The risk of accidents and spills at offshore production platforms: A statistical analysis of risk factors and the development of predictive models

This dissertation develops two models: expert and logistic regression. The models are used to predict the likelihood of accidents and spills at offshore production platforms and rank the platforms in terms of risk. Two sources of information are used to construct the models (1) databases maintained by the Minerals Management Service, (2) a survey of platform inspectors that was conducted in June 1998. This study covers ten years of data (1986–1995).; Every platform that had an accident during the ten-year period also had a spill at some point during that same period. However, not every platform that had a spill also had an accident Also, major complexes are over 12 times as likely as non-major complexes to experience either an accident or a spill over the ten-year period.; The logistic regression models routinely predict 50% of the accidents or spills that will occur in the top 15% of ranked platforms. In addition, platform complexity is the most important risk factor, inspection history is second, accident history is third and age of the platform, or experience of the operating company is fourth (out of four ranked risk factors). The models show that the relative merit of risk factors varies somewhat over time. However, there is no trend in model accuracy over time.; The expert models also routinely predict 50% of the accidents or spills in the top 15% of ranked platforms. The experts consider platform complexity as the most important risk factor, age of the platform or experience of the operating company is second, inspection history is third, and accident history is fourth (out of four ranked risk factors).; Both models (expert and logistic) are consistently good at ranking platforms, but the logistic regression model is significantly better (95% confidence level) than the expert model at predicting accidents. The logistic model is not significantly better than the expert model at predicting spills. Overall, a ranking based on an expert model risk index is much easier to calculate, and is only slightly less accurate than a ranking based on the logistic model.