|
Abstract:
|
Scores of chemists, toxicologists and environmental scientists flounder in analyzing data with values below a limit of detection (aka non-detects, or left-censored data). The U.S. Environmental Protection Agency and numerous other international agencies have developed contradictory and irrational guidelines for the analysis of data with non-detects. Several methods for left-censored data are available, but practitioners struggle with choices of parametric and nonparametric models. The non-parametric Turnbull estimator is poorly understood and implemented with incorrect plots in 3 major software packages. Statisticians, who might be helpful, are generally not taught how to handle left-censored data, which is considered 'rare' in many survival textbooks. Yet left-censored data, in the form of data below a detection limit, are abundant. Statistical methods for handing left-censoring would appear to be a trivial modification of right-censored data, but there are many pitfalls. Raising awareness of left-censored data among statisticians is crucial to both helping our colleagues in the lab sciences and further developing statistical methods and software for left-censored data.
|
