Inferences about the mean from censored water quality data

Several methods which are commonly used for making inferences about the levels of many metals and organic contaminants in ambient waters from type I censored data are critically evaluated, and their applications are illustrated using the concentrations of polychlorinated biphenyls in water samples from the Niagara River. Difficulties encountered in the application of all the methods, except the method of maximum likelihood, include ignoring the information available about the analytical detection limit and the unavailability of the standard errors for these estimates. Under the assumption that the distribution of the data is lognormal, the log regression method is modified to produce estimates of the censored values which use the detection limit information, and the properties of the maximum likelihood estimates for the lognormal mean are derived. Furthermore, an approximate confidence interval for the lognormal mean is given, and its use for estimating the load to and from the Niagara River is illustrated. The above methods along with a modification to the method of maximum likelihood are evaluated using Monte Carlo simulation.