EVALUATION OF STATISTICAL METHODS FOR CENSORED DATA: A METHYL TERT-BUTYL ETHER/TERT-BUTYL ALCOHOL CASE STUDY

Environmental measurements often include concentrations recorded as below detection limits or censored. Multiply censored data are common when samples are diluted or collected from the laboratories with different detection limits. To have a correct picture of the environmental conditions, an appropriate statistical method for multiply censored data must be applied. This study determines the appropriate statistical methods to evaluate multiply censored water quality data. An S-Plus module program is used. The case study investigated is the occurrence of methyl tert-butyl ether (MTBE) and tert-butyl alcohol (TBA) in groundwater impacted by leaking underground fuel tanks. Seventy three facilities were randomly selected from the greater Los Angeles basin, and a total of 629 groundwater samples were obtained. The data sets showed 28 % censoring of the MTBE and 66 % censoring of the TBA data. Two distributional methods, one regression, and one substitution method (assigning half of the detection limits to the censored values) were tested to estimate the distribution parameters for MTBE and TBA. The log-regression method outperformed other distributional methods which have distributional assumption and transformation bias. The substitution method produced very similar results to th