On estimating population parameters in the presence of censored data: overview of available methods

Abstract

This paper examines recent results presented on estimating population parameters in the presence of censored data with a single detection limit (DL). The occurrence of censored data due to less than detectable measurements is a common problem with environmental data such as quality and quantity monitoring applications of water, soil, and air samples. In this paper, we present an overview of possible statistical methods for handling non-detectable values, including maximum likelihood, simple substitution, corrected biased maximum likelihood, and EM algorithm methods. Simple substitution methods (e.g. substituting 0, DL/2, or DL for the non-detected values) are the most commonly used. It has been shown via simulation that if population parameters are estimated through simple substitution methods, this can cause significant bias in estimated parameters. Maximum likelihood estimators may produce dependable estimates of population parameters even when 90% of the data values are censored and can be performed using a computer program written in the R Language. A new substitution method of estimating population parameters from data contain values that are below a detection limit is presented and evaluated. Worked examples are given illustrating the use of these estimators utilizing computer program. Copies of source codes are available upon request.