Statistical Analysis of Environmental Data with Nondetects. A Solution for Practitioners

• Activity Number: 127
• Type: Contributed
• Date/Time: Monday, July 30, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Consulting
• Author(s): Breda Munoz*+ and Robert Truesdale
• Companies: RTI International and RTI International
• Keywords: non detects ; simulations ; MLE

Abstract

In general environmental data are positive and often measured values are less than a limit of detection. These values are referred as "non-detects' in the environmental sciences, and are reported as being less than some reported limit of detection, rather than as actual values. Sometimes detection limits vary from sample to sample, and even between individual observations. When non-detects are present in the data, it is not clear what are the best approaches for estimating descriptive statistics, hypothesis testing and modeling exercises. The size of the dataset, the proportion of non-detects and the distribution of the data are all relevant when selecting the best statistical approaches for a particular analysis. Available parametric and non-parametric approaches work well when the dataset is large and/or the proportion of non-detects is small. It is well documented that for small data sets (fewer than 30 to 50 detected values) methods based on maximum likelihood approach do not work well, leaving the analysis with non-parametric options. We proposed a method that combines Monte Carlo simulations, bootstrap, MLE approaches and non-parametric approaches to produce estimates, their