Did you know that a scientific table showing ‘not detected’ does not necessarily show a measurement of zero? This is one of the easiest ways a careful school investigation can accidentally turn a limitation of its instrument into a false conclusion.
A detection limit is a method’s threshold for reliably recognising the presence of a particular thing under specified conditions. A non-detect means the method did not establish detection under those conditions; it does not prove that the substance or signal is absent. When a result is reported as below a stated reporting limit, keep that limit and the qualifier alongside the result rather than silently changing it to 0. The exact meanings of detection and reporting limits depend on the method, so read the source’s legend.
What an instrument can say — and what it cannot
A ruler that marks centimetres cannot report a hair’s width with unlimited precision. In the same way, a measurement method may be capable of identifying a signal above some threshold while having too little evidence to quantify a smaller signal confidently. The words ‘detected’, ‘estimated’, ‘below the limit’ and ‘not tested’ therefore describe different states. School data tables often become unreliable when these categories are flattened into one number.
A measurement of exactly zero is a claim about the measured quantity. A non-detect is a claim about what an instrument or analysis could establish. A blank row may mean the sample was never processed at all. A result marked ‘invalid’ may mean a test failed its quality checks. None of those statements should be interchanged merely because the numerical cell would otherwise be empty.
The US Environmental Protection Agency’s data-quality guidance explicitly discusses observations below detection limits as non-detects rather than treating the result itself as a measured zero. The US Geological Survey’s laboratory reporting explanation distinguishes detection levels from reporting levels and explains why low-level results require careful qualification. These are technical references for the principle, not claims that a particular classroom device meets an EPA or USGS standard.
A worked example with invented classroom readings
Imagine a teacher provides three fictional water samples, called A, B and C, together with a simplified classroom test for a coloured tracer. For this exercise only, suppose the method reliably reports values of 0.20 milligrams per litre (mg/L) or higher and uses the report ‘less than 0.20 mg/L‘ for lower results. This is an assumed teaching rule, not a real manufacturer’s specification or an assertion about drinking-water safety.
The invented results are: Sample A = 0.60 mg/L; Sample B = <0.20 mg/L; Sample C = 0.40 mg/L. A hurried student types A = 0.60, B = 0, C = 0.40 into a spreadsheet, then averages all three. The calculation gives (0.60 + 0 + 0.40) ÷ 3 = 0.333… mg/L. The arithmetic is correct for those three entered numbers, but the data entry has invented an exact value for B.
Under the deliberately simplified threshold assumption for this exercise, if B is non-negative and less than 0.20 mg/L, the three-sample arithmetic mean could lie between 1.00 ÷ 3 = 0.333… mg/L and (1.00 + 0.20) ÷ 3 = 0.400 mg/L, excluding the upper endpoint. We do not know B’s exact concentration. The interval illustrates incomplete information, not a valid estimate for a real laboratory dataset. Actual detection limits involve probabilities, uncertainties and method-specific reporting conventions; a simple cutoff is not an absolute physical bound on the true concentration.
Notice the difference between saying ‘B was not detected by this method’ and ‘B contained exactly no tracer’. The first is a report about test evidence. The second would need stronger evidence than the non-detect provides. An honest investigation records A, B and C with their original qualifiers and, if an average is required, explains which statistical treatment is appropriate rather than silently inserting zero.
How to record the result so nobody loses the meaning
- Keep the sign: record
<0.20 mg/Lor the laboratory’s exact non-detect flag, not the number 0.20 alone and not 0. - Keep the units: milligrams per litre are not interchangeable with grams per litre. 0.20 mg/L equals 0.00020 g/L, if a unit conversion is actually needed.
- Keep the method identifier: note which instrument, test procedure or dataset produced the result and how its reporting limit was defined.
- Keep missingness separate: a sample with no test performed should be marked ‘not measured’, not ‘below detection limit’.
- Keep uncertainty visible: if an analysis has a quality flag or estimated value, retain the flag alongside the value.
- Keep original records: make analyses on a separate working copy rather than overwriting the teacher’s or laboratory’s results.
A useful learner diagnostic
Show students the invented table and ask them three questions before any calculations: ‘Was B measured as zero?’ ‘Can B be reported as exactly 0.10 mg/L?’ ‘What does the inequality sign tell you?’ A strong response is ‘No; no; the method reports B below its stated limit, not its precise value.’ A learner who supplies 0.10 by taking half the limit has selected a possible modelling convention without stating the assumption. It is not a laboratory measurement.
To identify the source of the error, ask whether the pupil knows the difference between a mathematical inequality and a number suitable for arithmetic. A student may understand the instrument’s limit yet still paste only numeric values into a graphing tool. The repair is to preserve the original category and choose an analysis that accommodates qualified observations.
Progressive practice — from labels to reasoning
- Guided: for a classroom threshold of 0.50 units, classify ‘0.80 units’, ‘<0.50 units’ and ‘not tested’. Which is a quantified reading, which is a below-limit report and which is missing?
- Paired: a partner changes ‘<0.10 mg/L’ to ‘0 mg/L’ in a fictional results table. Explain exactly what new claim was introduced and why the change is unsupported.
- Independent: four fictional samples contain 0.3, 0.7, ‘<0.2’ and 0.5 mg/L. Write a fair one-paragraph summary that states three quantified readings and one non-detect without fabricating the fourth reading.
- Transfer: compare ‘below the sound-meter threshold’, ‘no one pressed the record button’ and ‘the recording failed its calibration check’. Explain why these produce three different missing-or-uncertain data statuses.
- Challenge: in the simplified classroom cutoff model, two quantified values are 0.30 and 0.50 units, and one non-negative value is known only as ‘<0.20 units’. Explain why the total is at least 0.80 and below 1.00 units, without choosing an exact third value.
Why simply replacing non-detects with half the limit is not a universal rule
Some statistical and regulatory workflows use assumptions or specialised methods to analyse datasets containing non-detects. A convenient substitution such as one-half the limit is not an observed result and is not appropriate in every context. Its bias depends on the underlying distribution, number of non-detects and research purpose. Even regulatory guidance distinguishes different treatment choices instead of declaring one universal number for every case.
For schoolwork, the safest general action is to preserve the report and state that a numerical summary may require a specified method. If the teacher assigns a simplification solely for learning arithmetic, label it as an assumption in the calculation, never as a recovered measurement. For real environmental or health decisions, use the relevant professional method and reporting requirements rather than this classroom example.
What parents and teachers can do next
Teachers can give pupils one table containing a valid zero, an estimated result, a non-detect and a missing sample. Ask them to produce two outputs: an unaltered evidence table and a careful written interpretation. That checks understanding of what the data permit, not merely spreadsheet fluency.
Parents do not need laboratory equipment to discuss this. Try a household example: ‘We didn’t hear the very quiet sound’ is not always equivalent to ‘there was no sound’. Ask what the listening conditions were and what evidence could distinguish the possibilities. Do not turn the analogy into a claim about a medical diagnostic test.
Progress is visible when a student preserves units and flags, requests the method’s threshold, states an assumption before calculations, and refuses to claim a precise value from an imprecise report. Repeat the skill in an unfamiliar dataset to see whether the habit transfers.
Misconceptions, safeguards and limits
- ‘Less than 0.20 means 0.20.’ No. It is not an equality. Even the interpretation of the threshold requires the laboratory’s defined reporting convention.
- ‘Not detected means absent.’ No. Failure to establish detection is not proof of an exact zero.
- ‘Not detected means half the limit.’ No. Half a limit can be an explicit modelling choice under a justified procedure, but it is not the underlying observation.
- ‘A blank is the same as a non-detect.’ No. Missing measurements, failed tests and below-limit results need separate labels.
- ‘All instruments share one limit.’ No. Different methods, units, sample preparation and conditions can produce different detection and reporting limits.
- ‘The simple classroom interval is a regulatory decision rule.’ No. The example deliberately simplifies the reporting mechanism for mathematical understanding and must not guide real-world safety judgments.
Frequently asked questions
What does ND mean? It often means ‘not detected’, but check the table legend and laboratory’s actual definition before interpreting a particular dataset.
Why does a report sometimes show ‘<0.20’ instead of ND? The less-than format gives a reporting threshold as well as a flag that an exact number was not established. Some reports use a code and a limit in separate columns.
Can a real laboratory report a low concentration as estimated rather than non-detect? Yes, depending on its analytical and reporting framework. That is why a pupil must retain flags rather than collapsing all low values together.
Can a non-detect still matter in a graph? Yes. It may affect the supported conclusion and uncertainty. A graph should not place the observation at an exact zero unless that transformation has been explicitly justified and labelled.
What if the task demands an average? Ask which procedure to use, preserve the raw observations and explain the chosen treatment. The correct approach depends on the measurement context and how many results are censored.
The bigger lesson is to respect the boundary of evidence
A good scientist distinguishes what happened, what was observed and what the method was capable of revealing. A ‘less than’ sign is part of the evidence. Protecting it prevents an appealing but unjustified calculation from becoming an invented fact.
For linked learning, compare why no visible change need not mean zero effect and blank versus negative controls. Explore the eduKate Education learning hub and the Clementi diagnostic teaching reference. For primary technical definitions, see the EPA Data Quality Assessment guidance and USGS laboratory reporting explanation.