Two results from one sample can look reassuringly similar, or unexpectedly different. A duplicate check helps explain that relationship, but it answers a narrower question than whether a reported concentration is correct. Understanding what was duplicated, how agreement was calculated and which criteria apply makes the check more useful.
This article explains measurement concepts using US environmental and fish-monitoring guidance. Those sources do not establish EU cannabinoid regulatory limits. Their sampling frequencies and acceptance thresholds should not be transferred to another testing context. Information checked: 6 November 2026.
What is a duplicate sample?
“Duplicate” needs a description of the work repeated. The EPA UFP-QAPP manual, section 2.6.2.1, distinguishes laboratory duplicates, where two portions of the same sample are prepared and analysed separately, from analytical duplicates, where portions of an already prepared sample are analysed. Field duplicates also capture variation introduced during sampling.
Imagine a jar containing a powdered material. Taking two portions, extracting each independently and measuring both tests more of the laboratory process than measuring the same prepared extract twice. Neither arrangement automatically tells you how evenly the material was distributed across the original batch.
A useful first question is therefore: “At which stage did the two results become separate?” For the wider relationship between a tested portion and a batch, see batch sampling and measurement uncertainty.
What does RPD mean?
Relative percent difference, or RPD, expresses the difference between two results relative to their average. The EPA duplicate guidance uses this calculation:
RPD = absolute difference between A and B ÷ average of A and B × 100.
For a hypothetical analyte, suppose the results are 10 mg/kg and 12 mg/kg. The absolute difference is 2 mg/kg. Their average is (10 + 12) ÷ 2 = 11 mg/kg. RPD is therefore 2 ÷ 11 × 100 = approximately 18.2%.
That number describes agreement for this pair. It is not a concentration, a safety limit or an automatic pass or fail. Before interpreting it, check that both values describe the same analyte, use compatible units and share the same reporting basis.
Why can small numbers produce large percentages?
The denominator matters. When the average is close to zero, a small absolute difference can produce a large RPD. If the average is exactly zero, the formula is undefined. More decimal places do not resolve that mathematical limitation.
A result reported as “not detected” is also not a measured numerical zero. Substituting zero, or inventing an exact concentration below a reporting limit, can create a percentage the report does not support. Read detection versus quantification on lab reports before calculating with censored results.
Why do matching results not prove accuracy?
Precision describes agreement between measurements. Accuracy concerns how close a measured result is to the value of the quantity being measured. Two results can closely agree while sharing a systematic error. For example, a calibration problem affecting both measurements might shift both in the same direction.
NIST’s measurement terminology distinguishes repeatability under the same specified conditions over a short interval from reproducibility under specified changed conditions. A duplicate pair should not be presented as evidence of agreement across different laboratories unless that comparison actually happened.
Other checks address different questions. EPA quality-control guidance discusses reference materials, laboratory control samples and spikes. These contribute evidence about analytical performance beyond duplicate agreement. For context, read reference materials, certified values and traceability.
Keep the question attached to the evidence. “Were these two portions consistent?” is different from “Was the reference value recovered?” and different again from “Did the sample represent the batch?” A report can contain useful information about one question while leaving another unresolved. Asking the laboratory to identify the purpose of each check helps avoid treating one reassuring number as a complete assessment.
Does a failed duplicate check invalidate everything?
There is no universal RPD pass limit. Applicable criteria depend on the method or project and may distinguish matrices, analytes and concentration ranges. EPA’s fish-monitoring guidance also distinguishes RPD for two results from relative standard deviation for three or more results.
ATSDR guidance on evaluating sampling data calls for assessing quality-control findings against applicable requirements and intended use. A failed check warrants closer evaluation, including relevant qualifiers and consultation with an analytical chemist, rather than an automatic conclusion about every result. This does not permit required corrective actions to be skipped; the applicable method and quality system determine what must happen next.
When discussing a duplicate finding with the laboratory, ask:
- What preparation or measurement steps were repeated?
- Which acceptance criterion applies to this pair?
- Which results are affected, and what qualifiers or corrective actions apply?
- What does the finding change about the intended use of the data?
Read duplicate agreement as one piece of evidence. Its meaning comes from the work repeated, the values being compared and the documented quality-control framework.