The analysis of censored data: An application to toxic contaminant data from the South Saskatchewan River

In routine water quality monitoring of toxic contaminants, a number of difficulties are usually encountered in the statistical interpretation of the results. A major reason for this is the frequent occurrence of water sample concentrations below the limit of detection. In this paper the issue of dealing with values below the detection limit is discussed through a case study using data from the South Saskatchewan River. Some comments are also given outlining the properties and the difficulties associated with the commonly used ad hoc methods for estimating the population characteristics when the data are subject to type I censoring. Specifically inferences about the mean value, trend and seasonal changes are considered.