Ecological Risk Assessment of Diffuse and Local Soil Contamination Using Species Sensitivity Distributions - Dealing with Contaminated Sites: From Theory Towards Practical Application - page 652

CI-Left HC5 CI-Right

Environmental Concentration

Fig. 14.8 Presentation of statistical uncertainties of SSD results. The outer curves represent the
95% Confidence Intervals (CI) of the fitted SSD (the central curve). They are used to estimate
confidence limits of the HC p values (shown here: HC5). The dots are the original data and the fine
line is an empirical step function

14.9.3 Interpreting Statistical Confidence Intervals

If one fits various models to the same data set, multiple models often show a similar
degree of fit to the data. That is: all S-shaped mathematical functions fit well on a
(log) concentration scale, or none of them fit. What is the meaning of this, given the
possibility to present and interpret point estimates of HCp or confidence intervals?
When various models show a similar degree of fit (and overlapping confidence
intervals) the SSD-output (in the format of estimated values of e.g. an “acceptable
concentration” like the HC5) can nonetheless be numerically dissimilar, especially
when the required output is positioned in the tails of the distribution, where small
changes in estimated response are associated with relatively large differences in soil
concentration and where confidence bounds are wider.
This apparent contradiction (similar model fit, different outcomes) has caused
various intense debates at the interface of science and policy. Statistically indistin-
guishable differences in the numerical value of the HC5 (thus being scientifically
without serious implications) may practically mean the difference between permit-
ting and banning a pesticide or other newly developed chemicals. This problem is
illustrated in Fig. 14.9 . If one considers this graph in the context of the uncertainty
intervals that can be estimated (see Fig. 14.8 ), it is clear that the three estimated risk
limits (HC5) in Fig. 14.9 are scientifically estimates of the same “true” HC5 (that
is: all three numerical HC5-estimates are positioned within the confidence intervals
of all three models shown). For societal use, however, the three possible numerical