Methods for Deriving Pesticide Aquatic Life Criteria - Reviews of Environmental Contamination and Toxicology - page 79

The first step in the SSD methodology is to plot data in a cumulative frequency
distribution. One approach for doing this is to assume that those data are a random
sample of all species, and that if all species were sampled they could be described
by one distribution. The USEPA (1985) assumes a log-triangular distribution, while
the Netherlands methodology utilizes a log-normal distribution (Aldenberg and
Jaworska 2000). The USEPA Office of Pesticide Programs utilizes a log-normal
regression method for ecological risk assessment (Fisher and Burton 2003). Any
SSD method that utilizes all available data may be used either to determine the
percentage of species that could potentially be harmed by an expected environmental
concentration, or, conversely, to determine an environmental concentration that will
protect some percentage of species. The OECD methodology (1995) offers a choice
of the log-normal distribution method of Wagner and Løkke (1991), the log-logistic
distribution method of Aldenberg and Slob (1993), or the triangular distribution of
USEPA (1985), depending upon which distribution best fits the available data.
Figure 2 illustrates the character of the log-normal, log-logistic, and log-triangular
distributions.
The OECD method (1995) cites two advantages of the USEPA method (1985).
First, because it uses a subset of the lowest available values, it is not affected by
deviations of the highest values from the assumed distribution. Moreover, data
reported as “greater than” may be used, which is not possible with other methods
(Erickson and Stephan 1988). Okkerman et al. (1991) criticize the USEPA's selection
of the triangular distribution because it implies a toxicity threshold and possibility
of a 100% protection level, and it only uses four (usually the lowest four) data
points to calculate a criterion. The authors of the Australia/New Zealand guidelines

0.5

0.4

Log-normal

0.3

Log-logistic

Log-triangular

0.2

0.1

0

− 4 − 3 − 2 − 1 0 1 2 3 4

Standard Deviations

Fig. 2 Comparison of log-normal, log-logistic, and log-triangular distributions