Nondetects, Detection Limits, and the Probability of Detection.

When chemists cannot quantify the concentration in a field sample, they report nondetect instead of a numerical measurement (a detect). A data analyst might assume that all nondetects are zeros, all nondetects are smaller than the smallest detect, or that all nondetects are below the detection limit. This paper shows that these assumptions can be incorrect and develops better methods for describing the relationship between nondetects and detects. The crucial step is to separate two questions, which measurements are likely to be reported as nondetect and which field concentrations are likely to give measurements that are reported as nondetect? Since measurements are imprecise and often biased downwards, the answers to the two questions are different. The methods developed are applied to five soil pollutants from the 1988 Love Canal Study. These examples show that the detection limit is not always a good upper bound for nondetects. Moreover, in that study the estimated probability of detecting a field concentration at the detection limit was sometimes near one and sometimes near 0.5.