How to Handle an Anomalous PSLE Science Result Without Deleting It Just Because It Looks Wrong

Wait, What? The strange result may be the most important result in the table.

You repeat a PSLE Science investigation several times. Most readings are close together. One result is very different. It looks awkward. It breaks the pattern. It makes the graph less neat.

The tempting move is to cross it out.

Do not do that merely because the result looks wrong.

An unusual result can come from several different causes. A measurement may have been recorded incorrectly. The method may have been disturbed. One specimen may genuinely differ from the others. A controlled condition may have changed. The measuring instrument may have reached a limit. Or the result may reveal that the scientific relationship is more complicated than you expected.

Until you have evidence for what happened, “I do not like this number” is not a scientific reason to remove it.

Quick Answer

When one PSLE Science result looks anomalous, use this sequence:

KEEP: preserve the original result rather than silently deleting or changing it.
CHECK: verify the label, unit, time point, specimen, transcription and instrument reading.
COMPARE: look at the other results produced under the same intended condition.
INSPECT THE METHOD: ask whether anything was different during that run.
VERIFY: if you are designing or evaluating an investigation, repeat the appropriate test or measurement for a scientific reason.
EXPLAIN CAREFULLY: distinguish a documented error from a possible cause.
LIMIT THE CLAIM: make the conclusion no stronger than the full evidence allows.

A compact rule is:

UNUSUAL ≠ WRONG. CHECK BEFORE YOU REJECT.

The Exact PSLE Science Learning Job This Guide Owns

This guide owns one specific learner job: how to handle a single suspicious or anomalous PSLE Science result without cherry-picking the evidence.

Several nearby guides own different jobs:

Reasoning from unexpected experimental results deals with what to do when the overall outcome does not match your expectation.
Reading repeated results that do not match exactly deals with ordinary variation between repeated measurements.
Random variation versus a systematic shift deals with distinguishing scattered variation from a consistent bias or change.
Healthy scepticism in PSLE Science deals with questioning evidence without distrusting everything.

This page is narrower: one result stands out, and the learner must decide how to treat it without deleting inconvenient evidence or pretending to know its cause.

Why This Matters in the Current PSLE Science Frame

For the 2026 PSLE, SEAB identifies Science as revised and the official Science syllabus states that the examination assesses the 2023 Primary Science syllabus. The assessment objectives include applying scientific knowledge and scientific inquiry, including interpreting and analysing information, evaluating observations, information and methods, and communicating explanations and reasoning.

That means a student is not learning Science well if the method is simply:

“Keep the numbers that match my expectation and remove the one that does not.”

The evidence has to be read first. The method has to be checked. The conclusion has to remain answerable to what was actually observed.

What Does “Anomalous” Mean Here?

For this Primary 5/6 learning guide, an anomalous result means a result that stands out from the other relevant results or does not fit the pattern you expected.

We are not introducing a formal statistical rule for deciding outliers. A PSLE Science question may not provide enough data for that. The useful student job is more basic and more important:

notice that a result is unusual;
keep it visible;
check how it was produced;
compare it with the rest of the evidence;
avoid inventing a cause;
avoid pretending it never existed.

Anomalous Does Not Mean Incorrect

This is the most important distinction in the whole guide.

A result can be unusual and still be a genuine observation.

A result can also be unusual because something went wrong.

The appearance of the number alone does not tell you which situation you have.

Possible reason for the unusual result	What a learner should do
Real natural variation	Keep the result and recognise that similar specimens or repeated runs may not be identical.
Uncontrolled condition changed	Identify the specific condition that differed and explain how that could weaken the comparison.
Instrument or measurement limitation	Check range, resolution, reading method and whether the measurement could represent the true value adequately.
Recording or transcription error	Check the original observation or record. Correct only when there is evidence of the recording mistake.
Method disturbed the system	Identify the action that may have altered the setup and propose a method repair.
Genuine change in the relationship	Do not force the data into the expected trend. Consider whether a threshold, turning point or changing condition could matter.
Cause unknown	Say that the reason is not established. Preserve the result and limit the conclusion.

Worked Example 1: 12, 13, 12, 25

Imagine an original practice investigation in which the same outcome is measured in four independent runs under the same intended condition.

Trial	Measured result
1	12 units
2	13 units
3	12 units
4	25 units

The fourth result stands out.

A weak response is:

25 must be wrong, so remove it.

That is not yet scientific reasoning. Instead ask:

Was 25 copied correctly from the original reading?
Was the unit the same?
Was the same instrument used?
Was the instrument read from the correct scale?
Did anything happen differently during Trial 4?
Was the starting condition actually the same?
Could the system genuinely vary between runs?

If a notebook clearly shows that the instrument read 15 but the table was copied as 25, you have evidence of a transcription error. That is different from deleting 25 because it looks inconvenient.

If no error is found, the unusual result remains part of the evidence. If you are designing the investigation, another properly controlled independent trial may help you understand whether 25 was a one-off variation or part of a wider pattern.

Worked Example 2: One Seed Does Not Germinate

Imagine several similar seeds placed under the same intended conditions. Most germinate. One does not.

A learner says:

Ignore that seed because it was probably bad.

The phrase “probably bad” is an inference, not an observation. Perhaps the seed was not viable. Perhaps it experienced a slightly different local condition. Perhaps there was natural variation. Perhaps something else happened.

The important method question is whether the criteria for choosing similar specimens were defined before seeing the outcomes. If a learner removes a specimen only after it gives an inconvenient result, the evidence has been cherry-picked.

A better conclusion recognises the observed variation and avoids claiming that every similar seed must respond identically.

Worked Example 3: One Point Breaks a Graph Trend

Imagine a graph in which the measured outcome generally rises as the tested condition increases, except for one point that is lower than expected.

Do not redraw the point to make the line smoother.

Ask what the evidence actually shows:

Does the overall relationship still appear across most tested values?
Is the unusual point within a region where the relationship might be changing?
Was that condition tested more than once?
Is there information about the method or measurement that explains the difference?
Does the graph provide enough evidence to decide why the point differs?

If the reason is not given, do not invent one. Describe the pattern honestly and keep the irregular point visible.

This links to the separate guides on turning points, thresholds and plateaus. A point that initially looks wrong can sometimes be a clue that the relationship is not a simple straight-line story.

Worked Example 4: One Anomaly Versus a Whole Unexpected Pattern

Suppose a learner predicts that increasing a tested condition will increase an outcome.

Case A: five results generally follow the expected pattern and one point differs greatly.

Case B: every result shows the opposite direction from the prediction.

These are not the same problem.

Case A asks how to handle a suspicious individual result inside a broader pattern.
Case B asks whether the prediction, concept, condition or method needs re-examination because the overall evidence disagrees.

Calling all unexpected evidence “anomalous” can hide an important scientific difference.

Do Not Delete a Result Merely to Improve the Average

One unusual value can strongly affect an average. That does not give you permission to remove it automatically.

Before using an average, ask what the values represent and why they are being combined. The separate guide on reading an average without treating it as every trial or specimen develops that distinction.

For Primary-level learning, do not invent a statistical “outlier rule” unless the question gives one. The scientific job is to preserve the data, examine the method and make a conclusion that respects the uncertainty.

When Can a Value Be Corrected?

A value can be corrected when there is evidence that the recorded value is not the observation that was actually made.

For example:

the original instrument reading was 18 cm but someone copied it as 81 cm;
the value was entered in grams while the rest of the table is in kilograms and the original record confirms the conversion error;
a label was attached to the wrong specimen and the method record clearly identifies the mix-up.

Notice the difference. The correction is supported by a record or method check. It is not based on “this number ruins my pattern”.

When Should the Original Result Stay?

If you cannot establish that a recording or measurement error occurred, keep the original observation in the evidence set.

You can still say that it is unusual. You can still propose a repeat or method check. You can still limit the conclusion. What you should not do is silently rewrite the world to match the expected answer.

Anomaly, Random Variation and Systematic Shift

Pattern	What it looks like	What to investigate
Anomalous result	One or a small number of points stand apart from the relevant results	Recording, method, specimen, local condition, measurement or genuine unusual response
Random variation	Repeated values scatter around a general level without one consistent direction of shift	Natural variability, measurement resolution, repeatability and sample size
Systematic shift	Many or all values are displaced in a similar direction	Calibration, common method bias, changed condition or consistent procedural difference
Changing relationship	The pattern itself bends, turns, plateaus or changes across conditions	Whether the scientific system behaves differently across the tested range

One strange point should not automatically be called random variation. One strange point should not automatically be called an error either. First describe the pattern you actually have.

The Evidence-Integrity Protocol

Use this protocol when an investigation contains an unusual result:

IDENTIFY THE EVIDENCE UNIT: Is this one measurement, one trial, one specimen, one time point or one test condition?
CHECK IDENTITY: Is the result attached to the correct setup, specimen and condition?
CHECK THE RECORD: Verify number, unit, decimal place, label and transcription where possible.
CHECK MEASUREMENT: Was the instrument suitable? Was the reading within range? Was the same method used?
CHECK THE STARTING STATE: Was this run genuinely comparable with the others?
CHECK CONTROLLED CONDITIONS: Did a condition quietly differ?
COMPARE REPEATS: Is the result isolated or part of wider variation?
REPEAT APPROPRIATELY: If designing the investigation, perform a justified repeat rather than merely repeating everything blindly.
PRESERVE THE ORIGINAL: Keep the first result visible unless there is evidence of a specific recording correction.
LIMIT THE CONCLUSION: State only what the complete evidence supports.

How This Fits the PSLE Science Reasoning Chain

OBSERVE / READ GIVEN INFORMATION → IDENTIFY THE SCIENTIFIC OBJECT OR RELATIONSHIP → DISTINGUISH OBSERVATION FROM INFERENCE → SELECT THE RELEVANT CONCEPT → EXPLAIN THE CAUSAL MECHANISM → CONNECT TO THE QUESTION’S CONDITION → STATE THE OUTCOME → CHECK AGAINST THE EVIDENCE.

An anomaly makes the final step especially important. If the evidence is inconsistent, the conclusion may need to become narrower, more cautious or conditional.

What the Data Can and Cannot Support

Suppose four out of five results suggest one relationship and one result does not.

You may be able to say that the results generally show a relationship across the tested conditions. You may not be able to say that the relationship happened without exception in every trial.

You also may not be able to say why the unusual result occurred unless the question provides evidence for the cause.

This distinction protects three different claims:

Pattern claim: what the results generally show.
Anomaly claim: which result does not fit that pattern.
Cause claim: why the anomaly happened.

The first two may be visible in the data. The third often needs additional evidence.

What to Do in an Examination Question

If a PSLE Science question gives you a table or graph, treat the supplied values as evidence. Do not alter the data to fit your expectation.

If the question asks you to:

describe the results: include the unusual result if it matters to the pattern;
identify an anomalous result: state which result stands out and compare it with the relevant neighbouring or repeated values;
suggest a reason: give a scientifically plausible possibility without turning it into established fact unless evidence supports it;
improve the method: repair the specific weakness, such as measurement consistency, controlled conditions or justified repetition;
draw a conclusion: keep the conclusion within the full evidence rather than ignoring the inconvenient point.

Do not invent a universal marking phrase. Read the exact command and answer the scientific job requested.

A Better Way to “Repeat the Experiment”

“Repeat the experiment” is useful only when you can say what the repeat is meant to find out.

For an anomalous result, a repeat might help you ask:

Does the unusual value occur again under the same intended condition?
Do repeated results cluster around the original pattern or around the anomalous value?
Was the unusual value linked to one specimen?
Does changing or controlling one suspected condition remove the difference?

That is stronger than repeating without a question.

Do Not Confuse a Surprising Result With a Bad Method

A method can be well designed and still produce variable results. A result can be surprising even when no obvious mistake occurred.

Conversely, a neat pattern does not prove that the method was good. A systematic error can produce very consistent but misleading measurements.

Therefore, evaluate method quality from the design and evidence production—not from whether the graph looks pretty.

Failure Signatures You Can Observe

The learner deletes the largest or smallest number because it “spoils the pattern”.
The learner says “human error” without naming any specific action that could have affected the result.
The learner changes an unusual value to something closer to the others without evidence.
The learner averages all values immediately before deciding what they represent.
The learner removes a specimen after seeing an inconvenient outcome.
The learner assumes every unexpected result proves the method was wrong.
The learner assumes one anomaly disproves every general pattern in the data.
The learner invents a cause for the anomaly when the question provides no discriminating evidence.
The learner ignores the anomaly in the conclusion even though it limits the strength of the claim.

Earliest Weak-Link Diagnosis

When a learner mishandles anomalous data, locate the earliest failure:

Evidence-recognition failure: the learner does not notice that a point is unusual.
Evidence-integrity failure: the learner believes inconvenient data may be discarded automatically.
Identity failure: the result is attached to the wrong specimen, trial, time or condition.
Measurement failure: instrument range, resolution or reading procedure is not considered.
Method failure: a changed controlled condition or carryover effect is missed.
Inference failure: a possible reason is presented as a known cause.
Conclusion failure: the final claim is stronger than the complete evidence permits.

Repair from the first weak link instead of giving the child a sentence to memorise.

Misconception Repair

Misconception	Better scientific model
“An anomaly is a wrong result.”	An anomaly is an unusual result whose cause may or may not be known.
“Delete it before averaging.”	Preserve the evidence first. Decide how to summarise only after understanding what the values represent and why the result differs.
“Human error explains it.”	Name a specific possible procedural or recording problem and explain how it could change the result.
“One strange point means there is no relationship.”	Describe both the overall pattern and the exception. Judge claim strength from all evidence.
“A neat graph proves the method is reliable.”	Consistency is only one feature. A shared systematic problem can create a neat but biased pattern.
“If I expected a different answer, the data must be wrong.”	Prediction and result have different jobs. Unexpected evidence is a reason to re-check the model and method.

Retrieval and Practice Sequence

Practise the reasoning, not the definition.

Create a small original table with four similar values and one unusual value.
Without deleting anything, describe what is directly observed.
List three possible reasons for the unusual value and mark each as a possibility, not a fact.
Choose one method check that could distinguish between two of those possibilities.
Write a conclusion that includes the evidence limit.
Change the context from a physical measurement to a biological specimen or a graph.
Repeat the process without looking at this guide.

Unfamiliar Transfer Test

Transfer the skill across different representations:

one unusual bar in a graph;
one strange value in repeated measurements;
one specimen with a different response;
one time point that reverses a trend;
one setup whose result disagrees with a prediction.

You have learned the skill when your first move is no longer “remove it”, but “trace how this evidence was produced”.

Delayed Independent Return Test

After several days, use a fresh question containing one odd result. Without prompts, answer these questions:

Which result is unusual, and relative to what comparison?
What exactly was measured or observed?
Can I verify the record or unit?
Was the same method and starting condition used?
What possible causes remain?
Which cause, if any, is actually supported?
What additional check would be useful?
How should the anomaly change the strength of my conclusion?

If the learner can do this independently in a changed context, the repair has moved beyond a memorised phrase.

Answer and Checking Receipt

Evidence: I kept the original unusual result visible.
Identity: I know which trial, specimen, time point or condition produced it.
Measurement: I checked the unit, instrument and reading method where relevant.
Method: I checked whether a controlled condition or starting state differed.
Inference: I separated possible reasons from established causes.
Repeat: if I proposed repetition, I explained what the repeat would test.
Conclusion: I used all the evidence and did not hide the inconvenient result.

Parent and Tutor Teaching Guide

When a child wants to cross out an unusual result, ask one question first:

“What evidence tells you that the result is wrong rather than merely unusual?”

If the answer is “because the other numbers are different”, the distinction has not yet been learned.

Use a simple teaching routine:

Show four ordinary results and one unusual result.
Ask the learner to describe only what is observed.
Generate several possible explanations.
For each explanation, ask what evidence would support or weaken it.
Decide what method check would help.
Write a cautious conclusion using the whole data set.

Avoid teaching “remove anomalies” as a generic rule. Teach evidence provenance: where did the value come from, what could have affected it, and what can we honestly conclude?

Also avoid treating the phrase “human error” as an explanation. Ask for the action: Was the wrong scale read? Was timing started late? Was a specimen labelled incorrectly? Was the setup disturbed during measurement? Specific mechanisms can be evaluated; vague labels cannot.

Useful Internal Routes

Authoritative References and Evidence Boundaries

The National Academies material is general science-education and scientific-evidence context, not a PSLE marking rule. The examples in this article are original and do not reproduce national examination questions. The KEEP–CHECK–COMPARE–VERIFY routine and checking receipt are eduKate learning tools, not official SEAB answer templates.

Quiet Return

Science becomes trustworthy when the evidence is allowed to surprise us.

A strange result is not automatically a mistake. It is a question: What happened here? Sometimes the answer is a recording error. Sometimes it is a method weakness. Sometimes it is natural variation. Sometimes it tells you that your first model of the system was too simple.

Keep the result. Check how it was made. Follow the evidence. Then make the conclusion exactly as strong as the world in front of you allows.

Small Group Tutorials