How to Use All Relevant PSLE Science Repeat Results Without Cherry-Picking the Convenient Ones

Wait, What? A Result Does Not Become Invalid Just Because You Do Not Like It

A learner predicts that Set-Up A should give the largest result. Four repeats are recorded: 21, 22, 20 and 14 units. The learner circles the first three and quietly ignores 14 because it “must be wrong”.

That is not scientific evaluation. It is choosing evidence after seeing whether the evidence agrees with the prediction.

Use all relevant evidence first. Investigate unusual results second. Exclude a result only when there is an evidence-based reason to treat that measurement or trial as invalid—not because it is inconvenient.

Quick Answer

When PSLE Science gives repeated results, keep every relevant observation attached to its set-up, trial, specimen and condition. Describe the overall pattern and the variation that actually appear. If one result differs strongly, preserve it, check whether the question gives evidence of a recording mistake, method failure, wrong condition or other specific problem, and limit the conclusion if the cause is unknown. Do not choose only results that match your prediction, model answer or expected trend.

RECORD ALL RELEVANT RESULTS → CHECK PROVENANCE → DESCRIBE PATTERN AND VARIATION → INVESTIGATE UNUSUAL RESULTS → JUSTIFY ANY EXCLUSION → CONCLUDE FROM THE EVIDENCE THAT REMAINS.

The Exact PSLE Science Learning Job This Guide Owns

This guide owns one job: how a Primary 5 or Primary 6 learner uses repeated PSLE Science results as a complete evidence set instead of selecting only the repeats that support an expected answer.

It does not replace the guide on anomalous results, which focuses on diagnosing one unusual measurement. It does not replace the guide on choosing similar specimens, which prevents biased selection before data collection. This page owns the later evidence-selection problem: once relevant repeat results exist, which ones are allowed to count?

Why This Matters in the Current PSLE Science Frame

For examination from 2026, Standard PSLE Science is revised and assesses attainment in the 2023 Primary Science syllabus. The official assessment objectives include interpretation and analysis of information, evaluation of observations, information and methods, and communication of explanations and reasoning.

Scientific reasoning requires the explanation to answer to the evidence, not the evidence to be edited until it fits the explanation.

First Principle: Prediction and Result Have Different Jobs

A prediction is made from scientific knowledge before the result is known. A result records what was observed or measured. If the result does not match the prediction, the learner may need to inspect the method, the measurement, the condition or the scientific explanation.

The learner is not allowed to repair the prediction by deleting the result from memory.

The Evidence Set Comes Before the Story

Suppose five repeats under the same condition give 31, 30, 32, 21 and 31 units. The first job is not “find the four that agree”. The first job is:

all five are recorded results;
four are close to 30–32;
one is much lower at 21;
the evidence therefore contains a strong cluster and one unusual result.

Only after this description should the learner ask whether the question supplies evidence that explains the unusual result.

When Is Exclusion Scientifically Defensible?

There is an important difference between an unusual result and an invalid result.

Situation	What to do	Why
Result is surprising but the method was followed	Keep it	Surprise alone is not evidence of invalidity
Question states the instrument failed during that trial	Treat the affected reading according to the stated failure	There is an evidence-based reason
Result was copied into the wrong row and the original record is given	Use the corrected provenance	The recording error is established
Specimen was accidentally placed under the wrong condition and this is stated	Do not pretend it represents the intended condition	The trial did not test what the row claims
Result does not match the learner’s prediction	Keep it	Prediction mismatch is not a measurement failure

Do not invent hidden accidents simply to rescue a preferred trend.

Worked Example 1 — The Convenient Three

Original practice data:

Repeat	Set-Up A result
1	18 units
2	19 units
3	18 units
4	12 units

A learner wants to say A “always gives about 18–19 units”, so they omit Repeat 4. That statement is too strong. The actual evidence shows three similar results and one substantially lower result.

If no reason for the low result is given, the learner should keep it visible and make the conclusion appropriately cautious.

Worked Example 2 — Prediction Confirmation Bias

A learner predicts that increasing Condition X will increase Result Y. Five tested values mostly support the trend, but one repeat at the middle condition is lower than expected. The learner wants to plot only the readings that form a smooth line.

That would turn a real investigation into a drawing exercise. Plot or analyse the data the question actually gives. Then decide whether the unusual reading weakens confidence in a simple trend, suggests variation, or is explained by a stated method problem.

Scientific patterns are discovered from evidence; they are not tidied into existence.

Worked Example 3 — A Genuine Invalid Trial

Four trials are designed to use 100 mL of water. The question explicitly states that in Trial 3 the container leaked and only 60 mL remained before the test began.

Now there is a real reason to question whether Trial 3 represents the intended condition. A careful learner does not say, “Trial 3 is wrong because the number looks strange.” They say that Trial 3 no longer matches the planned condition because the stated water amount changed.

The reason for excluding or separately treating evidence must come from the method or data, not from the desired conclusion.

Worked Example 4 — The Quiet Danger of Choosing Only the First Results

Sometimes cherry-picking is not obvious. A learner looks at the first two repeats, sees the expected result and stops reading the table carefully. The later repeats include a different pattern.

The repair is simple: identify the complete data range before interpreting. In a table, scan every row and every condition. In a graph, identify every point or series. In prose, count every stated observation.

All Relevant Evidence Does Not Mean All Available Information

A question may contain background information that does not belong to the tested relationship. “Use all relevant evidence” does not mean force every sentence into the conclusion.

Relevant evidence must refer to the scientific object, condition, measurement or comparison needed for the question. The discipline is two-sided:

do not discard relevant inconvenient results;
do not add irrelevant information just because it is present.

The Evidence Provenance Pass

For every repeat, attach four labels:

Object: which specimen or set-up produced it?
Condition: under what tested condition?
Time: when was it observed or measured?
Method: was it produced by the intended measurement procedure?

Only after provenance is clear should the learner compare values or identify an anomaly. A result attached to the wrong condition is a different problem from a result that is unusual under the correct condition.

Failure Signature 1 — “Ignore It; It Is Probably an Error”

No evidence of error is supplied, but the learner removes the point.

Earliest weak link: confusing expectation with evidence quality.

Repair: write, “What fact shows this reading is invalid?” If the answer is “nothing”, keep it.

Failure Signature 2 — Using Only the Best-Looking Cluster

The learner chooses the three closest results out of five and treats them as the whole data set.

Repair: describe the cluster and the remaining values together. Variation is part of the evidence.

Failure Signature 3 — Deleting a Result After Seeing the Answer Key

The learner knows which conclusion a worked solution reaches and edits their reading of the data until it matches.

Repair: interpret the data first, then compare with the explanation. A model answer can help diagnose your reasoning; it cannot change the observations printed in the question.

Failure Signature 4 — Treating One Genuine Error as Permission to Ignore Anything Strange

Once a learner sees one stated instrument failure, they begin to label every unusual result as “probably faulty”.

Repair: each exclusion needs its own evidence. Scientific caution is local, not contagious.

Earliest-Weak-Link Diagnosis

Observable mistake	Earliest weak link	Repair
unexpected point is deleted	evidence discipline	preserve first, diagnose second
prediction determines which results count	prediction/result role confusion	freeze prediction before interpreting results
wrong-row value is treated as anomaly	provenance	attach result to object and condition first
one cluster hides wider spread	variation reading	scan all repeats before summarising
irrelevant background is added to compensate	relevance selection	use all relevant evidence, not all text

A Safe Four-Step Rule for Unusual Results

Preserve: keep the result in the evidence set.
Check: inspect units, labels, timing, method and stated conditions.
Explain only if supported: use a known method problem or scientific reason if the question supplies one.
Limit: if the cause remains unknown, state that the unusual result limits how strongly the pattern can be described.

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 CONDITION → STATE THE OUTCOME → CHECK AGAINST THE EVIDENCE.

The final step matters twice here. First, check the answer against the evidence. Then check whether you quietly removed any evidence because it made the answer less neat.

Original Transfer Challenge — The Uncomfortable Middle Point

An original data set shows Result R generally increasing as Condition C increases. At the middle condition, three repeats are 11, 18 and 12 units. The neighbouring conditions have tight clusters.

Do not select 11 and 12 and delete 18 merely to preserve a smooth trend. Ask whether the method gives any reason to treat 18 differently. If not, report that the middle condition shows greater variation in the available repeats. The overall relationship may still be supported, but the unusual value is part of the evidence and should influence how confidently the pattern is described.

Misconception Repair: “Science Wants a Perfect Pattern”

Real measurements can vary. A useful investigation tries to understand that variation rather than hide it. PSLE Science reasoning should respect the evidence provided in the question and avoid manufacturing a perfect straight line, perfect cluster or perfect prediction match.

Misconception Repair: “An Anomaly Is Automatically Wrong”

An anomaly is unusual relative to a pattern. It is not automatically false. It may come from natural variation, a method issue, a measurement problem, a real scientific effect, or an unknown cause. Without evidence, the learner should not choose among those possibilities as fact.

Retrieval and Practice Sequence

Take four close repeated values and one unusual value. Describe all five without explaining them.
Add a stated instrument failure to the unusual trial and explain why the evidence status changes.
Remove the failure information and practise keeping the unusual value again.
Use a prediction that the data partly contradict. Compare prediction and result without rewriting either.
Use a table with irrelevant background information and identify only the relevant evidence.
Practise one graph, one table and one prose data set.
Return after several days and interpret a fresh repeat set before seeing any suggested explanation.

Delayed Independent Return Test

Several days later, use an unfamiliar repeated-results table. Before answering any question, write:

all results that belong to each condition;
the main pattern, if any;
the amount of variation;
any result that is unusual;
the exact evidence—if any—that makes a result invalid;
one conclusion that stays within the evidence.

If you can do this before knowing which answer you hope to reach, you are practising scientific reasoning rather than answer-shaping.

Answer-Checking Receipt

I included every relevant repeat before describing the pattern.
I kept each result attached to the correct object and condition.
I did not delete an inconvenient result merely because it was surprising.
I separated an unusual result from a proven invalid result.
I used a method or recording fact to justify any exclusion.
I did not invent an accident that the question never states.
I did not let my prediction decide which observations count.
I kept variation visible.
I used relevant evidence rather than every available sentence.
I limited the conclusion when the evidence remains uncertain.

Useful Internal Routes

Parent and Tutor Teaching Guide

Create a small repeat set with one uncomfortable value. Ask the learner to describe the data before giving any prediction. Then reveal a prediction that the unusual value weakens. Watch whether the learner suddenly tries to remove it.

Next, create a second version in which a real method failure is explicitly stated. Ask what changed. The number itself did not become ugly; the evidence about how it was produced changed. That distinction is the lesson.

Finally, ask the learner to finish the sentence: “I am allowed to exclude or separately treat this result because ___.” Accept only a reason tied to the method, recording or stated condition—not “because it does not fit”.

Authoritative and Research References

The Quiet Ending

The strange result is not your enemy.

Sometimes it is the part of the evidence that teaches you the most—because it forces the explanation to stay answerable to what actually happened.

Small Group Tutorials