• A COMPARATIVE STUDY OF THE METHODS FOR ESTIMATION OF DISTRIBUTIONAL PARAMETERS FOR LEFT CENSORED DATA WITH SINGLE NON-UNIFORM DETECTION LIMITS

Abstract

A difficult step in dietary exposure assessment, which is a very important part of radiological/chemical risk analysis, is the handling of concentration data that has been reported below the detection limit (DL). These data are known as censored values or non-detects and therefore the resulting distribution of concentration values is left-censored. Handling left-censored data represents a challenge for statistical analysis of chemical/radiological data. Non detects have been so far  treated  with  widely  used  substitution  methods  recommended  by  international organizations. Based on simulation a comparative study has been carried  out to  assess  the performance  of  different  statistical  methods to handle non-detects, i.e. parametric Maximum  likelihood (ML) methods, and the log-probit  regression  method. Monte Carlo simulations were used to evaluate statistical methods for estimating mean and standard deviation of left-censored concentration data with non-uniform detection limits. Sample size and the percentage of censoring were allowed to vary randomly to generate a variety of left-censored data sets. The log probit regression was the method that yielded high correlation coefficient (r²= 0.92) between mean calculated using log probit method and that of the mean calculated using uncensored samples, similar were the results for the standard deviation.