at impacted and putatively non-impacted areas, or at one or a series of sites where before-and-after impact observations are available. Ideally, before-after and control-impact (BACI) observations can be made (Stewart-Oaten et al., 1992; Wiens and Parker, 1995; Peterson et al., 2001). The inherent assumption is that the variability of the ecosystem is sufficiently controlled (in the experimental sense) by these designs, which may or may not be correct. Controlling for impact by comparison of sites that have been affected and not affected allows for a variety of potentially important non-oiling variables to influence the system—such as differences in water temperature, salinity, or substrate type. For example, see Bowman (1978) for a case where high temperatures were documented to have a differential effect on intertidal invertebrate mortality, that might have otherwise been attributed to oil or dispersant toxicity. Usually an attempt is made to find study sites that are as similar as possible in factors suspected to be important. When effects are determined based on comparisons of before-impact conditions and after-impact conditions, it is possible that the ecosystem has changed in ways unknown to the observer. The chances of making errors can be lessened when: (1) multiple sites are used in each of the impacted and non-impacted sites, (2) multiple times are used in the time series, or, even better, (3) when both multiple sites and multiple times are available. Nevertheless, unreported factors not related to oil can interfere with ecosystem processes in ways that disguise the effects of pollution. Of course, with each additional kind of impact that is measured, the chance of making an error (Type I) rises.
At the same time, the mosaic of complex interactions and the resultant changes in ecosystems makes it possible to miss an impact that occurs (Type II error). For example, if an oil spill occurs when the pelagic larval stages of a fish species are developing near the sea surface, many or most of these larvae may die. If these larvae were to be the foundation of what would otherwise have been a strong year class for that fish species and whose population is maintained by infrequent large year classes, then the impact could be much larger than otherwise supposed. That would be a disproportional effect on a process that is temporally constrained. There are also examples of potential impacts on processes that are disproportionate because they are spatially constrained. For instance, a small spill around a seabird habitat where a large proportion of a population is gathered for breeding could have a disproportionately large impact. A good example of this occurred when an estimated 30,000 oiled seabirds washed up along the coasts of the Skagerrak following a small release of oil from one or two ships (Mead and Baillie, 1981). At the other extreme, the wreck of the Amoco Cadiz off the coast of Brittany, France, resulted in the release of 230,000 tonnes of crude oil into coastal waters and the death of less than 5,000 birds (Hope-Jones et al., 1978). These examples help illustrate that the volume of oil is only one factor determining mortality of birds and the