Ground water monitoring presents interesting statistical challenges, including controlling the risk of entering compliance monitoring, incorporating all modes of inherent variability into the statistical model on which tests are based, and taming the detection limit problem, all while maintaining demonstrable sensitivity to real contamination.
Some of these challenges exceed textbook statistics considerably, even when considered alone, and good solutions are scarce. When these challenges are combined, the task of developing good statistical procedures or good regulations can be formidable.
This article presents a number of realities of ground water monitoring that should be considered when developing statistical procedures. Recommendations made for addressing these realities include the following: (1) the false positive rate should be controlled on a facility-wide basis, rather than per well or per parameter as required in the proposed regulation (40 CFR §264); (2) multiple comparisons with control procedures are preferable to analysis of variance (ANOVA) for controlling the overall false positive rate; (3) retests can be made an explicit part of the statistical procedure in order to increase power and decrease sensitivity to distribution shape assumptions; (4) commonly used simple methods of handling below detection limit data with parametric tests, including Cohen's procedure as implemented in the U.S. EPA's Technical Enforcement Guidance Document (TEGD), should probably be avoided; (5) the statistical properties of practical quantitation limits for non-naturally occurring compounds should be studied carefully; and (6) so long as the facility-wide false positive rate is controlled, better sensitivity to real contamination is obtained by monitoring fewer well-chosen parameters at a smaller number of well-chosen locations.
An evaluation of the proposed revised §264 regulation with respect to these realities reveals that it seems to be a definite improvement over the current regulation, but that it may be quite difficult to develop an adequate statistical plan within its constraints.
[1]
A. Cohen,et al.
Simplified Estimators for the Normal Distribution When Samples Are Singly Censored or Truncated
,
1959
.
[2]
Robert D. Gibbons,et al.
Statistical Models for the Analysis of Volatile Organic Compounds in Waste Disposal Sites
,
1987
.
[3]
John A. Glaser,et al.
Trace analyses for wastewaters
,
1981
.
[4]
D. Owen,et al.
One-sided distribution-free simultaneous prediction limits for p future samples
,
1986
.
[5]
Robert G. D. Steel,et al.
A multiple comparison rank sum test: Treatments versus control.
,
1959
.
[6]
D. Helsel,et al.
Estimation of Distributional Parameters for Censored Trace Level Water Quality Data: 2. Verification and Applications
,
1986
.
[7]
A. Gleit,et al.
Estimation for small normal data sets with detection limits.
,
1985,
Environmental science & technology.
[8]
C. Dunnett.
A Multiple Comparison Procedure for Comparing Several Treatments with a Control
,
1955
.
[9]
Robert J. Gilliom,et al.
Estimation of Distributional Parameters for Censored Trace Level Water Quality Data: 1. Estimation Techniques
,
1986
.
[10]
R. Gibbons.
Statistical prediction intervals for the evaluation of ground-water quality
,
1987
.
[11]
Roger J. McNichols,et al.
One-Sided Intervals for at Least p of m Observations From a Normal Population on
,
1987
.