Approximate Likelihood and Bayesian Methods for Combining Multiple Data Sources

• Activity Number: 236
• Type: Contributed
• Date/Time: Tuesday, August 9, 2005 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics and the Environment
• Author(s): Darryl Cooney*+ and Montserrat Fuentes
• Companies: North Carolina State University and North Carolina State University
• Keywords: Bayesian ; Hierarchical ; Likelihood

Abstract

Combining network datasets with large numerical model output is an important environmental goal. These combined datasets allow for evaluating the numerical models and creating more reliable air pollution maps that cover the United States with a more dense field. It is critical that these combined networks are as accurate as possible because they are used to help set state regulations for pollution controls and follow the EPA's Clean Air Act. Specifically, nitrate is an important aerosol to study due to previous scientific research that has led to possible links to negative health effects. A Bayesian hierarchical model presents itself as a powerful method to model the biases and variation inherent in the networks compared to the true total nitrate values. Various covariates, such as temperature and relative humidity, also are incorporated into the hierarchical model. The major focuses of the research are the methods to approximate the likelihood for the large amounts of data to reduce computation time. The approximation involves clustering the data and forming conditional likelihoods for observation points based on others.