Bayesian Logistic Modeling in the Classification Analysis of Water Quality

• Activity Number: 298
• Type: Topic Contributed
• Date/Time: Wednesday, August 11, 2004 : 8:30 AM to 10:20 AM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Huizi Zhang*+ and Eric P. Smith and Keying Ye
• Companies: Virginia Polytechnic Institute and State University and Virginia Polytechnic Institute and State University and Virginia Polytechnic Institute and State University
• Keywords: Bayesian logistic modeling ; water quality ; MCMC method

Abstract

In water quality studies, logistic regression is used to predict toxic/nontoxic conditions of water systems given relatively available chemical and/or environmental information. When data display hierarchical structure, Bayesian logistic modeling provides as a natural alternative to either a grand model or several separate models for estimating location-specific parameters while taking into account the correlations among the data. Using matched data on sediment toxicity and chemical concentrations from the National Oceanographic and Atmospheric Administration from different coastal locations in the United States, we developed a predictive model for classifying water quality. Dimensionality reduction on the original 22 metal contamination and polycyclic aromatic hydrocarbon contamination variables was performed using Principal Components Analysis. Empirical Bayesian methods were used in developing the prior distributions and Markov chain Monte Carlo (MCMC) method was used to obtain predictions. Furthermore, goodness of fit and predictive ability of both ordinary logistical models and Bayesian logistic models were compared and their respective strength/weakness discussed.