Jennifer Niemann - Statistical Modeling of Environmental Data with Non-Detects

Abstract

In the environmental sciences, portions of collected data are often reported as non-detect, meaning that the actual data point is known only to be below the detection limit of a measuring device.  In mainstream statistics, this type of data is known as left-censored data. Oftentimes, data sets include two detection limits for various reasons. In  this  project, the performance of a variety  of substitution methods will be examined, as well as maximum  likelihood estimation and the Kaplan-Meier  method for estimating  summary statistics (primarily the mean) of left-censored data with respect to certain statistical criteria like “bias” and “mean squared error” for both one and two detection limit scenarios. The performance of these methods were also investigated in the context of construction of confidence intervals for mean and upper tolerance limits. After identifying the best method, the results were applied to a real life environmental data set provided by Neptune and Company, Inc.