The frequency of large-volume oil spills is considerably greater than is consistent with prediction based upon traditional methods. The reason for this phenomenon is that standard probability distributions of magnitude of spills do not have the flexibility to admit of very large coefficients of variation, especially for distributions which are highly skewed to the right. Hence, distributions which have large means relative to the median and which have long thick tails are prerequisites for an appropriate treatment of the problem. The class of stable laws provides a convenient method for investigating the empirical oil spill experience: several large spills dominate the total volume of spillage in virtually all accounting periods; e.g., quarterly. Our methodology involves a statistical assessment of “accident-proneness component;” if one exists, the data is further examined to identify insofar as is possible the genesis of the component (s); if none exists, we assess the frequency and severity of discharge for various geographic areas. A new approach has been utilized to fit these long, thick-tailed probability distribution to a U.S. Coast Guard data file on oil spills, the pollution incident reporting system (PIRS), with considerable success. We pay particular attention to the fitted upper tail vis-a-vis the actual upper tail. The agreement, where our methodology is deemed applicable, is very good We also indicate improvements to methodology and applications.
*Major, U.S. Air Force.
†Currently Visiting Professor, School of Urban and Public Affairs, Carnegie-Mellon University, Pittsburgh, Pa. Research sponsored in part by the U.S. Coast Guard. Department of Transportation, contract no. DOT-C6-23876-A.