Resampling

If you go to the doctor and he tells you that you need major surgery, what do you do? Get a second opinion. Retesting/resampling can help with various issues (false positives from results coming from the far ends of the populations, sampleing , or laboratory error). Resamples should NOT be collected immediately after the primary samples because the sample will not be independent (do not just collect twice the volume and put them on hold). However, the resample should be completed before the next compliance-sampling event otherwise it is not a resample it is just the next sample. In addition, resampling can help with the statistical power of the test.

Note that Section 5.3.3 of the Unified Guidance states:

“If a retest confirms an SSI, background should not be updated. Rather, some regulatory action at the site should be taken. But what if an initial exceedance of a prediction or control limit is disconfirmed by retesting? According to the logic of retesting (Chapter 19), the well passes the compliance test for that evaluation and monitoring should continue as usual. But what should be done with the initial exceedance when it comes time to update background at the well?

The initial exceedance may be due to a laboratory error or other anomaly that has caused the observation to be an outlier. If so, the error should be documented and not included in the updated background sample. But if the exceedance is not explainable as an outlier or error, it may represent a portion of the background population that has heretofore not been sampled. In that case, the data value could be included in the updated background sample (along with the repeat sample) as evidence of the expanded but true range of background variation.”

Based on the Unified Guidance referenced above, you should include exceedances in the background datasets if the following three criteria are met:

1. The initial exceedance is disconfirmed by retesting.

2. There is no evidence that the initial exceedance is the product of a testing error.

3. There is no evidence that the initial exceedance is an outlier.