ASTM D7782-13

5.2.1 Routinely Achievable Detection—The laboratory is able to attain detection performance routinely, using studied measurement systems, without extraordinary effort, and therefore at reasonable cost. This property is needed for a detection limit to be practically useful while scientifically sound. Representative equipment and analysts must be included in the study that generates the data to calculate the WDE.

5.2.2 Inclusion of Routine Sources of Error—If appropriate data are used in calculation, the WDE practice will realistically account for sources of variation and bias common to the measurement process and routine for sample analysis. These sources include, but are not limited to: 5.2.3 Exclusion of Avoidable Sources of Error—The WDE practice excludes avoidable sources of bias and variation, (that is, those which can reasonably be avoided in routine field measurements). Avoidable sources would include, but are not limited to: 1) inappropriate modifications to the method, the sample, measurement procedure, or measurement equipment, and 2) gross and easily discernible transcription errors (provided there was a way to detect and either correct or eliminate such errors in routine sample testing).

5.2.4 Low Probability of False Detection—Consistent with a measured concentration threshold (YC), the WCL is a true concentration that will provide a high probability (estimated at 99 %) of true non-detection (and thus a low estimated probability of false detection (α) equal to 1 %). Thus, when a sample with a real concentration of zero is measured, the probability of not detecting the analyte (that is, the probability that the measured value of the blank will be less than the WCL) would be greater than 99 %. To be most useful, this property must be demonstrated for the particular matrix being used, and not just for reagent-grade water.

5.2.5 Low Probability of False Non-detection—Where appropriate data have been used for calculations, the WDE provides a true concentration at which there is a high estimated probability (at least 95 %) of true detection (and thus a low estimated probability of false non-detection (β) equal to 5 % at the WDE), with a simultaneously low estimated probability of false detection. Thus, when a sample with a true concentration at the WDE is measured, the probability of detection would be estimated to be at least 95 %. To be useful, this property must be demonstrated for the particular matrix being used, and not just for reagent-grade water.

1.2.1 First, rather than making a single-point estimate of precision, the WDE protocol requires an estimate of precision at multiple points in the analytical range, especially in the range of the expected detection limit. These estimates are then used to create an appropriate model of the method’s precision. This approach is a more credible way to determine the point where relative precision has become too large for reliable detection. This process requires more data than has been historically required by single-point approaches or by processes for modeling the relationship between standard deviation and concentration.

1.2.2 Second, unlike most other approaches, the WDE process accounts for analytical bias at the concentrations of interest. The relationship of true concentration to measured concentration (that is, the recovery curve) is established and utilized in converting from as-measured to true concentration.

1.2.3 Third, most traditional approaches to detection limits only address the issue of false positives. Although false negatives may not be of concern in some data uses, there are many uses where understanding and/or control of false negatives is important. Without the false-negative-control information, data reported with just a critical-level value are incompletely described and the qualities of data at these levels incompletely disclosed.

1.2.4 Fourth and last, the WDE standard utilizes a statistical-tolerance interval in calculations, such that future measurements may reasonably be expected to be encompassed by the WDE 90 % of the time. Many older approaches have used the statistical confidence interval, which is not intended to encompass individual future measurements, and has been misunderstood and misapplied. Procedures using the confidence interval cannot provide the stated control when the detection-limit value is applied to future sample results; such application is the primary use of these values.