A probability model for predicting the occurrence of cargo spills was developed and quantified using a Bayesian statistical approach. The results of this study have allowed definition of the probable size, cause, and location of cargo spills. The most prollable cause and location characteristics of cargo spill changes with The size of a spill. Decision on port sites and regulatory measures should not be made solely on the probability of the occurrence of spills, but should also consider the consequences of a spill.
An important aspect of selecting, planning, and designing port sites and oil transportation systems is the evaluation of the probability of cargo spills. The identification of factors influencing the probability of spills and the extent of their influence can also be utilized to assist in evaluating the effect of possible actions (operational, regula- tory, legislative) that may be taken to reduce the likelihood of spills.
The need for a probabilistic approach to the prediction of the occurrence of accidents has been recognized in many fields. Gardenier (1972) discusses previous applications of probabilistic approaches to the problem of marine accidents. While there is a significant amount of data available (e. g. Poricellietal. 1971), Gardenier points out the difficulty in implementing classical probabilistic approaches because the data is generally not adequate for the development of a probabilistic model. Despite the complexity of the task, the probabilistic approach appears to be the only theoretically sound basis by which it is possible to assess the likelihood of a spill.
The purpose of this paper is to (i) present an approach to the development of a probabilistic model for the determination of likelihood of cargo spills as related to various causes and locations, (ii) illustrate how the probabilistic model can be quantified, (iii) illustrate the type of information obtained from such a model, and (iv) comment briefly on the possible utilization of the information generated in (iii).
For those problems where there is sufficient data on the occurrence of accidents under well defined conditions, it is possible to develop a probabilistic model using a classical (frequents) statistical approach. However, for many problems there is a lack of a suitable data base which precludes the use of the frequenters approach. The determination of the likelihood of cargo spills is such a problem. Under these conditions, it is necessary to approach the probabilistic problem from a Bayesian Viewpoint. The Bayesian a-approach considers, in addition to data, the subjective opinions of knowledgeable people, in determining probabilities.
Raiffa (968) has described the Bayesian, subjective or personalistic point of view of decision analysis as:
Roughly speaking, the Bayesians or subjectivists wish to introduce intuitive judgments and feelings directly into the formal analysis of a decision problem. The non Bayesian, or objectivists feel that subjective aspects are best left out of the formal analysis and should be used only, if at all, to bridge the gap between the real world and the objective results one obtains using a formal model.
