A Bayesian approach to evaluating site impairment

In the United States, each state is required to list water resources that are declared to be impaired under guidelines set by the Clean Water Act. Measurements are typically collected on a number of chemical constituents and compared with a standard. If there are too many measurements exceeding the standard, then the site is declared impaired. The approach is non-statistical but similar to a Binomial test. The Binomial approach would convert the measurements to binary data then test if the proportion exceeding the standard is excessive. Both methods convert measurements to binary values hence exclude potentially important information in the data. We present a statistical approach using a Bayesian model that uses the raw data instead of the binary transformed data. The population distribution of a family of location-scale parameter models is studied under the model. Posterior distributions from the Bayesian analysis are used in the decision-making process and error probabilities for the Bayesian and the Binomial approaches are compared for a normal population.