You are here:
A Bayesian network approach for causal inferences in pesticide risk assessment and management
Citation:
Carriger, J. AND M. Barron. A Bayesian network approach for causal inferences in pesticide risk assessment and management. Presented at SETAC North America 33rd Annual meeting, November 11 - 15, 2012.
Impact/Purpose:
Conference platform presentation for SETAC 2012
Description:
Pesticide risk assessment and management must balance societal benefits and ecosystem protection, based on quantified risks and the strength of the causal linkages between uses of the pesticide and socioeconomic and ecological endpoints of concern. A Bayesian network (BN) is a graphical representation of a joint probability distribution over a set of statistical variables and provide a mathematical approach for assessing causality. A BN can represent the best available knowledge of causal dependencies between variables important to risk management such as the influence of application events on contaminant exposure. Capabilities, advantages, and disadvantages of BNs to enhance assessments of causality in pesticide risk assessment management are demonstrated through a model development process. The process focuses on the incorporation of domain knowledge to represent and evaluate cause-consequence relationships between system variables and endpoints, and the integration of non-casual factors important to a risk problem such as measurement uncertainty, definitional interactions, and reconciliation of independent models. For the causal linkages, conditional and marginal probabilities in the BN represent the strength of evidence on causal associations and permit the incorporation of a wide range of statistical model output and expertise. The model building process illuminates the potential for BNs to facilitate the identification of new management options, improve risk communication, assimilate competing hypotheses, and enhance public participation in risk management tasks. Some of these features of BNs are exhibited in a case study of lampricide risks and benefits in Great Lakes region. The example BN will demonstrate predictive reasoning on the efficacy and potential risks to non-target organisms from several uses and diagnostic reasoning for updating beliefs on potential causes from observed effects.