Semiparametric Estimation of the Size of Oil Tanker Spills

This paper estimates the determinants of the size of oil tanker spills without dis- tributional assumptions on the error terms. We employ semiparametric estimation techniques to estimate the parameters of a sample selection model and compare them to the estimates from a sample selection model with normal errors. We nd that al- though parameter estimates are sensitive to the assumption of normality and to the semiparametric technique used. Major ndings that are qualitatively supported by all methods are: groundings and collisions result in larger spills if there is a spill, but the likelihood that there will be a spill due to a grounding or collision is very low; tanker size has only a marginal e ect on the probability of a spill and a dubious e ect on spill size; US ag tankers and new tankers have a lower probability of causing spills, compared to foreign ag and old tankers, respectively.