PSLE Science Reality Lab Vol No.155 | “This Dot Is Beyond the Box-Plot Whisker” — Does That Mean the Measurement Is Wrong?

PSLE-SCI-REALITY-0155

Wait, What? The Software Drew a Dot Outside the Whisker

A science report compares the drying times of a coating. Most results cluster between 28 and 36 minutes. One result is 49 minutes.

The report shows a box plot. The 49-minute value appears as a separate dot beyond the upper whisker.

A student points at it and says:

“The graph has marked that measurement as wrong, so we should delete it.”

The graph has not proved that.

Many box-plot conventions use a mathematical rule to flag unusually distant observations. A point beyond a whisker can tell us, “Investigate this value.” It does not automatically tell us, “This value is a mistake.”

The measurement might be a recording error. It might also be a real but unusual event, a different specimen, a changed condition, a method problem, or an important clue about the system.

Reality Lab habit: A symbol that flags an unusual observation is a request for investigation, not permission to erase inconvenient evidence.

Quick Answer

A point beyond a box-plot whisker may be flagged as an outlier under the plot’s chosen rule.
“Outlier” means unusual relative to the rest of the displayed data; it does not automatically mean incorrect.
Different box-plot conventions can place whiskers differently, so read the legend or method before interpreting the symbol.
Investigate the original observation, sample identity, unit, instrument, recording and experimental conditions.
If independent evidence shows a transcription, unit or measurement error, correction or exclusion may be justified and should be documented.
If the observation is valid, deleting it merely because it weakens a neat pattern would damage the evidence.
A verified unusual result may still be rare rather than typical.
Keep “unusual”, “wrong”, “excluded” and “important” as four different ideas.

The Exact Learner Job This Article Owns

This article owns one real-world evidence-transfer job: how a Primary 5/6 learner should interpret a box plot that marks a separate outlier point without treating the display symbol as proof that the observation is erroneous.

It does not teach box plots as a mathematics syllabus owner. It also does not replace the PSLE Science owner for handling anomalous results. Reality Lab applies that existing skill to a common scientific communication object produced by data software.

Original Reality Lab Case: The Sunleaf Drying Trial

The following data are invented for teaching.

A materials team applies the same experimental coating to twelve test tiles and records drying time under controlled laboratory conditions:

Tile	Drying time / min
1	29
2	31
3	30
4	32
5	34
6	33
7	35
8	31
9	30
10	36
11	32
12	49

Software displays the 49-minute observation as a separate circle beyond the upper whisker.

Before anyone deletes it, the team returns to the investigation record. They discover that Tile 12 was measured with the same timer and the written time really is 49 minutes. The unit is minutes, not seconds. The sample label is correct. The room record shows no obvious temperature change.

Now what?

The value remains unusual. But the checks have removed several simple error explanations. The scientific response is to keep investigating, perhaps repeat comparable tiles and inspect Tile 12’s preparation, rather than deleting the point simply because software drew it outside the whisker.

Displayed, Observed, Inferred and Claimed

Layer	Sunleaf example
Observed	Tile 12 was recorded as taking 49 minutes to dry
Displayed	The box-plot rule places 49 as a separate point beyond the whisker
Reasonable inference	49 minutes is unusually high relative to the rest of this data set under this display rule
Question to investigate	Why is Tile 12 different?
Unsupported claim	The graph proves the 49-minute measurement is wrong

What the Outlier Symbol Actually Means

One widely used box-plot variation uses the interquartile range, or IQR, to define “fences”. NIST describes a convention in which points beyond an inner fence are marked as outliers, with more extreme points beyond a farther fence.

The crucial word is rule.

The software is comparing the observation with a mathematical boundary derived from the data. The software is not travelling back in time to inspect the experiment. It does not know whether someone mistyped 49 instead of 39, whether Tile 12 had a genuine surface difference, or whether a hidden environmental condition changed.

That requires scientific investigation beyond the symbol.

First Check the Plot Convention

Not every box plot uses the same whisker convention.

Some school box plots show the minimum and maximum at the whisker ends. Some statistical software instead shortens the whiskers to values inside a particular fence and draws farther observations separately. Other specialist plots use additional rules.

Therefore the first representation question is:

What do the whiskers and separate symbols mean in this specific figure?

A learner who memorises “whisker equals maximum” can misread software that uses an outlier convention. A learner who memorises “dot equals bad measurement” can make the opposite error.

Unusual Is a Comparison, Not a Diagnosis

Imagine a class in which eleven pupils are between 145 cm and 158 cm tall and one pupil is 171 cm. The taller pupil may be unusual relative to that small group. It would be absurd to conclude that the person’s height must therefore have been measured incorrectly.

Scientific measurements work the same way. An unusual value can arise because:

the observation is genuinely rare;
the specimen differs physically from the others;
a hidden condition changed;
the measurement method failed;
a unit or transcription error occurred;
the sample came from a different source;
the system occasionally produces extreme outcomes.

The box plot alerts us to the difference. It does not choose among those explanations.

The Provenance Check: Can You Trace the Dot Back to a Real Observation?

A scientific graphic compresses the data. To investigate an unusual point, return to its provenance.

Which specimen or trial produced the value?
When was it measured?
Which instrument or observer recorded it?
What unit was used?
Was the value entered manually?
Were any method changes recorded?
Did the sample label remain correct?
Is the raw record still available?

If the separate dot cannot be traced back to the underlying observation, the graph is harder to audit.

The Unit Check: 4.9, 49 or 490?

Suppose every drying time is recorded in minutes except one value copied from a system that stores seconds. A conversion mistake can produce a dramatic outlier.

That would be independent evidence of an error. Correcting it would not be “deleting an outlier because it looks ugly”. It would be repairing a documented unit problem.

The difference matters: scientific correction needs a reason tied to the measurement record.

The Method Check: Did the Measurement Process Change?

An outlying observation can reveal a method change. Perhaps one balance was replaced, one sample was measured after a long delay, or one temperature sensor briefly lost contact.

If the unusual point coincides with a documented method change, the team has a plausible explanation to investigate. Again, the graph itself did not reveal the cause; it helped identify where to look.

The Specimen Check: Could the Difference Be Real?

Suppose Tile 12 has a microscopic patch of residue that the other tiles lack. That physical difference might genuinely slow drying.

If a repeat test on similarly prepared tiles produces several long drying times, the original “outlier” may have been the first clue to a scientifically meaningful subgroup.

This is why NIST advises investigating outliers carefully: unusual observations can contain valuable information about the process or about the data-gathering procedure.

The Repetition Check: Repeat the Question, Not the Desired Answer

Repeating a comparable measurement can help distinguish a one-off recording problem from a recurring phenomenon. But repetition must be designed fairly.

If Tile 12 is simply measured again and the first value is discarded whenever the second is more convenient, the procedure becomes selective.

A stronger approach records both results, checks whether the repeated method is comparable and explains why any value is rejected.

Deleting a Point Can Change the Story

Imagine a claim that a coating always dries in under 40 minutes. Eleven values support that statement; the 49-minute observation contradicts it.

If the researcher deletes 49 merely because it is inconvenient, the cleaned graph may falsely suggest perfect consistency.

If independent evidence proves 49 came from a broken timer, excluding or correcting it can be appropriate.

The visible action—removing a point—can therefore be scientifically responsible or scientifically misleading. What matters is the reason and documentation.

Worked Case 1: The Decimal-Point Error

A box plot shows measurements near 2.1, 2.3, 2.4 and 2.5, plus one point at 24.0. The original notebook clearly says 2.40 and the spreadsheet entry says 24.0.

Here there is direct evidence of transcription error. Correcting the spreadsheet is justified. The reason is not “the point was outside the whisker”; the reason is “the stored value disagreed with the original record”.

Worked Case 2: The Rare but Real Specimen

Most seed pods contain 8–12 seeds. One intact pod contains 21, and recounting confirms 21. A photograph of the opened pod and a second observer support the count.

The value may remain unusual. It is not automatically invalid. The correct next question is how often such pods occur and what conditions might explain them.

Worked Case 3: The Wrong Unit

Eleven masses are entered in grams. One is entered as 0.031 because the instrument exported kilograms. Software flags the value.

The audit trail shows the unit mismatch. Converting the measurement properly is scientifically justified. The outlier symbol helped reveal the problem but did not, by itself, prove it.

Worked Case 4: The Changed Condition

One unusually slow evaporation trial occurred after the room’s humidity-control system failed. The maintenance record confirms the failure.

The point may not answer the intended “normal humidity” question, but it could be useful evidence about a different condition. A transparent report might analyse it separately rather than silently erase it.

Worked Case 5: The Outlier Disappears When More Data Arrive

With only eight observations, a value of 20 appears isolated. Twenty more observations are later collected, including 18, 19, 20, 21 and 22.

The context has changed. What looked unusual in the smaller data set may no longer be unusual in the larger one. Outlier status is relative to the distribution and the rule being used.

Worked Case 6: The Sensor Saturation Point

Most readings are between 70 and 80 units, but one value is exactly 100, the instrument’s maximum display. The graph flags it as unusual.

The important clue is not merely the outlier symbol. The value equals the instrument ceiling. That suggests a measurement-limit problem worth investigating. The true underlying value might even have been above 100.

Worked Case 7: The Tempting Deletion

A group wants to claim its new method reduces time in every trial. One result shows the opposite and is plotted as an outlier. There is no documented error.

Deleting it because it weakens the claim would be poor evidence practice. Keep it, investigate it and let the conclusion acknowledge variation.

What Evidence Strengthens a Decision to Correct or Exclude a Point?

A raw record proving a transcription error.
A confirmed unit mismatch.
A documented instrument malfunction.
A sample-identity error.
A method deviation that makes the observation inapplicable to the intended comparison.
A pre-specified exclusion rule applied consistently.
Independent records supporting the reason for exclusion.
Transparent reporting of what changed and why.

What Weakens It?

“It looked too far away.”
“The software put a dot around it.”
“It ruined our average.”
“The trend became nicer after deleting it.”
“We repeated only the strange result until we got a normal one.”
“We cannot find the original record, but we removed it anyway.”

Tempting Reasoning That Fails

“Outlier means error.” An outlier is an unusual observation under a rule; error needs separate evidence.
“If it is real, it cannot be an outlier.” Real values can be rare.
“If it is an outlier, it should always be deleted.” Deletion requires scientific justification.
“If a repeat is normal, the first value must have been wrong.” Variation can produce different valid results.
“If a point is inside the whiskers, it must be correct.” Ordinary-looking values can also contain errors.
“Box plots are misleading because they flag points.” No. The flag is useful when its meaning is understood.

How Far Can the Conclusion Travel?

A separate box-plot point can support:

This observation is unusually distant from the central portion of these data under the plot’s stated rule and deserves investigation.

It cannot, by itself, establish:

the observation is a recording error;
the instrument failed;
the specimen is impossible;
the point should be deleted;
the unusual value is typical of the wider population;
the experiment’s main claim is false or true.

PSLE-Style Transfer Case

A student measures the cooling time of 15 identical-looking containers. A software box plot displays one 27-minute result beyond the upper whisker. The student removes it and writes, “The computer showed that 27 minutes was wrong.”

Explain why this is not justified.

Answer: The box plot only shows that 27 minutes is unusually high relative to the other measurements under the plot’s outlier rule. The student should check the raw record, unit, container identity, method and measuring conditions. The value should not be removed unless there is independent evidence that it is invalid or a justified exclusion rule applies.

Suppose the notebook shows the timer actually read 17 minutes and 27 was typed into the spreadsheet. What changes?

Now there is direct evidence of a transcription error, so correcting the spreadsheet to match the original record is justified.

Explained Practice

Practice A: A point sits outside a whisker. What is the first safe description? It is unusually distant under this box-plot convention.

Practice B: The raw notebook confirms the value and the unit. Is the point automatically wrong? No.

Practice C: The original notebook proves a typing error. Can the database be corrected? Yes, with the correction documented.

Practice D: Why check the figure legend? Whisker and outlier conventions can differ.

Practice E: A valid unusual value is confirmed. Does it become the general rule? No. Validity and typicality are different questions.

Delayed Independent Return: F-L-A-G

When a graph marks an unusual point, try four questions:

F — Figure rule: What does the symbol mean in this plot?
L — Link: Can the symbol be linked back to the raw observation?
A — Alternative explanations: Error, changed condition, rare specimen, or something else?
G — Grounds: What independent grounds justify correction, exclusion or retention?

The letters are a memory aid, not an examination requirement.

Parent and Tutor Teaching Guide

Write nine numbers close together on cards and one noticeably larger number. Ask the learner to identify the unusual card. Then ask a different question: “Which card is wrong?” The learner should realise that the list alone cannot answer the second question.

Next secretly create one genuine recording error by copying 14 as 41. Give the learner the original note as independent evidence. Now correction is justified. Contrast that with a second unusual value that exactly matches its original record.

The teaching goal is not calculation of quartile fences. It is the deeper distinction between a statistical flag and a scientific diagnosis.

Authoritative Sources

NIST notes that outliers should be investigated carefully and can contain valuable information about the process or about data gathering and recording. SEAB’s 2026 PSLE Science objectives include interpreting and analysing information and evaluating observations, information and methods. MOE’s syllabus advocates healthy scepticism rather than automatic acceptance or automatic rejection.

The Quiet Return

The dot sits outside the whisker.

That is not the end of the scientific story.

It is where the story becomes interesting enough to check.

Flag the unusual result. Trace it. Test explanations. Do not let a plotting symbol decide the truth for you.

Small Group Tutorials