Statistical approaches to groundwater monitoring

Current and proposed regulations require that some form of Student's t-test be used for evaluation of groundwater pollution detection parameters. Several problems arising from that requirement lead to an inflated false alarm rate. Among these are the use of variability among replicates as the estimate of random sampling error, and the failure to take proper account of spatial and temporal sources of variation. In addition, conditions essential for the valid application of the t-test are usually lacking. The requirement that a separate t-test be conducted for each detection parameter, each reporting period, causes an additional severe increase in the likelihood of a false alarm. A dummy variable analysis of covariance is suggested as a desirable alternative. After statistically removing temporal variation from the data the averages of the upgradient and downgradient wells are compared. Examples of the use of this form of analysis are given.