PSLE Science Reality Lab Vol No.238 | “RPD = 5%” — Does That Mean the Result Is 95% Accurate?

Stable internal ID: PSLE-SCI-REALITY-0238

Wait, what? A laboratory report gives two duplicate results: 48 mg/L and 50 mg/L. Beside them is a neat quality-control number: RPD = 4.1%. Someone points at the small percentage and says, “Excellent. That means the result is about 96% accurate.”

That sounds reasonable because 100% − 4.1% = 95.9%. It is also the wrong scientific move. The percentage is being asked to do a job it was not designed to do.

This is exactly the kind of evidence-transfer problem that matters in PSLE Science. The current 2026 PSLE Science assessment framework asks pupils not only to know scientific ideas, but also to interpret and analyse information, evaluate observations, information and methods, and communicate explanations and reasoning. The 2023 Primary Science syllabus also explicitly values healthy scepticism: question the observation, the method, the process and the data before deciding what follows.

Quick Answer

No. Relative percent difference, usually shortened to RPD, is commonly used to describe how closely a pair of duplicate results agree. It is a precision or agreement check. It does not, by itself, tell you how close either result is to the true or accepted value. A small RPD can occur even when both results are wrong in the same direction.

The central habit for this Reality Lab is simple:

Before interpreting a percentage, ask: percentage of what scientific relationship?

The Owned Learner Job

This article owns one narrow real-world evidence-transfer job: how to evaluate a laboratory or monitoring report that presents RPD for duplicate results without turning that precision statistic into an accuracy score.

It does not replace the existing eduKateSengkang owners for repeated measurements, accuracy, reliability, validity, sampling, measurement uncertainty, graph reading or general answer construction. Use this page when the communication object itself contains an RPD or duplicate-agreement percentage. Route the underlying core skill back to the established PSLE Science guides.

Rebuild the Evidence Object

Imagine an original classroom-quality laboratory report. A water sample is analysed twice as a duplicate pair.

Entry	Reported result
Original analysis	48 mg/L
Duplicate analysis	50 mg/L
Relative percent difference	4.1%
Laboratory control limit	RPD must be 10% or lower for this check

The pair passes the stated duplicate-agreement rule. What can we say?

The two duplicate results are close to each other relative to their average.
The RPD meets the stated control limit in this invented report.
The pair provides evidence about repeat agreement under the conditions represented by the duplicate.

What can we not say from the RPD alone?

That the true concentration is 49 mg/L.
That the laboratory is 95.9% accurate.
That the sample was collected correctly.
That no shared bias affected both measurements.
That every other result in the batch is correct.
That the method is suitable for every sample type or concentration range.

Observed, Claimed and Inferred

Layer	What belongs here?
Observed / reported	48 mg/L, 50 mg/L, RPD 4.1%, stated control limit 10%
Supported claim	The duplicate pair agrees within the stated precision criterion
Possible inference needing more evidence	The measurements are close to the true value
Unsupported leap	RPD 4.1% means 95.9% accurate

This separation matters because scientific reports often put several percentages next to one another. The human brain likes to treat all percentages as if they live on the same scale. They do not. A recovery percentage, completeness percentage, confidence level, defect percentage and RPD can all contain the % sign while answering completely different questions.

What RPD Is Actually Comparing

For two duplicate values, a common RPD calculation compares the absolute difference between the two results with their average. In child-friendly language:

How far apart are these two results compared with the size of the results themselves?

If the pair is 48 and 50, the difference is 2 and the average is 49. The difference is therefore a little over four percent of the average. That is where the RPD comes from. The calculation never asks, “What is the true value?” So the answer cannot magically become a statement about closeness to truth.

Precision Is Not Accuracy

Use two separate questions:

Precision question: Do repeated or duplicate measurements agree reasonably closely with one another?
Accuracy question: How close is the measurement to an accepted reference or true value?

The US Environmental Protection Agency makes this distinction directly in its laboratory QA/QC guidance: precision concerns agreement among replicate measurements without assuming knowledge of the true value, and RPD is a common estimate of precision for duplicate samples. Accuracy requires a different comparison to a reference or accepted value.

Worked Case 1: Tiny RPD, Shared Bias

A reference material has an accepted value of 90 units. A laboratory measures the same prepared material twice and obtains 100 and 102 units.

Question	Evidence	Judgement
Do the duplicates agree?	100 and 102 are close	Yes, precision may be good
Are they close to the reference?	Both are well above 90	No, there appears to be positive bias

This is the most important counterexample in the article. Two measurements can march together in the wrong direction. Shared calibration error, contamination, an interfering signal or another systematic problem can make the pair agree while both are biased.

Therefore:

Agreement with each other is not the same as agreement with reality.

Worked Case 2: Wider Pair, Good Average

The accepted value is 50. Two results are 40 and 60. Their average is exactly 50.

Would it be sensible to say, “The average is perfect, so the measurement process is excellent”? No. The pair is very spread out. The average happened to land on the reference, but the individual results show poor precision. If the next sample is measured only once, which side will it land on?

This case reverses the first one. Case 1 had close duplicates but poor closeness to the reference. Case 2 has an average close to the reference but poor duplicate agreement. Good scientific evaluation keeps both dimensions visible.

Worked Case 3: Same RPD, Different Scientific Consequence

Suppose two different methods each produce an RPD of 5%. Does that mean the evidence quality is equally acceptable? Not necessarily. The acceptable precision depends on the method, concentration, sample type, purpose and established control limits. A five-percent difference may be excellent in one measurement context and unacceptable in another.

This is why you should never memorise a universal rule such as “RPD below 10% is always good.” The control limit belongs to a defined method and quality plan. The scientific move is to find the relevant criterion, not invent one.

What Was Actually Duplicated?

The word duplicate can hide an important design question. Did the laboratory:

take one collected sample and analyse two portions;
prepare the sample twice;
collect two separate field samples side by side;
repeat only the instrument reading;
repeat the entire collection-to-analysis chain?

These designs test different sources of variation. Two instrument readings can be beautifully consistent while sample collection is poor. Two field duplicates include more steps and therefore may expose variability that a narrow laboratory repeat cannot see.

This is not a reason to distrust duplicates. It is a reason to ask what part of the evidence chain the duplicate actually checks.

The Representation Check

When you see a QA/QC table, do not read the number first. Read the headings first.

What are the two measurements?
What unit do they use?
What does the percentage column represent?
Is there a stated control limit?
Does PASS refer to duplicate precision, recovery, contamination, calibration or something else?

A green PASS box can be visually powerful. But a PASS under one quality-control check is not a magical certificate for the whole scientific result. The pass belongs to the exact check named in the table.

The Baseline Check

RPD is built from the two duplicate values themselves. It does not contain an external truth baseline. Accuracy evidence needs something that can act as a reference: for example, a certified or accepted reference value, a properly prepared spike with known addition, or another suitable standard defined by the method.

Ask: What number in this evidence object represents reality independently of the pair? If there is no such number, the table may be telling you about agreement rather than truth.

The Method and Variable Check

A good duplicate pair cannot rescue a method that measures the wrong quantity. Imagine a sensor that always responds to both substance X and substance Y, but the claim says it measured only X. Repeating the measurement can produce a tiny RPD because the sensor is consistently responding to the same mixture. The duplicate result is precise, yet the method may not be selective enough for the claim.

The same logic appears throughout PSLE Science: repeating an invalid comparison does not make it valid. More careful repetition strengthens evidence only when the method itself is aligned with the question.

Alternative Explanations for a Small RPD

Suppose a duplicate pair agrees closely. Several explanations remain possible:

The method is genuinely stable and the sample is homogeneous.
The same systematic bias affected both measurements.
The duplicate was made from the same well-mixed portion, so field variability was not tested.
The instrument rounded both values to a coarse reporting resolution.
The sample concentration is high enough that a small absolute difference produces a small relative percentage.

The point is not to invent doubt forever. It is to match the strength of the conclusion to the evidence actually available.

What Evidence Would Strengthen the Accuracy Claim?

A suitable reference material with an accepted value.
Recovery evidence from a known addition where appropriate.
Calibration checks that bracket the measurement range.
Blanks that test relevant contamination routes.
Independent methods with different likely biases.
Evidence that the method is validated for the sample type and range.
Repeated quality-control results over time rather than one attractive pair.

Notice how each extra check has a different job. Science becomes stronger when several independent checks constrain different failure modes. It does not become stronger merely because the report contains more numbers.

What Evidence Would Weaken the Claim?

Reference materials repeatedly read high or low.
Blank contamination appears in the same batch.
The duplicate was not actually independent in the way the report implies.
The sample was outside the validated concentration range.
The RPD exceeds the method’s stated control limit.
The report silently averages badly disagreeing duplicates.
Another laboratory using an appropriate independent method finds a materially different result.

How Far Can the Conclusion Travel?

If one duplicate pair has an acceptable RPD, the safest conclusion is local:

These two duplicate results agreed within the stated precision criterion for this check.

Do not automatically expand that into:

the whole laboratory is accurate;
the sampling programme has no error;
every result in the batch is correct;
the product claim is scientifically proven;
the measured value is within 5% of truth.

Scientific reasoning is often the art of refusing an unnecessary upgrade in certainty.

Tempting but Invalid Reasoning

“RPD is 5%, so accuracy is 95%.” Wrong job. RPD compares duplicates, not each result with truth.
“The duplicates match, so the result must be correct.” Shared bias can make wrong results agree.
“The duplicate passed, so QA/QC passed.” One check does not stand for every quality-control check.
“A smaller RPD is always better science.” A tiny RPD cannot rescue the wrong method, wrong sample or wrong question.
“10% is the universal RPD limit.” Control limits are method- and purpose-specific.

PSLE-Style Transfer Case

A pupil tests the mass of salt in a solution using Method P. Two repeated analyses give 7.8 g and 8.0 g. The pupil says, “Because the two results are close, Method P is accurate.” Another pupil says, “The results show that Method P is repeatable under these conditions, but we need a trusted reference value before judging accuracy.”

Which response is stronger?

The second. The evidence shows agreement between the repeats. It does not supply a known correct mass. The best answer names what the evidence does support and identifies what additional evidence would be needed for the stronger claim.

Practice Set

Practice 1

Duplicate results are 25.0 and 25.2 units. There is no reference value. A report says, “The measurement is accurate.” What is missing?

Explained answer: The pair supports close agreement, but there is no independent reference showing closeness to the true or accepted value. Accuracy has not yet been established by this evidence object.

Practice 2

Duplicate results are 105 and 106. A reference material should read 100. Which two statements can both be true?

Explained answer: The method may show good precision because the duplicates agree closely, and it may show positive bias because both readings are above the reference.

Practice 3

Two laboratories each report RPD = 6%. May you conclude that they have equal data quality?

Explained answer: No. The laboratories may use different methods, sample types, ranges and control criteria. RPD is one quality indicator, not a complete score for the whole measurement system.

Delayed Independent Return

Come back to this tomorrow without rereading the article. Explain this sentence in your own words:

Two arrows can land close together and still miss the centre.

If you can connect “close together” to precision and “centre” to accuracy without using memorised wording, the reasoning has transferred.

Route to the Canonical PSLE Science Owners

Parent and Tutor Teaching Guide

Do not begin by teaching the formula. Begin with three pairs of dots on paper. Pair A lands close together but far from a target centre. Pair B spreads around the centre. Pair C is both close together and close to the centre. Ask the learner which pair is most consistent and which is closest to the target.

Only after the distinction is secure should you introduce an RPD table. Ask four questions in order: What two values are being compared? What does the percentage represent? What external reference, if any, tests accuracy? What conclusion is justified without adding certainty?

The teaching goal is not vocabulary performance. A learner has understood this article when a new percentage no longer triggers automatic subtraction from 100. Instead, the learner first identifies the scientific relationship represented by the number.

Authoritative Sources

Quiet Return

The important habit is smaller than the formula and more durable than the acronym. When a report shows RPD = 5%, do not ask, “What percentage correct is that?” Ask, “What two results are being compared, and what scientific job does this percentage actually perform?”

That one question keeps a precision check from pretending to be a truth meter.

Small Group Tutorials