WORK PLANS - Laserfiche WebLink

Home Browse Search

an example, background threshold values (BTVs) and exposure point concentration(EPC)terms should be estimated by reliable statistics (and not distorted statistics) obtained using data sets representing the main dominant population under study (e.g., site,background). The low probability high outlying values contaminate the underlying left-censored or uncensored full data set from the population under study. The inclusion of outliers in a background data set needs to be justified before performing other relevant statistical analyses including the estimation of BTVs. If possible, all interested parties should be involved in decision making about the disposition(inclusion or exclusion) of outliers in a background data set. Typically, outlying locations (if any) with elevated concentrations need separate investigation. It should be noted that the objective is to compute reliable background statistics based upon the majority of a defensible background data set representing the dominant background population. In the process of estimating the BTVs, it may not be desirable to accommodate a few low probability outlying observations (if any) by using a lognormal distribution(Singh, Singh, and Iaci, 2002). The use of a lognormal distribution often accommodates outliers and multiple populations, which in turn yields inflated UCLs and background statistics such as UPLs, percentiles, and UTLs. The proper identification of multiple outliers is a complex issue based upon robust statistical methods, and is beyond the scope of ProUCL 4.0. For details of the robust outlier identification procedures, refer to Barnett and Lewis (1994), and Singh and Noccrino (1995). Amore complicated problem arises when the collected background data set may represent a potentially mixture data set including observations from some of the site areas. The occurrence of mixture samples is quite common in many environmental applications. This is especially true when data sets are collected from large federal facilities (e.g.,Navy Sites). For such cases,the underlying data set may consist of samples from the background areas as well as from some other potentially contaminated site areas. In this situation, first, one has to separate the background observations from the other site related observations. After the background data set has been properly extracted from a potentially a mixture sample, one can proceed with the computation of background statistics as available in ProUCL 4.0. Appropriate population partitioning techniques (e.g., see Singh, Singh, and Flatman (1994)) can be used to extract a background data set from a potentially mixture data set. However, the population partitioning methods are beyond the scope of ProUCL 4.0. It should be noted that some of those methods will be available in Scout(EPA, 2000) software which is currently under revision and upgrades. For methods as incorporated in ProUCL, it is assumed that one is dealing with a sample from a"single"population representing a valid site-related background data set. Therefore, before using statistical methods to compute the various limits such as UCLs, UTLs, and UPLs, it is suggested that the user pre-processes the data set to identify potential outliers and mixture populations (if any). Outlier Tests ProUCL 4.0 has a couple of classical outlier test procedures, such as the Dixon test and the Rosner test. Additionally,ProUCL 4.0 software has exploratory graphical methods including