A survey of drinking water results is only as sharp as the lowest number the laboratories could report. When a sample lands below the method reporting limit, the lab records a qualified value or a non-detect, and that record carries the limit into every table built from it. A Circle of Blue report on a new study, dated October 23, 2026, makes the general point: "When PFAS concentrations in drinking water fall below laboratory reporting limits" the picture can understate what is present. The published summary does not reproduce the study's full methods, so the discussion below concerns what this kind of finding implies for anyone who pools results, not the study's own figures.
Why pooling inherits the floor
Every reporting limit is a property of a method, a laboratory, and often a specific instrument run. Two utilities sampling the same aquifer through two laboratories may report different floors, even when both labs are competent and both follow accredited procedures. A pooled dataset does not erase those differences. It averages them, or more often it treats a non-detect at one floor as equivalent to a non-detect at another. The result looks like a single measurement but is really a blend of several.
The direction of the bias usually runs toward zero for the samples that matter most. Results near the floor are the ones most likely to be censored, and they are also where a rising concentration first appears. A dataset that drops or flags those values will show fewer detections than the water contains. Whether the size of that gap is small or large is a question no summary can settle.
Handling censored values before the table is built
Analysts have several conventions for censored data. Substitution with zero, or with half the limit, is simple and easy to audit, but it fixes the answer before the data speaks. Maximum likelihood approaches treat each non-detect as a value somewhere below its own limit and estimate the distribution that best fits the remaining results. Both methods require the reporting limit for each record. A pooled file that discards the limit field has already lost the information needed to choose among them.
The practical step is to keep the limit, the method identifier, and the laboratory name on every row. A column that records only the qualifier "ND" reads cleanly in a spreadsheet and cannot be corrected later. Researchers who inherit a file without these fields should treat any prevalence estimate as a lower bound until they have checked the source records, and they should say so in the methods section rather than in a footnote.
What a reader of a published figure should check
For a professional reader, the useful questions are narrow. Which method and floor produced each value? Were non-detects counted as absent, as censored, or replaced with a substituted number? Did sampling use a single laboratory across the period? If a study cannot answer these, a headline percentage describes its sampling choices more than the aquifer. That is not a criticism of any particular study, which may have answered them in full. It is a reminder that a secondary account rarely carries the methods.
Utilities and home water professionals should apply the same discipline to their own records. A household that receives a non-detect on one test and a detection on another may be looking at two floors rather than a change in the water. Anyone weighing an in-home treatment unit against a retest should first confirm which floor applies, then read the warranty and service terms, including whether any coverage for home systems and appliances applies to the equipment in question.