PSLE Science Reality Lab Vol No.078 | “The Bars Look Far Apart” — Did the Axis Start Close to the Data?

PSLE-SCI-REALITY-0078

Wait, What? The same two measurements can look slightly different—or wildly different—depending on where a bar chart begins.

An advertisement shows two tall bars. Product A reaches almost to the top of the chart. Product B looks much shorter. The caption says, “A produced dramatically better results.”

Then you read the scale. Product A scored 96 units. Product B scored 94 units. The vertical axis begins at 92, not zero.

The data may be genuine. The two-unit difference may matter. But the picture has made that difference occupy most of the visible height of the chart. The scientific job is not to accuse the chart-maker of dishonesty. It is to separate the measured difference from the visual impression created by the representation.

Reality Lab Vol No.078 teaches one transfer habit: when a real-world bar chart is used as evidence for a scientific or product-comparison claim, inspect the numerical axis before deciding how large the measured difference really is.

Quick Answer

Read the actual values. Do not estimate the scientific effect from bar height alone.
Find the zero point. In a bar chart, bar length visually represents magnitude, so a missing zero can greatly exaggerate the apparent difference.
Calculate the real difference. Ask “96 compared with 94” before asking “how much taller does the bar look?”
Check the quantity and units. A two-unit difference is meaningless until you know what was measured.
Check the comparison conditions. Honest scaling cannot repair an unfair test.
Match the conclusion to the evidence. “A was 2 units higher in this test” is more defensible than “A is massively better” unless broader evidence supports that stronger claim.

Reality Lab habit: A chart can faithfully print every number and still create a misleading sense of size if the visual scale is poorly chosen.

The Owned Learner Job — and the Boundary

This page does not become the general owner of graph scales, axes or data plotting. Those skills already have canonical owners in the eduKateSengkang PSLE Science estate. This Reality Lab applies them to a specific real-world communication object: a bar chart used to make a scientific comparison look larger than the measured values justify.

The underlying graph-reading skill stays with those pages. Vol No.078 asks a narrower transfer question: what should you do when the representation itself is trying to persuade you that a modest measured difference is enormous?

Original Reality Lab Case: The Cooling-Pad Advertisement

Consider an original composite teaching case. No real advertisement or commercial product is being copied.

Two reusable cooling pads are tested under the same stated room conditions. After a fixed cooling period, a laboratory records a performance score on a defined 100-point test scale.

Pad	Measured score	Bar-chart axis shown
Pad A	96	92 to 97
Pad B	94	92 to 97

On that cropped axis, Pad A’s bar rises four units above the chart baseline while Pad B rises only two. Visually, A’s bar can appear about twice as tall. But the actual measured scores are 96 and 94. The scientific difference is two score units, not “twice the performance.”

Observed, Claimed and Inferred

Layer	What it says
Observed	Pad A scored 96 and Pad B scored 94 under the stated test.
Represented	The chart displays bars beginning at 92, making the visible bar lengths very different.
Claimed	Pad A is dramatically better.
Inferred	The large visual gap in bar height is being treated as if it were the size of the measured scientific effect.

A careful learner keeps those layers separate. The chart is a representation of the evidence, not an extra measurement.

Why Bar Charts Are Special

A bar chart uses the length of each bar as part of its message. Your eye compares how far one bar extends from a common baseline. That is why official data-visualisation guidance from the UK Office for National Statistics advises that bar-chart axes start at zero. If the baseline is moved close to the data, a small numerical difference can occupy a large fraction of the displayed bar length.

This does not mean every graph in science must start at zero. Line charts and scatter plots often use a restricted axis so readers can see a pattern more clearly, because the marks are points or lines rather than filled lengths growing from the axis. The important lesson is not a magic rule. It is to understand what the visual shape is encoding.

Representation Check: Rebuild the Chart in Your Head

When a bar chart feels dramatic, mentally remove the bars and keep only the numbers.

Ask: “If the graphic were replaced by the sentence ‘A = 96, B = 94’, would I describe the difference in the same dramatic language?”

If the answer changes sharply, the representation may be doing more persuasive work than the data.

Baseline Check: Where Does the Bar Actually Begin?

Suppose two measurements are 51 and 49.

With an axis from 0 to 60, the bars differ only slightly in height.
With an axis from 48 to 52, one visible bar segment is three units tall and the other one unit tall.

The underlying measurements have not changed. Only the displayed reference point has.

That is the key scientific habit: do not let a graphical reference point silently replace the physical or numerical reference relevant to the claim.

Comparison Check: Difference, Ratio and Meaning Are Not the Same Thing

If A = 96 and B = 94, the difference is 2 units. It is wrong to say A is “twice as high” merely because the visible segment from 92 to 96 is twice the segment from 92 to 94. That ratio belongs to distances measured from the chart’s cropped display baseline, not necessarily to the scientific quantity itself.

To make a ratio claim, the zero point of the scientific quantity and the meaning of the scale both matter. Many scores and indexes do not even support simple “twice as much” interpretations.

Method Check: A Better Chart Cannot Rescue a Bad Experiment

Imagine Pad A was tested for ten minutes but Pad B for fifteen. Or A was tested in cooler air. Or only one trial was performed for each pad. Changing the chart to start at zero would make the picture more honest, but the underlying comparison would still be weak.

Reality Lab reasoning therefore works in two directions:

Is the evidence itself fit for the claim?
Is the representation faithful to the evidence?

Both matter. A fair test can be communicated badly. A beautiful chart can be built from an unfair test.

Source and Provenance Check

Before accepting an impressive chart, ask where the numbers came from. Does the chart name the test, measurement unit, sample size and source? Can the original table or report be found? A chart with a perfect zero baseline still cannot tell you whether the underlying measurements were selected fairly or reported completely.

The best evidence object lets you travel backward: graphic → values → method → observations.

Worked Case 1: 84% Versus 80%

An infographic compares two filtration materials. Material A removes 84% of a test substance; Material B removes 80%. The bars begin at 78%.

The chart can make A’s displayed bar segment look three times the height of B’s segment: 6 units above the cropped baseline versus 2. But the measured difference is four percentage points. Before deciding whether that difference is scientifically important, you still need the test conditions, repeat evidence and uncertainty.

Worked Case 2: A Large-Looking Difference That Is Real but Small

Two materials have measured strengths of 101 N and 100 N. A bar chart begins at 99.5 N. The difference is real: one newton. Whether one newton matters depends on the use, measurement quality and decision threshold. The chart cannot answer that scientific question merely by making the gap look large.

Worked Case 3: A Non-Zero Axis That Is Not Automatically Wrong

A line graph follows temperature from 29.0°C to 30.2°C over six hours. Starting the vertical axis at 28.5°C may help a reader see the small time pattern. Because a line chart marks positions rather than using filled bar length as magnitude, this can be reasonable if the scale is clear.

This is why the learner habit is not “all axes must start at zero.” It is “understand how this representation turns numbers into visual size.”

Worked Case 4: The Chart Is Fine; the Claim Is Still Too Strong

A zero-based bar chart honestly shows A = 70 and B = 55. That is a clear measured difference. But the caption says, “A will always outperform B.” One test cannot support the word “always.” Representation integrity and conclusion scope are separate gates.

Alternative Explanations to Keep Alive

The measured difference may be ordinary trial-to-trial variation.
The products may have started in different states.
One test condition may favour one design.
The selected measurement may capture only one aspect of performance.
The displayed values may be averages hiding individual variation.
The chart may show a real difference but exaggerate its practical importance.

None of these possibilities proves the claim false. They define what further evidence would help.

What Evidence Would Strengthen the Claim?

A bar chart with a zero baseline or another representation that does not use cropped bar lengths.
Exact numerical labels beside the bars.
A clear measurement unit and operational definition.
Comparable test conditions.
Repeat results or individual measurements, not only one summary value.
A source that lets the reader inspect the original data and method.
A conclusion whose wording matches the measured size and scope of the result.

What Would Weaken It?

A bar axis beginning very close to the smallest value with no clear visual warning.
No numerical labels.
A caption using “twice as good” when only cropped bar length doubled.
Different measurement conditions for the compared groups.
Only the most favourable trial shown.
A broad product claim built from one narrow laboratory outcome.

How Far Can the Conclusion Travel?

If the evidence shows that A scored 96 and B scored 94 under one defined test, the safest conclusion begins there. You may be able to generalise further if repeated testing, multiple specimens and a relevant method support it. You may not jump from “two points higher in this test” to “dramatically better in every real situation” simply because the bars look dramatic.

Model and Measurement Limits

Even a perfectly designed bar chart compresses information. It may show only an average, hide variation, omit uncertainty or combine several measurement steps into one number. A chart is a model of the dataset for communication. It is not the entire investigation.

Primary 5/6 learners do not need advanced statistics to use this idea. They only need to keep asking what the mark on the chart stands for and what evidence was left outside the picture.

PSLE-Style Transfer Case

An original practice graph compares the average height of seedlings grown under two lamp settings. The averages are 18.2 cm and 17.8 cm. The bar chart begins at 17.5 cm. A student writes, “The seedlings under Lamp A grew much taller because its bar is more than twice as tall.”

Evaluation: The reasoning is invalid. The measured difference is 0.4 cm. The cropped baseline makes the visible bar segments 0.7 cm and 0.3 cm above the chart minimum, but those display lengths are not the seedling heights. The learner should use the actual values, then consider whether the method and repeated evidence support a meaningful difference.

Tempting Reasoning That Fails

“The taller-looking bar means a much larger scientific effect.” Check the scale.
“If an axis starts above zero, the data must be fake.” The data may be genuine; the representation may simply exaggerate bar-length differences.
“Every graph must start at zero.” Chart type matters. Bars encode length from a baseline; line and scatter plots can reasonably use narrower ranges when clearly labelled.
“The values differ, so the stronger product claim is proven.” You still need fair comparison, repeat evidence and scope control.
“A zero axis guarantees an honest chart.” Other design choices and missing context can still mislead.

Explained Practice

Practice A: Bars show 72 and 70 on an axis beginning at 68. What should you state first? The actual measured difference is 2 units.

Practice B: A bar chart begins at zero but compares averages from groups tested under different temperatures. Is the comparison scientifically secure? No. The representation is clearer, but the method is still confounded.

Practice C: A line chart begins at 19°C and shows readings from 19.2°C to 20.0°C. Is that automatically misleading? No. Read the clearly labelled scale and judge whether the chart supports the stated pattern.

Delayed Independent Return: The V-A-L-U-E Check

V — Values: What are the actual numbers?
A — Axis: Where does it begin, and what intervals are used?
L — Length: Is bar length being interpreted as magnitude?
U — Units: What scientific quantity was measured?
E — Evidence: Do method, repeats and scope support the claim?

Do this later on a different chart. If you can recover the measured comparison without being led by the visual drama, the transfer has stuck.

Parent and Tutor Teaching Guide

Use four invented numbers such as 92, 94, 95 and 96. Draw one bar chart from zero and a second from 90. Do not tell the learner that the numbers are identical. Ask which chart appears to show the bigger differences, then reveal that both use exactly the same data.

Next, remove the bars and show only the values. Ask the learner to write one sentence that is definitely supported and one sentence that would be too strong. Finally, give a line graph with a narrowed axis so the learner learns the important distinction: the goal is not a memorised “zero rule,” but an understanding of how visual geometry carries numerical meaning.

A useful tutor prompt is: “What would you conclude if the picture disappeared and only the table remained?”

Authoritative Sources

SEAB’s current 2026 assessment objectives include interpreting and analysing information, evaluating observations, information and methods, and communicating explanations and reasoning. The ONS guidance is useful here because it explains why bar lengths should begin at zero while other chart types can sometimes use a cropped range. Reality Lab turns those principles into a Primary Science evidence habit.

The Quiet Return

A good scientific graph helps your eye notice something true in the data. A poor one can make your eye feel a difference that the numbers do not support.

So when two bars look far apart, do not begin with the picture’s drama.

Begin with the values. Then ask whether the representation earned the impression it created.

Small Group Tutorials