Assessing impacts of dredge spoil disposal using equivalence tests: implications of a precautionary (proof of safety) approach

Abstract

Equivalence tests evaluate whether a treatment effect lies within or outside a predetermined equivalence interval. The equivalence interval might be determined in relation to a precleanup value, reference values, or unimpacted controls. Such tests are little known in ecology, but offer advantages over the common tests of point-null hypotheses. They come in 2 forms. The first, testing the equivalence hypothesis, constitutes a proof of hazard approach by postulating that a difference lies within the interval. The second, testing the inequivalence hypothesis, also known as bioequivalence testing, constitutes a proof of safety approach by postulating that a difference lies beyond the interval. The proof of safety option provides a formal mechanism for the implementation of the precautionary approach. We demonstrate the usage of these tests for a subtidal rocky reef dataset evaluating the impact of dredge spoil disposal at New Plymouth, New Zealand. Sampling of conspicuous subtidal organisms was done once before, and twice after spoil disposal. Abundances of subtidal organisms were compared between 6 sites in the predicted path of the dredge spoil and 6 distant control sites. Comparison of mean numbers of species and of individuals on and off the dumpground before and after disposal demonstrated that greater effort was required for proof of safety, though there was no proof of hazard. The most important conclusion was that much greater sampling effort than is common in ecological study was required to demonstrate safety. Because equivalence tests are readily calculated, test a realistic hypothesis, and provide an outcome that is directly interpretable in terms of a biological endpoint, we suggest that they should be more widely adopted.

Fish−habitat associations in New Zealand: geographical contrasts

“…No formal statistical tests of data from the habitat manipulations were undertaken because p-values from tests of point null hypotheses do not summarize evidence (Carver 1993, Cole & McBride 2004, Gerrodette 2011; the data are plotted as mean ± SE, and trends are discussed.…”

Quantitative characterisation of reef fish diversity among nearshore habitats in a northeastern New Zealand marine reserve

“…the relative abundance of the discriminating species best separating the a priori defined habitats were tested for statistical significance using standard univariate techniques, where statistical significance can be an artefact of sampling size (Cole & McBride 2004). However, in this study the differences detected between the five main habitats were generally owing to large changes in abundance, for example U. lineatus numbers on sponge flats, or by the absence/near absence of fish in some habitats, for example the lack of species in the sand habitat.…”

A Statistical Test to Show Negligible Trend

“…One primary motivation was to compare properties of generic and name-brand drugs (Wellek 2003:6). Equivalence tests are relatively unknown in ecological and environmental applications, although they have been applied to assess remediation success (McDonald and Erickson 1994), the assumption of equal detectability (MacKenzie and Kendall 2002), and the lack of environmental impact McDonald 1995, Cole andMcBride 2004).…”

“…It is not appropriate for identifying the absence of an important trend. Failure to reject the null hypothesis of no trend does not imply that the null hypothesis is true (Anderson and Hauck 1983, Millard 1987, Dixon 1998, Johnson 1999, Parkhurst 2001, Cole and McBride 2004. A nonsignificant result may be due to a small sample size, large random fluctuations in abundance, a poor choice of test, a trend that is close to zero in a practical sense, or the true absence of trend.…”