Methods for Deriving Pesticide Aquatic Life Criteria - Reviews of Environmental Contamination and Toxicology - page 86

When enough data are available, SSD methodologies provide a reasonable way
to estimate ecosystem-level effects based on single-species data. Several criticisms
have been directed at SSDs and their use in setting regulatory limits. Most of the
criticism stems from the underlying assumptions in SSD methodologies, some of
which are discussed in the OECD (1995) and Australia/New Zealand guidelines
(ANZECC and ARMCANZ 2000). The most general of these assumptions are
discussed here. First, is the assumption that the ecosystem is protected if 95% of
species in the ecosystem are protected. This assumption may be particularly prob-
lematic if so-called keystone species are among the most sensitive to a toxicant.
Any criterion derived by any method must be evaluated in the context of species
considered to be important for ecological, commercial or recreational reasons. If
data indicate that important species will be harmed by the derived criterion, then an
adjustment of the criterion is in order. The USEPA methodology (1985) stipulates
that, if a species mean acute (chronic) value (SMAC or SMCV, respectively) of a
commercially or recreationally important species is lower than the calculated FAV
(FCV), then the SMAC (SMCV) is used as the FAV (FCV).
Another assumption discussed by the OECD (1995) and the Australia/New
Zealand (ANZECC and ARMCANZ 2000) guidances is that the distribution of
toxicity data is symmetrical. If insensitive species give very high toxicity values
and sensitive species give very low values, then this bimodal distribution will prob-
ably have a very large standard deviation. The large standard deviation, resulting
from such data, will portend a very low estimate of the 5th percentile level. The
best way to handle bimodal distributions is to scrutinize data and determine if
outliers should be removed from the set (ANZECC and ARMCANZ 2000); alter-
natively, data can be split into two distributions and the more sensitive data used
to derive criteria (RIVM 2001; ECB 2003). A third assumption (ANZECC and
ARMCANZ 2000; OECD 1995) is that toxicity data represent independent,
random samples from the distribution, which is generally not true; data are normally
collected from species that are easy to handle in the laboratory, or were selected
for their sensitivity to a particular toxicant. For some species, there are many data
and for some there are none.
Posthuma et al. (2002a, b) point out a number of advantages, disadvantages, and
ongoing issues in the use of SSD methods. Advantages include (1) SSD methods
are conceptually more transparent and scientifically more defensible than AF
methods; (2) they are widely accepted by regulators and risk assessors; (3) they are
understandable; (4) they allow risk managers to choose appropriate percentile lev-
els and confidence levels; (5) they use commonly available ecotoxicity data; (6) they
rely on relatively simple statistical methods; (7) they provide a way to assess mix-
tures; (8) they can be used to determine effects on species or on communities; and
(9) they provide clear graphical summaries of assessment results. Disadvantages
include (1) SSD methods have not been proven to be more (or less) reliable than
alternatives; (2) they require relatively large data sets; (3) they rely on statistics with
no mechanistic components; (4) distributional assumptions may not be true; (5)
multimodal species distributions are problematic; (6) criteria based on lower confi-
dence limits are overprotective; (7) test species are not randomly sampled; (8) there