variables (Stewart 1976, Stewart and Kennedy 1978). Devanney and Stewart (1974) 



estimated the number of spills using the negative binomial distribution. Since 



the exposure variable (volume of oil) associated with an OCS sale is less than 

 the historical exposure, the Poisson distribution serves as an excellent 



approximation to the negative binomial distribution (Smith et al. 1982). The 

 Poisson distribution is defined by its single parameter, the expected number of 

 spills . 



The oil spill data bases used in the DOI model were initially assembled by 

 Devanney and Stewart (1974) and Stewart (1975, 1976). OCS production records 

 are maintained by USGS, as are records of platform and pipeline spills. Oil 

 spill incidences are currently under review by the Futures Group (1982) under 

 contract with BLM. 



There have been 10 OCS platform spills greater than or equal to 1,000 

 barrels since 1964; eight of these occurred prior to 1974. A critical analysis 

 of these platform spills and production since 1964 has quantified a decrease in 

 platform spill rate (Nakassis 1982). Pipeline and tanker spill rates are also 

 under review by personnel in BLM and USGS. 



Spill incidence rates used in the DOI model receive considerable criticism 

 from numerous groups. Generally the criticisms suggest that the spill rates 

 are either too high or too low; that the rates are not applicable to some OCS 

 regions (recall that all of America's OCS production to date has been from the 

 western and central Gulf of Mexico and southern California); and that there 

 must be better exposure variables. Those of us responsible for the oil spill 

 model have adopted the following policy: 



1) the past OCS experience is the best available quantitative basis from 

 which to predict future events; 



2) the number of spill incidences is of record and will be made available 

 to those who desire testing their hypotheses regarding spill incidence; 



3) if alternative exposure variables can be demonstrated to be more 

 precise and of more utility to decisionmakers they will be incorporated; 



4) it is reasonable to use a length of past record comparable with the 

 length of forecast (this, incidently, is necessary if we are to examine 

 the record for changing rates); 



5) as for regional applicability, the spill rates used in the model have 

 been examined against the records of experience in Cook Inlet (pro- 

 duction from State leases) and in Prohdue Bay (onshore production) - 

 the experiences in these two locations are not different, in a sta- 

 tistically significant sense, than our experiences in the Gulf of 

 Mexico ; and 



6) lastly, we are able to define the circumstances (required production 

 without spills) needed to achieve a difference from our past 

 experience. 



95 



