A Modified Bayesian Network model of a scenario study for risk management related to shark-human interactions

In our recently published article “A scenario study of the acceptability to ocean users of more and less invasive management after shark-human interactions”, Bayesian Network (BN) was chosen as the principal analysis tool for identifying patterns from a survey study of 1769 valid responses. A modified BN model (26 nodes, 54 links) was proposed in this poster to improve the prediction performance of those four risk management policy/strategy variables in the original BN model (26 nodes, 44 links). At the cost of substantive increase in number of model parameters (7200 conditional probabilities estimated for the original model versus 19872 for the modified model), the modification was justified by the consistently lower error rate results in predicting the outcomes of the four focus strategy variables. The modified BN model has showed a higher sensitivity in quantifying the impacts of predictor variables (e.g., Scenarios and the characteristics of respondents) on the response variables (preference levels for four types of risk management strategies). The overall patterns of the much higher proportions of support for the Education and Noninvasive strategies versus the much lower proportions of support for the Invasive and Population Reduction strategies remained the same for both BN models but to a slightly lesser extent in the modified model. Since the modified model counted for both the direct and indirect effects of the predictor variables on the response variables, the changes of the proportions of support for each of the four possible risk management strategies were more sensitive to the changes in the predictor variable such as Scenario, Age, Gender, or Frequency of Beach Visit. Hence the modified model was deemed to be an improved version of the original BN model.