Statistical analysis of water quality data affected by limits of detection
Date
1986
Authors
Porter, Paul Steven, author
Ward, Robert C., advisor
Bell, Harry F., committee member
Loftis, Jim C., committee member
Skogerboe, Rod, committee member
Journal Title
Journal ISSN
Volume Title
Abstract
Many water quality problems are related to substances which are present at concentrations too low to be measured precisely. Obtaining information from a monitoring system which produces many results near the fringes of analytical capabilities is not straightforward. This thesis is a discussion of the concerns one should have when statistically analyzing water quality data from such a system. Two general approaches are discussed. The traditional approach is to regard all measurements as precise or imprecise. Precise results are simply numerical responses, for which statistical analysis may lead to valid and sound monitoring information. Imprecise results are reported as "ND", or not detected, with criteria for reporting based on categories of measurement precision. Measurement error which leads to censoring is described. The impact of this error on the statistical characteristics of water quality data is illustrated using a model appropriate for analyte concentrations near the limit of detection. It is shown that the statistical properties of a set of measurements may not resemble the population from which samples were taken. This suggests the use of statistical methods which acknowledge observation error. Loss of information due to censoring is demonstrated and it is proposed that a numerical result be reported for all measurements. It is also suggested that an estimate of data precision accompany all results. This would permit the data user to censor at levels of uncertainty chosen by the user, rather than having information censored by the measurement process. When the number of results with significant observation error is small, or when data has been censored and no information is available regarding such error is available, statistical methods intended for censored data may appropriately be used. Such methods covering a variety of water quality problems are reviewed. Numerical examples of many methods are provided.
Description
Rights Access
Subject
Water quality -- Measurement