BTT Mathematics / Primary Mathematics Learning Hub / Statistical Thinking

Primary Mathematics: Data Claims, Sampling, Misleading Graphs and Evidence | Worked Learning Guide

A data display can be numerically accurate and still create a misleading impression. Good data reasoning checks the question, the sample, the scale, the summary and the strength of the final claim.

Modern learners encounter graphs, percentages and claims everywhere. “Most pupils prefer…”, “scores increased by 50%”, “Group A doubled Group B”, “this chart proves…”. The arithmetic may be correct while the conclusion is too strong, the sample is biased or the visual design exaggerates a small difference.

This guide develops a practical evidence audit for upper-Primary learners. It is a bridge into later statistical literacy and is not presented as a compulsory formal statistics syllabus for every Primary level.

Claims · Samples · Misleading graphs · Averages and percentages · Association and cause · Evidence strength · 24 questions · Worked solutions

1. Start by separating claim from evidence

Claim: “Most pupils in the school walk to school.”

Evidence: 18 of 25 surveyed pupils said they walk.

Before accepting the claim, ask who the 25 pupils were and whether they represent the whole school.

The calculation 18/25=72% may be correct while the school-wide claim remains unsupported.

2. One anecdote is not a group pattern

“My friend improved after using Method X” describes one case.

It does not establish that Method X improves outcomes for all pupils.

Group claims require group evidence.

3. Sample selection matters

To study school transport, surveying only pupils standing at a bus stop creates obvious bias.

A more representative approach would select pupils across classes or times in a way that does not systematically favour one transport mode.

4. Large biased samples can still mislead

Surveying 500 cyclists at a cycling event gives a large sample but poor evidence for how the general population travels.

Sample size cannot repair a selection process that systematically excludes important groups.

5. Small samples can vary greatly

In a sample of four pupils, one response changes a percentage by25 percentage points.

In a sample of100 pupils, one response changes it by1 point.

Small samples can produce unstable percentages.

6. Non-response can matter

If 100 surveys are sent and only40 returned, the 40 responses describe respondents.

If non-respondents differ systematically, the result may not represent all 100.

Report response count, not just percentages.

7. Truncated axes can exaggerate differences

Bar heights 98 and100 look similar on a0–100 axis.

On an axis starting at97, the 2-unit difference can appear enormous.

The numerical difference is still2. The visual impression has changed.

Worked example

Average scores A=78, B=82.

Difference=4 points.

A graph starting at75 may make B’s bar appear several times as tall above the baseline. Read the scale before judging effect size.

8. Unequal intervals destroy visual meaning

If an axis labels 0,10,20,100 at equal physical spacing, the scale is inconsistent.

Distances on the graph no longer represent equal numerical changes.

Unless a special scale is explicitly defined, this is misleading.

9. Pictograms can exaggerate by area

Suppose one icon represents10 pupils. If a designer doubles both height and width of an icon to represent20 pupils, its area becomes four times as large.

The picture visually suggests four times the amount, not twice.

For quantity comparison, repeated equal-size icons are safer.

10. 3D bars can distort length comparisons

Perspective and volume make front bars appear larger than rear bars even when heights are equal.

Use simple 2D bars when bar height is meant to encode value.

11. Missing baselines can hide absolute size

A graph showing “50% increase” without original values hides whether the change was 2 to3 or200 to300.

Relative change needs context.

12. “Average” must name which average

Mean, median and mode are different summaries.

A claim saying “average household size is4” should ideally specify the statistic used.

For numerical data, mean is often intended, but statistical precision benefits from naming it.

13. A mean can be pulled by extreme values

Data: 10,11,12,13,54.

Mean=20, median=12.

Calling20 “typical” without showing the distribution is questionable.

14. Percentage changes depend on the base

Increase from20 to30:

Change=10.

Percentage increase=10/20=50%.

Do not divide by30 unless the question specifically uses the final value as reference.

15. Percent and percentage points are different

A proportion rises from40% to50%.

Increase = 10 percentage points.

Relative percentage increase = 10/40 = 25%.

Both statements can be correct but answer different questions.

16. Counts and percentages can tell different stories

Group A: 18 of20 succeed =90%.

Group B: 80 of100 succeed =80%.

B has more successful individuals by count; A has higher success proportion.

State which comparison matters.

17. Association does not automatically prove cause

Suppose pupils who read more minutes tend to score higher.

The data may show an association.

They do not by themselves prove that extra reading caused the score difference. Prior attainment, access to books, support and other factors may also matter.

This is an enrichment-level caution in evidence interpretation.

18. Reverse causation can be possible

If pupils who enjoy mathematics practise more, practice may support confidence; confidence may also encourage practice.

A simple association does not tell us the direction of cause.

19. Coincidence is more plausible in tiny datasets

Two variables matching across three observations can happen by chance.

Larger, well-designed datasets provide stronger evidence, though design quality still matters.

20. Match claim strength to evidence strength

Evidence	Safer conclusion
one class survey	“In this class…”
several selected classes	“In these sampled classes…”
whole-school census	“Among pupils in this school at this time…”

Do not generalise farther than the data collection supports.

21. Use uncertainty language honestly

Useful phrases include:

• “The data suggest…”
• “In this sample…”
• “By this measure…”
• “We cannot conclude…”
• “Further data would be needed…”

Careful language is a strength, not hesitation.

22. Ask what evidence would change your mind

If a claim says “Route A is more reliable,” ask what data would count against it.

A strong claim should be testable rather than protected from all counterevidence.

23. A data audit should inspect the full chain

1. Was the question clear?
2. Was the sample appropriate?
3. Were data collected consistently?
4. Is the graph scaled honestly?
5. Is the summary statistic suitable?
6. Does the conclusion match the evidence?

An error anywhere in the chain can weaken the final claim.

24. Common data-claim errors

• Generalising from an unrepresentative sample.
• Reporting percentages without sample size.
• Using truncated axes to exaggerate differences.
• Using unequal intervals on ordinary axes.
• Scaling pictogram height and width together.
• Using “average” without identifying the measure.
• Confusing percentage points with percent increase.
• Claiming cause from association alone.

25. A data-claim audit protocol

1. Claim: What exactly is being asserted?
2. Source: Who or what produced the data?
3. Sample: Who was included and who was not?
4. Display: Are axes, intervals and symbols proportional?
5. Summary: Which statistic is used, and is it appropriate?
6. Context: Are counts, percentages and baselines visible?
7. Strength: Does the conclusion go beyond the evidence?

26. Practice: 24 original questions

Questions 1–8: Sampling and graphs

1. A transport survey samples only pupils at bicycle racks. Identify the bias.
2. A survey of500 cyclists is used to describe national transport habits. Why can size fail to fix the problem?
3. In a sample of4, one response changes the percentage by how many points?
4. In a sample of100, one response changes it by how many points?
5. A graph compares98 and100 using an axis from97 to100. What visual risk appears?
6. An axis labels0,10,20,100 at equal spacing. What is wrong?
7. An icon’s height and width are both doubled to represent twice the quantity. What happens to its area?
8. Why can 3D bars be harder to compare accurately?

Questions 9–16: Averages and percentages

9. Find mean and median of10,11,12,13,54.
10. Which summary is closer to most values in Question9?
11. A value rises20→30. Find percentage increase.
12. A proportion rises40%→50%. Find increase in percentage points.
13. Find relative percentage increase for Question12.
14. Group A:18/20 succeed. Group B:80/100. Compare success proportions.
15. Which group has more successful individuals by count?
16. Explain why counts and percentages answer different questions here.

Questions 17–24: Claims and evidence

17. A class survey finds72% walk to school. Rewrite the claim so it does not generalise beyond the class.
18. Give one reason an association between reading time and test scores may not prove cause.
19. What phrase can be used when evidence suggests but does not prove a conclusion?
20. Why should sample size accompany a reported percentage?
21. Give one example of a graph design that exaggerates a small difference.
22. Write one question that should be asked before accepting an “average increased” claim.
23. Create a strong data claim and then weaken it to match a small sample.
24. Use the seven-step audit protocol on any graph or claim you have seen recently.

27. Worked solutions

1. Cyclists are overrepresented. 2. The selection process still excludes or overrepresents groups systematically. 3. 25 percentage points. 4. 1 percentage point.

5. The 2-unit difference may look dramatically larger because the baseline is truncated. 6. Equal physical intervals represent unequal numerical changes. 7. Area becomes4 times as large. 8. Perspective and volume can distort perceived bar size.

9. Mean20, median12. 10. Median12. 11. 50%. 12. 10 percentage points.

13. 25% relative increase. 14. 90% versus80%; A higher. 15. B has80 successes versus18. 16. Counts measure number of cases; percentages compare shares relative to group size.

17. “Among pupils surveyed in this class,72% reported walking to school.” 18. Prior attainment, support or other variables may influence both reading and scores. 19. “The data suggest…” or “In this sample…”. 20. The same percentage from a tiny and large sample may carry different stability and context.

21. Example: start the vertical axis just below the smallest value rather than zero without clearly signalling the scale. 22. Ask which average, from what baseline, and across what group/time period. 23. Example strong: “All pupils prefer X.” Weaker: “In this sample, most respondents preferred X.” 24. Answers vary; each audit should inspect claim, source, sample, display, summary, context and claim strength.

28. Evidence laboratory: two true numbers, one misleading story

School A: 90% pass rate from20 pupils →18 passes.

School B: 80% pass rate from200 pupils →160 passes.

Statement 1: “School A has the higher pass rate.” True.

Statement 2: “School B has more pupils who passed.” True.

Misleading statement: “School A produced more successful pupils.” False by count.

The same data support different claims depending on whether the quantity is proportion or absolute number. Good evidence language names the quantity explicitly.

29. Parent and tutor guide

Use everyday graphs as critique material, but focus on mathematical features rather than brand or politics: axes, baseline, group size, percentage base, sample selection and wording.

Ask the learner to rewrite overstrong claims into evidence-matched statements. This builds both mathematical precision and responsible data literacy.

30. Mastery receipt

• I separate a data claim from its evidence.
• I check whether a sample can represent the target group.
• I inspect axes, intervals and pictograms for distortion.
• I distinguish counts, percentages and percentage points.
• I choose averages carefully.
• I know association alone does not prove cause.
• I use uncertainty language when evidence is limited.
• I audit the full chain from question to conclusion.

Sources and scope

NCTM — Mathematics Education / Statistics Study in Elementary Schools outlines a statistics cycle of formulating questions, collecting data, analysing data and interpreting results, with attention to variability and group comparisons.

Illustrative Mathematics — Identifying Statistical Questions provides the foundational distinction between fixed-answer questions and questions requiring variable data.

For Singapore Primary curriculum scope, use the MOE Primary Mathematics Syllabus P1–P6, updated October 2025. This guide is an upper-Primary enrichment/transition bridge.

Continue the Statistical Thinking collection

• Statistical Questions, Data Collection and Variability
• Distributions, Center, Spread and Typical Values
• Comparing Data Groups, Consistency and Variability
• BTT Primary Mathematics Learning Hub

The Quiet Return

Good data literacy is not suspicion of every graph. It is knowing exactly what must be true before a graph or percentage deserves the conclusion attached to it.