PSLE Science Reality Lab Vol No.117 | “The Error Bars Are Tiny” — Does That Mean the Measurement Is Accurate?

PSLE-SCI-REALITY-0117

Wait, What? Five Wrong Readings Can Sit Very Close Together

A graph compares two temperature sensors. Each bar has a tiny vertical error bar. An infographic says, “The error bars are tiny, so these measurements are highly accurate.”

That conclusion may be wrong.

Imagine a thermometer that is miscalibrated and always reads about 2°C too high. It might give 27.9, 28.0, 28.0, 28.1 and 28.0°C when the reference value is actually 26.0°C. Those readings cluster tightly. A graph could therefore show a small spread. Yet the measurements are consistently far from the reference value.

Small error bars can be evidence of low variation or a precise estimate, depending on what the bars represent. They do not automatically prove accuracy.

The Reality Lab habit is: first ask what the error bars mean; then ask what evidence connects the measurement to a trusted reference.

Quick Answer

Find the graph caption or methods section and identify what the error bars represent.
Separate low spread, repeatability or uncertainty of an estimate from accuracy relative to a reference.
Check for calibration, reference measurements and possible systematic bias.
Do not use error-bar size alone to claim that a sensor, product or method is “correct”.
When comparing groups, avoid universal shortcuts such as “bars overlap, so there is no difference” unless the statistical meaning of those bars and the analysis justify that conclusion.

The Exact Learner Job This Page Owns

This page owns one evidence-transfer problem: a scientific graph uses small error bars, and a reader mistakes small variation or a precise estimate for proof of measurement accuracy.

It does not replace the canonical PSLE Science owners for precision versus accuracy, repeated readings, uncertainty or graph interpretation. Instead, it applies those micro-skills to a familiar real scientific communication object: a plotted mean with vertical or horizontal bars.

Original Reality Lab Case: The Two Thermometers

This is an original teaching case using constructed data.

A reference bath is maintained at 20.0°C. Two thermometers are tested five times.

Trial	Thermometer A	Thermometer B
1	22.0°C	19.5°C
2	22.1°C	20.4°C
3	22.0°C	20.0°C
4	21.9°C	19.8°C
5	22.0°C	20.3°C

Thermometer A’s readings have very little spread. Thermometer B’s readings vary more. But A is consistently about 2°C above the reference, while B is centred much closer to 20.0°C.

If a graph used standard deviation error bars, A might have the smaller bars. That would support a claim of better short-term repeatability, not a claim of better accuracy.

Observed, Claimed and Inferred

Layer	Statement
Observed	The plotted error bars for Thermometer A are small.
Possible meaning	The repeated measurements cluster closely, or the estimated mean is precise, depending on the bar definition.
Claim	Thermometer A is accurate.
Missing evidence	Comparison with an appropriate reference and investigation of systematic bias.

Error Bars Are Not One Universal Thing

Different graphs use error bars for different quantities. They may show a standard deviation of the observations, a standard error of the mean, a confidence interval, a stated measurement uncertainty or another defined interval.

That means you cannot interpret the picture correctly until you read the caption. A tiny standard-error bar and a tiny measurement-uncertainty bar are not identical statements about the evidence.

The Primary-level rule is simple: never guess what an error bar means from its shape alone.

Precision and Accuracy Are Different Questions

NIST distinguishes short-term variability or precision from accuracy and bias. Measurements can agree closely with one another while sharing a systematic offset. The familiar target-board analogy works because it separates two questions:

Are the shots close together? That resembles precision or repeatability.
Are the shots close to the centre? That resembles accuracy relative to the target.

Small error bars often tell you more about the first question than the second, unless the bars explicitly describe total measurement uncertainty tied to an appropriate reference chain.

The Shared-Bias Problem

A systematic bias can move an entire cluster together. If every measurement is shifted high by the same calibration problem, repeated results can still look beautifully consistent.

NIST’s measurement-process guidance describes consistent bias as a difference that can persist across repeated measurements. This is why repetition alone does not guarantee closeness to a reference value.

Small Bars Can Also Come From a Large Sample

If error bars show uncertainty in an estimated mean, increasing the number of independent observations can make the estimate of the mean more precise even when individual observations still vary widely. A graph can therefore have a narrow confidence interval around the average while the underlying measurements remain spread out.

This is why the individual data points can be valuable. They show the learner whether “tiny bars” came from a tightly clustered process or from estimating the average of many variable observations.

The Error-Bar Overlap Shortcut

Students sometimes learn an unsafe visual rule: “If error bars overlap, the groups are the same; if they do not overlap, the groups are different.” That shortcut is not universally valid because the meaning depends on what the bars represent and on the statistical comparison being made.

If the scientific claim requires a formal comparison, use the appropriate analysis rather than inventing a universal rule from bar overlap. Reality Lab Vol No.116 applies this same caution to the phrase “no statistically significant difference”.

The Representation Check: Did the Graph Hide the Reference?

A graph can show tiny error bars but omit the known reference value. Without the reference line, a consistent bias may be invisible.

Imagine Thermometer A plotted alone at 22.0°C with a tiny bar. It looks impressively stable. Add a horizontal reference line at 20.0°C and the scientific story changes immediately. The graph now shows both repeatability and disagreement with the reference.

The Scale Check: Tiny Compared With What?

Error bars can look visually tiny or huge depending on the vertical axis. A graph spanning 0–1000 units can make a ±5-unit interval appear almost invisible. A graph spanning 95–105 makes the same interval dominate the picture.

Always read the numerical scale. Visual size is not the measurement itself.

Worked Case 1: The Biased Scale

A balance repeatedly reads a 100 g reference mass as 103.0, 103.1, 102.9 and 103.0 g. The spread is tiny. The balance is repeatable but biased high. Small error bars would not prove accuracy.

Worked Case 2: The Noisy but Unbiased Method

A second balance gives 98, 101, 102, 99 and 100 g. The values spread more widely, but their centre is close to the 100 g reference. Depending on the job, this method may be less precise but less biased.

Worked Case 3: Many Measurements, Narrow Mean Interval

A hundred plant heights range from 15 to 25 cm, but the estimated average has a narrow confidence interval. A small bar around the mean does not mean every plant has nearly the same height. It means the study estimates the group average relatively precisely under that statistical definition.

Worked Case 4: Two Product Bars

Product X has a mean of 50 with tiny bars; Product Y has a mean of 52 with tiny bars. Can we conclude Y is scientifically superior? Not until we know what was measured, what the bars represent, whether the comparison was fair, whether the 2-unit difference is meaningful and whether any shared bias affects both measurements.

Tempting Reasoning That Fails

“Tiny error bars mean the experiment is accurate.” They may reflect precision or uncertainty of an estimate, not closeness to a reference.
“Big error bars mean the experiment was badly done.” Natural systems can genuinely vary widely.
“No overlap proves a difference.” The validity of that visual rule depends on the bar type and analysis.
“Overlap proves no difference.” Also too strong.
“All error bars show measurement uncertainty.” Many graphs use standard deviation, standard error or confidence intervals instead.
“If repeated readings agree, calibration is unnecessary.” A systematic bias can make all readings agree around the wrong value.

What Evidence Would Strengthen an Accuracy Claim?

the error-bar definition is stated clearly;
the measurement is compared with an appropriate reference or standard;
calibration and known corrections are documented;
possible systematic bias is investigated;
individual observations are shown where useful;
the method’s uncertainty is appropriate for the claimed difference;
independent methods or reference checks give compatible results.

What Would Weaken It?

the caption does not define the bars;
only the mean and bars are shown while individual data are hidden;
there is no reference value for an accuracy claim;
all sensors share the same calibration source or systematic bias;
the y-axis makes bars look tiny without showing their numerical magnitude;
the public claim says “accurate” when the study measured only repeatability.

Model and Measurement Limits

An error bar compresses information. It cannot show every source of uncertainty, every outlier, every distribution shape or every systematic effect unless the underlying analysis includes them. Even formal uncertainty intervals must be interpreted according to their stated definition and assumptions.

The graph is a communication object, not a replacement for the method.

How Far Can the Conclusion Travel?

Small bars can legitimately strengthen claims about consistency or precision when they represent the relevant quantity. If calibration and uncertainty evidence also show closeness to a suitable reference, they can contribute to an accuracy claim. But bar size alone cannot carry that conclusion.

The transferable scientific habit is to keep spread, uncertainty, bias and accuracy from collapsing into one vague word: “good”.

PSLE-Style Transfer Case

A thermometer measures a 25.0°C reference five times and gives 27.0, 27.1, 27.0, 26.9 and 27.0°C. The repeated values have little spread.

A student says: “The small spread proves the thermometer is accurate.”

Reasoned answer: The repeated readings are precise because they agree closely, but they are about 2°C above the 25.0°C reference. The thermometer may have a systematic bias, so small spread does not prove accuracy.

Explained Practice

Practice A: A graph shows mean ± standard deviation. What do the bars mainly describe? The spread of observations around the mean, not automatically measurement accuracy.

Practice B: A graph shows a narrow confidence interval around the mean. Does that mean every observation is near the mean? No. A mean can be estimated precisely even when individual observations vary.

Practice C: Two sensors have tiny error bars but both read 3 units above a certified reference. What is the likely evidence pattern? Good consistency with a shared positive bias.

Delayed Independent Return: The B-A-R-S Check

B — Bar definition: What exactly do the bars represent?
A — Accuracy evidence: Is there a trustworthy reference or calibration check?
R — Repeated spread: Are the bars mainly describing variability or precision?
S — Systematic bias: Could all results be shifted together?

Parent and Tutor Teaching Guide

Draw two target boards. Put five dots tightly clustered far from the bullseye on one target, and five more scattered dots centred around the bullseye on the other. Ask which group is more precise and which is more accurate. Then translate the dots into repeated measurements and error bars.

Next show two identical bar graphs with tiny error bars. Tell the learner that one graph uses standard deviation and the other uses confidence intervals. Ask why the caption matters. This prevents the learner from treating a graphical symbol as if it had one fixed meaning everywhere.

Authoritative Sources

The official Singapore Science frame asks learners to interpret and analyse information, evaluate observations and methods, consider uncertainty and communicate explanations responsibly. Error bars are exactly the kind of compressed scientific representation that should trigger questions rather than automatic trust.

The Quiet Return

A tight cluster tells you that measurements agree with one another.

To know whether they agree with reality, you still need a reference and a trustworthy method.

When the error bars look impressively small, ask what they are small around.

Small Group Tutorials