puts predictor variables into a context that supports better environmental policy-making. In particular, using both regression and recursive partitioning, we are able to confirm a spatial interaction with the air quality variable PM2.5, a critical variable in this application domain. We also determine the relative importance of this variable in comparison to others used in air pollution research. We show that there is no association between PM2.5 and mortality west of Chicago and that where there is an association between decreased PM2.5 and increase longevity, it is much less important than other variables such as income and smoking. Our findings point to somewhat different policy recommendations from those developed by previous researchers.