Prior elicitation for Bayesian generalised linear models with application to risk control option assessment

Highlights: • A pragmatic approach to prior elicitation for generalised linear models is proposed. • Subjective probability distributions are elicited from experts conditional on scenarios. • Models link the elicited data to important covariates and factors. • Models can be used for prediction and to coherently assimilate empirical data. • Application to risk control options for shipping suggests key covariates and interactions. - Abstract: A pragmatic approach to prior elicitation was developed to elicit the parameters and model structure for Bayesian generalised linear models. Predictive elicitation of subjective probability distributions was used to evaluate Risk Control Option (RCO) effectiveness for reducing the risk of ship collisions in Australia's Territorial Sea and Exclusive Economic Zone. The RCOs considered were pilotage, Vessel Traffic Services (VTS) and Ships' Routeing Systems (SRS). Predictive relationships with key covariates were documented. Distance from the Territorial Sea Baseline was important for all RCOs, and aggregate measures of shipping traffic patterns such as volume and the distribution of course over ground headings were related to the effectiveness of both VTS and SRS. A synergistic interaction between pilotage and VTS effectiveness was predicted. The elicitation method enabled a practical approach to eliciting subjective probability distributions while accounting for the complexity and myriad factors that contribute to challenging problems. The approach supports coherent updating given new information, and so can be used to support evidence based decision making.