PSLE Science Reality Lab Vol No.070 | “Three Sensors Agree” — Can They All Share the Same Bias?

PSLE-SCI-REALITY-0070

Wait, What? Three thermometers can show almost the same temperature and all three can still be wrong in the same direction.

Imagine three digital temperature sensors sitting side by side. One reads 24.7°C. The second reads 24.8°C. The third reads 24.7°C.

That looks convincing. Three devices agree. Surely the room must be about 24.7°C.

Now imagine that all three sensors came from the same production batch, were adjusted using the same incorrect reference, and are each about 0.8°C low. A carefully checked reference instrument placed beside them reads 25.5°C.

The three sensors still agree with one another. What changed is our understanding of what that agreement proves.

This is an important scientific habit: agreement among measurements is evidence about consistency, but agreement with an appropriate reference is a different question.

Reality Lab Vol No.070 teaches you how to evaluate a public claim such as “three sensors confirmed the result” without being impressed by the number three alone. The real job is to ask whether the checks are genuinely independent and whether they can reveal a shared error.

Quick Answer

When several sensors agree, ask four questions:

What exactly agrees? Are the readings close to one another?
What are they being checked against? Is there an appropriate reference value or independently checked instrument?
What do the sensors share? Same calibration source, same model, same placement, same software correction, same environment or same preparation?
Could the shared feature move all readings in the same direction? If yes, agreement does not rule out a shared bias.

Reality Lab rule: Several agreeing measurements are not automatically several independent checks.

What This Guide Owns — and What It Does Not

This guide owns one real-world evidence-transfer job: how to evaluate a claim that multiple sensors, instruments or measuring devices “agree” when those devices may share the same source of measurement bias.

It does not re-teach the full PSLE Science micro-skills of precision, accuracy, random variation, systematic shift or instrument checking. Those jobs already have canonical owners in the eduKateSengkang Science estate. Use this Reality Lab as the transfer layer, then route back to the underlying skills when you need them:

The new learner job here is not “define accuracy.” It is: when a real-world claim says several devices agree, decide whether the agreement supplies genuinely new evidence or repeats the same dependency.

The Original Reality Lab Case: The Three Greenhouse Sensors

A school greenhouse team wants to check the air temperature near three trays of seedlings. They attach three identical digital sensors to a horizontal rail, only a few centimetres apart. After the readings settle, the display shows:

Sensor	Displayed temperature
A	28.1°C
B	28.2°C
C	28.1°C

A poster is prepared: “Three sensors independently confirm that the greenhouse temperature is 28.1°C.”

Before accepting that sentence, inspect the evidence object more closely.

All three sensors are the same model.
They were bought together.
They were adjusted using the same reference thermometer.
They are mounted on the same rail.
The rail is close to a warm lamp.
The reference thermometer used for adjustment is later discovered to have been reading 0.6°C low.

Now the phrase “independently confirm” is too strong. The devices are separate physical objects, but their errors are not necessarily independent. A shared reference can transmit the same offset to all three. A shared warm location can also make all three measure the same local condition rather than the greenhouse as a whole.

Observed, Claimed and Inferred

Layer	What we can say
Observed	The three displayed readings are 28.1°C, 28.2°C and 28.1°C at the recorded time.
Claimed	“Three sensors independently confirm 28.1°C.”
Inferred	The sensors are accurate, independent and representative of the greenhouse.

The observation is real. The problem is that the claim adds properties the display alone does not prove.

This is exactly the kind of move PSLE Science reasoning helps you notice. You separate what the evidence directly shows from the extra conclusion someone wants you to carry away.

Agreement Has More Than One Meaning

“The sensors agree” can mean different things. A careful reader asks which one is intended.

Kind of agreement	Question	What it can support
Reading-to-reading agreement	Are the displayed values close?	Consistency at that moment.
Repeat agreement	Does the same instrument give similar values repeatedly?	Repeatability under those conditions.
Instrument-to-instrument agreement	Do different devices give similar values?	Agreement across devices, if the comparison is fair.
Reference agreement	Do the readings agree with an appropriate checked reference?	Evidence about measurement correctness relative to that reference.

The first three do not automatically create the fourth.

Why Shared Bias Is Different From Random Scatter

Suppose three rulers each have a worn zero end. You line an object up with the damaged edge and measure from there. The three readings may agree because the three rulers have the same kind of problem.

Taking more readings can help you see ordinary scatter. If your hand position changes a little each time, repeated measurements may vary above and below a central value. Averaging can sometimes reduce the influence of that random variation.

A shared systematic shift behaves differently. If every reading is moved upward by the same cause, collecting more of the same kind of reading does not guarantee the shift disappears. You can become very consistent about the wrong value.

NIST measurement guidance makes this distinction important: measurement results can contain random and systematic effects, and checking against standards or reference values is one way bias can be identified. The useful Primary Science version is simple: close agreement tells you one thing; closeness to a suitable reference tells you another.

The Dependency Map: What Do the Sensors Share?

Before counting “three sensors” as three strong confirmations, draw a dependency map. Ask whether the devices share any of these:

Calibration source: all adjusted using one reference instrument;
Manufacturing batch: a shared production issue could affect all units;
Measurement principle: the same method may respond similarly to the same interference;
Position: all devices may be measuring one unusual local spot;
Environment: sunlight, vibration, humidity or airflow may affect all devices together;
Software: one correction rule or conversion formula may be applied to every reading;
Operator: the same person may make the same reading or setup mistake each time;
Reference data: all devices or analyses may ultimately depend on the same external value.

Shared features are not automatically defects. Three identical sensors may be exactly what an investigation needs. The point is narrower: if you are using agreement as evidence of correctness, you must ask whether the agreement could have been produced by a shared dependency.

A Useful Ladder of Evidence

Think of the following as an evidence ladder, not a magic scoring system.

One sensor, one reading. Tells you what that sensor displayed.
One sensor, repeated readings. Adds evidence about repeatability.
Several sensors of the same kind. Adds instrument-to-instrument agreement, but may retain common dependencies.
Different instruments or methods. Can reduce some shared dependencies if they genuinely differ in relevant ways.
Comparison with an appropriate reference. Directly tests whether the measurement agrees with a value whose role is to check the process.

Higher on this ladder does not mean “perfect.” A reference itself has uncertainty. A different instrument can have its own problems. Science improves evidence by making dependencies visible, not by pretending any check is absolute.

Worked Case 1: Three Kitchen Scales

Three kitchen scales measure the same sealed bag. They show 502 g, 501 g and 502 g. A claim says: “Three scales prove the bag has a mass of 502 g.”

Better reasoning:

The scales agree closely.
That supports consistency among those scales for that measurement.
It does not by itself establish that 502 g is the best reference value.
Check whether the scales were zeroed correctly and whether a suitable known mass gives the expected reading.

If all three scales have the same incorrect zero setting, their agreement can repeat the same shift.

Worked Case 2: Three Light Sensors Under One Shadow

Three light sensors placed together report nearly identical brightness. The poster says: “Three sensors show the whole classroom has the same brightness.”

The problem is no longer mainly calibration. It is representativeness. Three devices in one place can all measure the same local patch accurately while failing to represent other parts of the room.

This shows why a scientific claim can fail at more than one level. Agreement may be genuine, but the conclusion may travel too far.

Worked Case 3: Two Sensors and a Reference Check

Sensor A reads 19.8°C. Sensor B reads 19.9°C. A checked reference instrument reads 20.0°C. The three readings are close.

This is stronger than simple A-versus-B agreement because the comparison includes a reference with a different evidential job. You still preserve uncertainty and the conditions of the check, but you have tested the shared-error possibility more directly.

Worked Case 4: Three Identical Sensors and One Shared Formula

Imagine three electronic sensors measure a raw signal correctly. Their software then converts that signal into a displayed value using the same formula. If the formula contains the same incorrect constant, all three final displays can agree because they share the same calculation.

The physical sensors are separate. The evidence chain is not fully separate.

Count evidence chains, not just devices.

The Reference Check Is Not a Magic Oracle

A sensible reference improves the test, but do not replace one overclaim with another. Reference values are themselves established through measurement systems and have uncertainty. NIST explicitly notes that reference values are not literally error-free in real work.

For a Primary 5/6 learner, the important lesson is not advanced metrology. It is this:

Choose a reference because it has a clear checking role.
Know what it can check.
Keep the conclusion proportional to that check.

A known 100 g classroom mass can check whether a balance reading is sensible around that range. It does not prove the instrument behaves perfectly for every mass, environment and future date.

What Evidence Would Strengthen “Three Sensors Confirm It”?

A clear statement of what quantity each sensor measured.
Individual readings rather than only “all agreed.”
Information about calibration or a reference check.
Evidence that the sensors were not all affected by the same unsuitable placement.
A check using a genuinely different instrument or method when that difference helps reveal shared bias.
Repeated checks across relevant conditions, not only one convenient moment.
Disclosure of corrections or shared software processing.
A conclusion limited to the place, time and conditions actually measured.

What Would Weaken It?

Three devices adjusted from the same questionable reference.
Three readings from one instrument presented as “three independent confirmations.”
Several devices clustered in one unusual location but used to describe a whole area.
Agreement reported without the actual values.
Identical processing applied to every instrument with no independent check.
“They agree, therefore they are accurate” with no reference or other evidence.

How Far Can the Conclusion Travel?

Suppose three sensors agree at 2:00 p.m. in one place. What can you safely conclude?

You may be able to say that the three sensors produced similar readings under those conditions at that time. If they were checked against a suitable reference, you may have further evidence about the quality of those readings.

You cannot automatically conclude that:

the entire room has that value;
the sensors will remain correct next month;
another sensor type would agree;
the measuring method has no systematic error;
the reading is exact to every displayed decimal place.

Good scientific communication does not make evidence smaller than it is. It also does not make it bigger.

PSLE-Style Transfer Case

A student uses three identical temperature probes to measure water in the same beaker. The probes show 42.1°C, 42.2°C and 42.1°C. The student writes:

“The water is definitely 42.1°C because three probes gave almost the same answer.”

A stronger evaluation is:

The similar readings show that the probes agree closely, but agreement alone does not prove the temperature is exactly 42.1°C. The probes should also be checked using an appropriate reference or another suitable method, because a shared measurement bias could shift all three readings in the same direction.

Notice what this answer does not do. It does not accuse the instruments of being wrong. It identifies what the existing evidence supports and what extra check would test the remaining uncertainty.

Tempting Reasoning That Fails

“Three devices cannot all be wrong.” They can share a calibration, method or environmental bias.
“The readings are close, so the true value must be inside that tiny range.” A shared shift can move the entire cluster.
“Different serial numbers mean independent evidence.” Physical separation does not guarantee independent calibration or processing.
“If the reference disagrees, the reference must be wrong because three beats one.” Evidence quality is not a vote count. Investigate the measurement chains.
“A reference instrument is perfectly correct.” References also have uncertainty and scope.
“More sensors always fix the problem.” More of the same dependency can produce more confidence without removing the dependency.

Practice 1: The Shared Zero

Three digital balances are set to zero while an empty tray is already sitting on each one. A student later removes the trays and weighs identical 50 g reference masses. All three balances display about 45 g.

Question: Does the agreement among the balances show that 45 g is correct?

Answer: No. The balances agree, but the known reference provides evidence that the shared setup or zeroing procedure introduced a systematic shift.

Practice 2: Same Place, Same Air

Four temperature sensors in one corner of a classroom all read 27.0–27.2°C. Can the class conclude that every part of the classroom is about 27.1°C?

Answer: Not from those readings alone. The sensors provide repeated information about one location. Other locations must be sampled if the claim is about the whole classroom.

Practice 3: Different Method

Two electronic probes agree closely. A separately checked glass thermometer also gives a similar result under the same conditions. Why can that extra check be useful?

Answer: A different measurement chain may not share every dependency of the electronic probes. Agreement across appropriately different methods can test some shared-error possibilities, although it still does not make the result uncertainty-free.

Practice 4: What Would Change Your Mind?

A product demonstration says, “Five sensors all recorded the same improvement.” Name one piece of information that would strengthen the claim and one that would weaken it.

Possible strengthening evidence: the sensors were checked against a suitable reference and placed to measure genuinely separate relevant locations.

Possible weakening evidence: all five were corrected using the same faulty reference or all five occupied one unrepresentative location.

Delayed Independent Return: The Four-Box Audit

Later, when you encounter a fresh claim involving several sensors, draw four boxes without looking back at this page:

READINGS — What values were actually recorded?
REFERENCE — What, if anything, checks correctness?
SHARED — What dependencies do the measurements share?
SCOPE — How far can the agreement travel?

If you can fill those four boxes, the phrase “three sensors agree” stops being a conclusion and becomes the beginning of an investigation.

Teaching Guide for Parents and Tutors

You can teach this idea without special laboratory equipment. Use three rulers, three household thermometers or three simple measuring containers. The aim is not to create a perfect calibration exercise. The aim is to make dependency visible.

Start by asking the learner to predict: “If three tools agree, what does that prove?” Do not correct immediately. Let the learner state the strongest claim they think is justified.

Then introduce one shared dependency. For example, measure lengths from a deliberately displaced starting line printed on a sheet of paper. Every ruler can agree because every measurement began at the same wrong line. Ask what stayed independent and what did not.

Next, give a reference object whose value is known for the lesson. Ask the learner to decide whether this new evidence changes the conclusion. The key teaching move is to make the learner revise the claim rather than merely announce that the first answer was “wrong.”

Finish with a changed context: three clocks, three light sensors, three balances or three weather readings. If the learner can ask “What do these measurements share?” without prompting, the reasoning has transferred.

Authoritative Sources

The Quiet Return

Science does not distrust agreement. Agreement is useful.

Science simply asks one more question: why did the measurements agree?

If several readings agree because several independent checks point to the same result, confidence can grow. If they agree because every measurement inherits the same hidden shift, counting them as independent confirmations would exaggerate the evidence.

So when a headline, product demonstration, dashboard or experiment says “three sensors agree,” do not stop at the number three. Trace the measurement chains. Find the shared dependencies. Look for the reference. Then let the evidence—not the vote count—decide how strong the claim is.

Small Group Tutorials