TRUE AND FALSE POSITIVE RATES IN MAXIMUM CONTAMINANT LEVEL TESTS

The U.S. Environment Protection Agency (EPA) is promulgating a revised national primary drinking water regulation (NPDWR) which includes a monthly sample size and maximum contaminant level (MCL) for total coliform bacteria in public water systems. No previous quantification has been made of the coliform content that must be present in the sampled water in order for an MCL to be exceeded. This paper presents a method for evaluating the coliform level an MCL will detect with likelihood P. Our approach is to treat an MCL as a decision rule, with Type I (false positive) and Type II (false negative) error rates. The stringency of an MCL is quantified as the mean coliform level in the sampled system that it will detect with likelihood P. MCLs are contrasted on stringency by comparing the mean coliform level each targets for detection, with fixed error rates. Interim rules (NIPDWR), in effect since 1975, are shown to vary widely on the coliform content each targets for detection, that is, on stringency. Yes/no decisions on contamination have not been decided based on mean coliform content. Coliform levels permitted in monitored public water systems have been determined by the particular MCL used for testing. The same coliform level will test positively with one MCL 90 times in 100 yet be guaranteed 95% nondetection by a second MCL. EPA's reasonably safe standard for drinking water is reformulated on our stringency criteria. Its proposed monthly MCL is evaluated on its capability for maintaining this standard. Smaller systems will not provide its users this level of protection under the new rule. In addition, our evaluation of the safe water standard on stringency and the rationale for a monthly MCL require that coliform levels be identically distributed (i.d.) across month and sampled system. Empirical data strongly refute this model and question the utility of a monthly MCL. This work suggests an alternative, single sample MCL, with repeat sampling for verification, which can be configured to provide monitoring to discover mean coliform values at any level, in any size of system, at minimal extra cost.