A statistically meaningful approach to the setting of environmental standards
Most countries aim to regulate and protect the state of our environment under a system
of environmental standards to control the level of certain pollutants present in various
media of concern. Many such standards are often set without due consideration of
uncertainty and variation and based on poorly defined principles. A sound statisticallybased
approach to setting environmental standards can be based on the statistically
verifiable ideal standard (SVIS) of Barnett and O'Hagan (1997).
The SVIS is developed and practical implications of its use considered in terms of
applications to pollution situations in air, water and soil, working in co-operation with
relevant bodies. Developments include a non-parametric binomial approach and
quantile testing for several simple parent pollutant distributions; properties of these
approaches are examined in detail. A best linear unbiased quantile estimator (BLUQE)
is examined, and 5% and 1 % critical values for the 0.95 and 0.99 BLUQE tabulated for
use in an approximate significance testing procedure. This work is extended to a
BLUQE for ranked set sampling, demonstrating impressive efficiency gains.
Assessment of the SVIS using composite sample data is also investigated, with major
improvements in test perrormance over the use of the commonly accepted 'divide-by-n'
rule for critical value calculation.
Following Barnett and O'Hagan (1997), the problem of setting directly equivalent
compatible standards at different stages of the pollutant cause-effect chain is
investigated. A statistically verifiable ideal guard point standard with two levels is also
developed to avoid benefit of the doubt in testing procedures for standards, and its use
demonstrated for both normal and gamma parent pollutant distributions. A reference
point standard is proposed for a spatially dependent pollutant variable, with a krigingbased
testing procedure. Finally, a 'hotspot' identification procedure is also developed,
using outlier methods and composite sampling. The work concludes with suggestions
for further related research.