Keep detection limits distinct from observed values

Record the laboratory information faithfully

A result below a detection or quantification limit is not necessarily zero. It may indicate that the assay cannot distinguish or quantify the value reliably under the stated conditions. Begin by obtaining the laboratory's definitions, units and flags rather than assuming that every “less than” result has the same meaning.

The USGS discussion of nondetect substitution explains how replacing nondetects with a fixed fraction of a limit can distort statistical relationships. The appropriate analysis depends on the measurement process, the scientific target and the amount and pattern of censoring.

Preserve values and limits separately

Create separate fields for the reported result, censoring indicator, relevant limit, units and assay or batch identifier. Keep the original text result for traceability. Check whether limits vary between samples, runs or study periods. A change in assay sensitivity can create an apparent change in the outcome distribution.

Do not convert a string such as “<5” into the observed number five without retaining its qualifier. Distinguish a nondetect from a missing specimen, failed assay or result outside the upper range. Those mechanisms may require different handling and should not share one generic missing code.

Hypothetical example

Two fictional specimens are reported as below two units and below ten units. Replacing both with zero would erase their different information bounds. Replacing each with half its limit would create values of one and five, but those numbers were not measured concentrations.

An analysis using such substitutions must disclose them as assumptions and examine their influence. The example does not prescribe a distribution or establish that a particular censored-data model will be suitable.

Choose an analysis that matches the information

Summarise the proportion censored and the distribution of limits by relevant groups. Consider methods designed for censored measurements when their assumptions are defensible, and use sensitivity analyses to show how conclusions depend on handling choices. Avoid applying a universal percentage cutoff for substitution without a study-specific rationale.

Report the laboratory limits, flags, exclusions and statistical method. Use the data dictionary change-impact worksheet to preserve changes in assay definitions. If the data contain too little information below the limits to support the intended comparison, state that limitation rather than manufacturing precision through arbitrary replacement values.

Source verification: 6 November 2026. Examples are hypothetical.