Handling Non-Detects

It is important to understand how to properly handle samples that are classified as non-detects. A non-detect is an analytical sample where the concentration is deemed to be lower than could be detected using the method employed by the laboratory. Non-detects are accommodated in EVS for analysis and visualization using a few very important parameters that should be well understood and carefully considered. These parameters control the clipping non-detect handling in all of the EVS modules that read chemistry (.apdv, or .aidv) files. The affected modules are 3d estimation, krig_2d, post_samples, and file_statistics.

Non-detects should "almost" never be left out of the data file. They are critically important in determining the spatial extent of the contamination. Furthermore, it is important to understand what it means to have a sample that is not-detected. It is not the same as truly ZERO, or perfectly clean. In some cases samples may be non-detects but the detection limit may be so high that the sample should not be used in your data file. If the lab (for whatever reason) reports "Not detected to less than XX.X" where that value XX.X is above your contaminant levels of interest, that sample should not be included in the data file because doing so may create an indefensible "bubble" of high concentration.

As for WHY to use a fraction of the detection limit.  At each point where a measurement was made and the result was a non-detect, we should use a fraction of the detection limit (such as one-half to one-tenth).  If we were to use the detection limit, we would dramatically overestimate the actual concentrations.  From a statistical point of view, when we have a non-detect on a site where the range of measurements varies over several orders of magnitude, it is far more probable that the actual measurement will be dramatically lower than the detection limit rather than just below it. Statistically, if the data spans 6 orders of magnitude, then we would actually expect a non-detect to be 2-3 orders of magnitude below the detection limit! Using ONE-HALF is inanely conservative and is a throwback to linear (vs log) interpolation and thinking.

When you might drop a specific Non-Detect: If your target MCL was 1.0 mg/l, and the laboratory reporting limit for a sample were 0.5 mg/l, you would be on the edge of whether this sample should be included in your dataset. If you plan to use a multiplier of one-half, it would make the sample 0.25, which is far too close to your MCL given that the only information you really have is that the lab was unable to detect the analyte. If you use a multiplier of one-tenth, it is probably acceptable to include this sample, however if the nearby samples are already lower than this value, we would still recommend dropping it.

Recommended Method: The recommended approach for including non-detects in your data files is the use of Less Than signs "<" preceding the laboratory detection limit for that sample. In this case,the Less Than Multiplier affects each value, making it less by the corresponding fraction.

Otherwise, you can enter either 0.0 or ND for each non-detect in which case, you need to understand (and perhaps modify) the following parameters:

• The number entered into the Pre-Clip Min input field will be used during preprocessing to replace any nodal property value that is less than the specified number. When log processing is being used, the value of Clip Min must be a positive, non-zero value. Generally, Clip Min should be set to a value that is one-half to one-tenth of the global detection limit for the data set. If individual samples have varying detection limits, use the Recommended Method with "<" above. As an example, if the lowest detection limit is 0.1 (which is present in the data set as a 0), and the user sets Clip Min to 0.001, the clipped non-detected values forces two orders of magnitude between any detected value and the non-detected values.

• The Less Than Multiplier value affects any file value with a preceeding "<" character. It will multiply these values by the set value.

• The Detection Limit value affects any file values set with the "ND" or other non-detect flags (for a list of these flags open the help for the APDV file format). When the module encounters this flag in the file it will insert the a value equal to (Detection Limit * LT Multiplier).