Abstract
Air quality data often contain several observations reported only as below the analytical limit of detection (LOD), resulting in “censored” data sets. Such censored and/or truncated data sets tend to complicate statistical analysis. This paper discusses various procedures for estimating mean concentration and Its 95 percent confidence bounds from air contaminant data sets which contain values that are reported as below the LOD, A quantitative approach for assessing the uncertainty Inherent In the estimated mean concentration due to (a) presence of values below the LOD In the data set, and (b) natural variability of atmospheric concentration data, is described. The utility of this approach In the analysis and interpretation of ambient pollutant concentration data Is demonstrated for both hypothetical and observed singly-censored data sets, and for a multiply-censored, multi-pollutant observed concentration data set. The methodologies discussed here should be particularly useful In verifying compliance with environmental regulations, and In estimating the risks associated with long-term exposure of populations to toxic air contaminants and assessing the uncertainty associated with these estimates.