A water-quality spreadsheet contains a value written as “<0.05”. It is tempting to replace it with zero so a chart or average will work. That convenient edit changes the meaning of the observation before the analysis has even begun.
The entry indicates a qualified result relative to a stated limit, in the units used by that dataset. It does not report a measured concentration of exactly zero. The meaning of the limit also depends on the laboratory method and reporting conventions, which should be checked rather than inferred from the symbol alone.
Keep the qualifier beside the number
The USGS guide to groundwater-quality web services describes separate fields for an analyte value not detected, the detection limit and its units. That structure is a useful reminder: the qualifier is part of the result, not decoration that can be discarded during import.
In a fictional working table, preserve the original text “<0.05 mg/L”. You might also store the numerical threshold 0.05, the unit mg/L and a separate qualifier in distinct columns. Label the threshold accurately; it is not a newly measured concentration.
Leave a genuinely missing result distinguishable from a qualified laboratory result. “No sample collected” and “sample analysed but reported below a limit” describe different events, even if neither gives you an ordinary concentration value for plotting.
Why a row of zeros can mislead
Suppose two fictional methods report limits of 0.05 and 0.5 mg/L. Replacing all their non-detects with zero hides the different sensitivity and reporting information. A line chart may then suggest identical measurements where the records actually contain different bounds or qualifications.
Nor does choosing half the limit automatically solve the problem. Any substitution used in a statistical method needs a stated rationale and an assessment of its suitability for the dataset. The USGS explanation of detection limits also shows why detection is connected to uncertainty, rather than an absolute boundary between substance present and substance absent.
Before calculating a summary, count the qualified observations, check whether limits changed across samples, and decide how the intended analysis handles them. If that method has not been chosen, an honest first output is a table of reported values and qualifiers. Preserving uncertainty makes later analysis possible; silently converting it into zero makes the original evidence harder to recover.

