Simultaneous nonparametric prediction limits

Prediction limits are often attractive for carrying out multiple-comparison s-with-control hypothesis tests. We present algorithms for computing simult aneous confidence levels of partially sequential nonparametric prediction l imit tests used in environmental monitoring. Algorithms are given for "p-of -m" (m chances to get p observations "inbounds" at each of r locations to " pass"), "California" (either the first or all of the next m - 1 observation s inbounds), and "modified California" (either the first or two of the next three observations inbounds) resampling strategies. The prediction limit c an be any order statistic of the control or background sample. This methodo logy is particularly useful when a high proportion of observations are "non detects." We demonstrate its use in evaluating groundwater chemistry measur ements at facilities monitored under U.S. Environmental Protection Agency r egulations. Regulatory and practical considerations and limitations in this application area are discussed.