Sec 2 G2 Maths Bukit Timah | Sampling and Survey Bias

Did you know that interviewing more people does not automatically remove survey bias? A thousand replies from an open online poll can still miss students who never saw the poll or had no reason to respond. By contrast, a properly designed smaller random sample can provide more useful information about a defined population—although it still carries sampling variation and possible nonresponse.

For Secondary 2 G2 Mathematics learners in Bukit Timah, the important first question is often not ‘What percentage did you calculate?’ but ‘Of whom?’ If eighty students are invited, fifty-six reply and thirty-five choose an option, the response rate is seventy percent, while support among respondents is 62.5 percent. The same small set of counts produces different legitimate numbers because their denominators answer different questions.

This guide explores populations, sampling frames, convenience and voluntary responses, leading questions, response rates, missing answers and cautious interpretation of charts. It connects Mathematics with real-world data literacy without pretending a school exercise is a scientific census of Singapore students.

Quick answers: the first five questions to ask about a survey

Question	Useful answer
Who is the target group?	The population about whom the report seeks information.
Who was actually measured?	The sample or respondents, not necessarily the whole population.
How were they chosen?	The selection method, such as random, convenience or voluntary response.
What percentage is reported?	A numerator divided by its correctly named denominator.
Who failed to respond?	Missing answers that may matter to interpretation.
What can be stated exactly?	Observed sample counts and calculations.
What is an estimate?	A conclusion about people not directly measured, based on assumptions.

Data analysis and statistical literacy in Singapore G2 Mathematics

Secondary G2 Mathematics develops data presentation, frequency, averages, probability and reasoning about information. How a dataset was collected is a useful broader application of these skills. Formal sample design, weighting and population inference can be extensions rather than assessed core chapters in a particular school. Under Full Subject-Based Banding, the student’s G2 subject level does not define performance in all subjects.

Curriculum context: MOE’s G2 Mathematics syllabus and a 2026 Secondary 2 G2 school outline. All survey figures below are fictional for teaching.

Twenty-four ideas that make survey conclusions trustworthy

1. The population is the whole group a question concerns

A population is the complete set of people or objects whose characteristics we want to understand. If the question is ‘How do all 1,000 students at a school travel to lessons?’, the population is those 1,000 students, not the twenty who happen to be in one classroom. An answer can become misleading even with perfect arithmetic if the group being described quietly changes during the analysis.

Teaching move: Ask the pupil to state the target population as a complete noun phrase before looking at any percentages.

Common mistake: Do not assume answers from one convenient class automatically describe every student in the school.

2. A sample is a smaller subset used to learn about the population

A sample contains some of the population members. Choosing 100 of 1,000 students for a survey can be quicker than asking everyone, but the sample’s composition influences how useful its results are. The aim is not only to collect a number but to choose people in a manner relevant to the question. Even a well-chosen sample can differ from the whole population through natural sampling variation.

Teaching move: Draw a large box labelled population and a smaller set of dots labelled sample.

Common mistake: A sample percentage is an observation about the sampled group, not proof of the exact population percentage.

3. Sample size is not the same as population size

A population of 1,200 people may be studied through a sample of 120. The sample fraction is120/1200=10%, but this fraction is not the percentage who support a proposal. That support percentage must be calculated using the number of favourable answers among those actually observed. Students should keep sampling fraction, response rate and result percentage in separate rows.

Teaching move: Put population N, selected sample n and favourable count f in distinct columns.

Common mistake: Dividing favourable answers by the whole school population gives a different quantity from the sampled response share.

4. A census attempts to measure the whole population

If a teacher successfully collects one relevant response from all forty pupils in a class, that is a census of that particular class, not a sample of it. It still does not form a census of an entire school. A census can remove sampling uncertainty for its defined population, but reporting mistakes, nonresponse and ambiguous wording can still damage data accuracy. Naming the population precisely remains essential.

Teaching move: Compare asking every member of one class with sampling students across the school.

Common mistake: A complete count of a small class is not a complete count of a larger school.

5. A simple random sample aims to avoid selective inclusion

In an ideal simple random sample, every subset of the specified size has the same selection chance. A school might number all eligible students and use a proper random mechanism to choose a set of names. This method reduces deliberate choice bias, though the resulting sample can still differ from the population by chance and may suffer nonresponse. Random selection is not the same as selecting whoever first opens a link.

Teaching move: Describe drawing numbered tokens or using an appropriate random selection procedure from a complete list.

Common mistake: Calling a social-media poll ‘random’ does not make visitors to that page a simple random sample of all students.

6. A sampling frame is the list from which selections are made

A sampling frame should include members of the target population appropriately. If a survey about every student uses only the list of students in one after-school club, pupils who never join that club have no chance to be selected. The bias arises before questionnaire design or arithmetic. A larger sample from the wrong frame may still omit the same important groups.

Teaching move: Write the target group and the actual list of eligible names side by side.

Common mistake: A complete club list is not necessarily a complete school list.

7. Convenience sampling can be useful but has limits

Asking the first thirty students entering a library may be easy. It can describe those particular thirty respondents, but it may overrepresent students who use the library and miss people who go straight home. This does not make their answers dishonest. It means the method does not justify broad conclusions about all pupils without further support. Good data literacy includes stating the convenience clearly.

Teaching move: Ask how a change in survey location could change who is approached.

Common mistake: Thirty genuine answers do not automatically establish a representative sample.

8. Voluntary-response surveys invite self-selection

An online poll lets people choose whether to participate. People with strong opinions, frequent internet access or special interest may respond more readily. A large number of voluntary responses may still underrepresent those who never saw or chose the poll. Describing the results as ‘of the people who answered this poll’ is more accurate than declaring them the preferences of every local resident.

Teaching move: Contrast an open link with a randomly selected invitation list.

Common mistake: Thousands of votes cannot correct selection bias simply by becoming numerous.

9. Response rate measures who actually replied

If eighty randomly selected people are invited and fifty-six provide usable responses, the response rate is56/80=70%. The survey result percentage should usually use the fifty-six usable responses as its denominator when describing respondents’ answers. Response rate is a different number and asks a different question: how many invited participants answered. Both figures matter for interpreting a report.

Teaching move: Write invited, returned and positive counts on separate lines.

Common mistake: Do not use the invited count as the denominator for a percentage described explicitly as ‘among respondents’.

10. Nonresponse can introduce bias even after good random selection

Suppose a representative-looking invitation list is random, but only people who enjoy Mathematics choose to reply to a Mathematics-club survey. The answered set may no longer reflect the random invited group. Increasing the number of reminders or offering several response methods may help participation but does not guarantee removal of bias. A responsible report states the response rate and considers how nonrespondents could differ.

Teaching move: Compare preferences of early responders with those who might never respond.

Common mistake: Randomly selecting invitees does not ensure that the final respondents remain unbiased.

11. A respondent share has a precise denominator

If fifty-six students reply and thirty-five choose an option, the observed respondent share is35/56=62.5%. It is not35/80 when eighty were invited unless the question explicitly asks for confirmed supporters among invitees. Those two fractions describe different things. The report should name the denominator directly in its sentence, for example ’35 of 56 respondents’.

Teaching move: Write a sentence that includes the numerator and denominator before converting to a percentage.

Common mistake: A bare ‘62.5% of students’ may falsely imply the entire school has been measured.

12. Missing responses limit what can be known about invitees

Suppose eighty people are invited, sixty reply, and forty of the sixty support a proposal. Among respondents the observed support share is40/60=66⅔%. Among the eighty invitees, support could be as low as40/80=50% if none of the twenty nonrespondents supports it, or as high as60/80=75% if all do. These are simple arithmetic bounds for invitees, not a confidence interval or a claim about the population.

Teaching move: Count the unknown nonrespondents as a separate category before interpreting results.

Common mistake: Do not substitute a statistical confidence statement for information the survey never collected.

13. Selection bias differs from random sampling variation

Random sampling variation means a properly chosen sample may by chance contain slightly more or fewer members of a group than the population. Selection bias means the method systematically offers different inclusion chances or response opportunities to relevant groups. A larger sample can reduce some random variability, but it does not automatically correct a biased selection mechanism. Knowing which problem is present determines what improvement is sensible.

Teaching move: Compare an unlucky but properly random sample with a huge convenience sample from only one club.

Common mistake: Do not claim any difference between sample and population automatically proves deliberate bias.

14. Stratification can help represent important subgroups

If a school population is 40% lower-secondary and 60% upper-secondary, a proportional sample of fifty could include twenty lower- and thirty upper-secondary students, with random selection within each relevant group. This makes the sample’s broad composition match the school on that chosen characteristic. It does not guarantee that every other characteristic or nonresponse pattern is balanced. The method is an extension in statistical design, not a promise of perfect representativeness.

Teaching move: Build a table showing40% of fifty and60% of fifty.

Common mistake: A proportional grade distribution cannot by itself remove bias in who responds within each grade.

15. An unbalanced sample may need careful weighting for population estimates

Suppose 40% of a target population is group A and60% group B. A disproportionate sample happens to include thirty A and ten B. If80% of A responders and20% of B responders favour an idea, the raw sample share is(24+2)/40=65%. A population-composition weighted estimate would be0.4×0.8+0.6×0.2=44%, assuming those within-group rates are suitable estimates. This example shows why simply aggregating an imbalanced sample can mislead.

Teaching move: Calculate the raw and composition-weighted results separately, then explain their differing denominators.

Common mistake: A weighted estimate is not automatically accurate if the within-group samples are themselves biased.

16. Leading questions can change the answers collected

Compare ‘Do you support our excellent additional revision sessions?’ with ‘Would you prefer additional revision sessions, no change, or another type of support?’ The first contains a positive judgement before the respondent answers. Both are understandable English questions, but the first can push the survey toward agreement. Wording is part of the measurement instrument and deserves the same attention as sample selection.

Teaching move: Ask learners to identify opinionated adjectives before rewriting a neutral version.

Common mistake: A balanced sample cannot repair a questionnaire that systematically suggests which answer is expected.

17. Double-barrelled questions mix multiple decisions

‘Should school start earlier and finish later?’ asks about two separate changes in one yes/no response. A student might support one part but oppose the other, making the answer hard to interpret. Separate the topics into different questions or offer clear alternatives. Statistical honesty begins by ensuring that each recorded response refers to a well-defined decision.

Teaching move: Turn one compound question into two neutral independent items.

Common mistake: Reporting a yes response as support for both parts may misrepresent what was intended.

18. Timing and location can affect who is available

A lunchtime survey taken only at one canteen may overrepresent pupils eating there and miss those with other schedules. An evening online survey may exclude families without easy access or time. These constraints can matter even when the questionnaire itself is fair. A responsible study considers where, when and how people have an opportunity to be selected or reply.

Teaching move: List three groups who could be systematically absent from a proposed survey location.

Common mistake: Convenience at collection time is not evidence of balanced population coverage.

19. Survey categories should be mutually exclusive when counting one answer each

If students select one primary mode of travel, a response should enter exactly one category. But if the question asks them to tick all ways they travel in a week, one student can contribute several category counts. Those counts can sum above the number of students. A pie chart dividing people into exclusive categories would then be inappropriate unless a new whole, such as all selections, is explicitly defined.

Teaching move: Check whether the form permits one response or multiple selections.

Common mistake: Percentages in a multiple-answer question need not total100% of respondents.

20. Privacy and consent are part of trustworthy data practice

School surveys involving children should avoid unnecessary identifiable personal details and follow the institution’s applicable privacy and consent procedures. A Mathematics exercise about sampling does not need students’ phone numbers, medical information or private family circumstances. Aggregate, hypothetical classroom data are usually sufficient to practise percentages and chart interpretation. Good method includes respect for people, not only arithmetic.

Teaching move: Design a safe anonymous question about preferred revision formats without collecting names.

Common mistake: Do not publish or invent identifiable student data just to make a worked example seem realistic.

21. A sample result is not automatically a population fact

If a randomly selected sample of one hundred students has sixty-two who prefer one option, the observed sample percentage is62%. Under suitable conditions it can inform an estimate for a larger population, but the precise population count is not thereby known. A statement such as ’62 of100 sampled students chose this option’ is fully supported. ‘Exactly62% of the entire school’ is stronger and needs additional evidence.

Teaching move: Ask which sentence can be concluded exactly from the data and which is an estimate.

Common mistake: A neatly calculated sample proportion cannot eliminate uncertainty about unobserved people.

22. Results should distinguish facts, estimates and assumptions

Facts might include ’56 usable responses were recorded’ and ’35 respondents selected option A’. Calculations such as35/56=62.5% are exact for those recorded responses. Extrapolating that share to unobserved students is an estimate that relies on assumptions about selection and response. A good report can use all three categories, but should label them rather than blur them into one confident claim.

Teaching move: Write fact, calculation, estimate and assumption in separate rows for one toy survey.

Common mistake: Do not give a sample-based estimate the same certainty as a directly counted total.

23. One survey cannot answer every related question

A survey of preferred bus versus walking travel does not directly establish commute duration, safety, family income or general satisfaction with transport. Those are different variables requiring separate questions and possibly different sampling considerations. Students sometimes infer too much from one attractive chart. Ask what was actually measured before discussing possible causes or recommendations.

Teaching move: Name one conclusion supported by a given survey and two that are not measured.

Common mistake: Correlation, preference and causal explanation should not be treated as interchangeable.

24. A report should show enough method for another reader to judge it

A useful school-level data report states its target population, who was invited, how they were selected, number of replies, question wording and calculation denominator. The resulting percentages and charts then have context. This does not require advanced statistical jargon; one honest paragraph explaining how the data were collected can prevent major misunderstandings. Concise transparency is a mathematical communication skill.

Teaching move: Draft a short methods note for a hypothetical 60-person survey.

Common mistake: A polished graph without source, total or selection explanation can look precise while remaining hard to trust.

Thirty worked data and sampling problems

Worked example 1: A hundred-person sample

Question: In a random sample of100 of a school’s1000 students,62 selected walking. State exact sample share.

Solution or interpretation: 62/100=62% of sampled students chose walking.

What it shows: The observed count62 is exact for that sample.

Wrong inference: Do not assert exactly620 of all1000 prefer walking.

Changed question: Explain how sampling variation could affect the whole-school share.

Worked example 2: Sampling fraction

Question: A school population is1200 and120 are sampled. Find sample fraction.

Solution or interpretation: 120/1200=0.10=10%.

What it shows: This describes what proportion was sampled, not who supports an idea.

Wrong inference: Do not confuse sampling fraction with survey support percentage.

Changed question: Try population800, sample80.

Worked example 3: Response rate

Question: Of80 invited students,56 reply. Find response rate.

Solution or interpretation: 56/80=70%.

What it shows: Twenty-four did not reply.

Wrong inference: Do not describe70% as support for an option.

Changed question: What if60 reply?

Worked example 4: Result among respondents

Question: Of56 respondents,35 favour option A. Find observed support share.

Solution or interpretation: 35/56=5/8=62.5%.

What it shows: The denominator is respondents56.

Wrong inference: 35/80 would instead be a share of all invited with confirmed positive responses.

Changed question: Find how many respondents did not favour A.

Worked example 5: Nonrespondent bounds

Question: Eighty people invited; sixty answer; forty of sixty support A. Give minimum and maximum support among the eighty invitees if the twenty missing answers are unknown.

Solution or interpretation: Minimum40/80=50%; maximum60/80=75%.

What it shows: There are twenty unknown invitees, so confirmed supportive count lies between40 and60.

Wrong inference: These are arithmetic bounds for invitees, not a confidence interval for the school.

Changed question: What if only ten are missing?

Worked example 6: One-class census

Question: A class has40 students and all40 answer an anonymous preference question. Is it a census?

Solution or interpretation: It is a census of the defined forty-student class, but only a sample of a larger school population.

What it shows: Every class member was counted.

Wrong inference: Do not call it a census of the entire school.

Changed question: What if only thirty-six reply?

Worked example 7: Convenience versus population

Question: A survey of30 pupils entering the library asks about library use. What is the main limitation?

Solution or interpretation: Library visitors are disproportionately included, so the results may not describe students who rarely or never use the library.

What it shows: The frame is conditioned on entering the library.

Wrong inference: Thirty real responses need not be representative of the whole school.

Changed question: Propose a better school-wide selection method.

Worked example 8: Voluntary response

Question: An open social-media poll gets2000 votes about tuition. Does its size alone ensure representativeness?

Solution or interpretation: No. Respondents self-select and may differ from nonparticipants, even with many votes.

What it shows: Selection mechanism, not just n, matters.

Wrong inference: A large online poll is not automatically a simple random sample.

Changed question: Describe the population it truly observes.

Worked example 9: A leading question

Question: Is ‘Do you support our excellent new programme?’ neutral?

Solution or interpretation: No. The positive word ‘excellent’ can influence answers. A less leading version asks whether respondents support, oppose or are undecided about the proposed programme.

What it shows: Neutral categories reduce suggestion.

Wrong inference: Do not combine praise with the measurement question.

Changed question: Write a balanced three-option version.

Worked example 10: Double-barrelled item

Question: A question asks whether school should start earlier and finish later. Why might yes/no be hard to interpret?

Solution or interpretation: It combines two changes. A respondent might favour only one. Ask separate questions.

What it shows: The answers become more meaningful when one decision is measured at a time.

Wrong inference: One yes does not establish agreement with both changes.

Changed question: Draft two separate questions.

Worked example 11: One response per student

Question: Forty students may each select one mode: bus16, walk12, car8, cycle4. Is a pie chart possible?

Solution or interpretation: Yes. Counts sum40 and categories are exclusive, making parts of one whole.

What it shows: Sector shares total100%.

Wrong inference: One student choosing multiple modes would change the data interpretation.

Changed question: Find bus sector144°.

Worked example 12: Multiple answers

Question: Twenty pupils each select every subject they enjoy; subject checkmarks total47. Can47 be a percentage denominator for people?

Solution or interpretation: Forty-seven is number of selections, not people. Percentages of respondents choosing each subject use denominator20 and can total above100%.

What it shows: One person may contribute several selections.

Wrong inference: Do not force the subject shares of people to sum100%.

Changed question: State one defensible alternative chart.

Worked example 13: Proportional sampling

Question: A school’s target population is40% group A and60% group B. For a proportional sample of50, how many from each?

Solution or interpretation: A:20, B:30.

What it shows: Twenty plus thirty equals50 and proportions match40/60.

Wrong inference: Choosing composition proportionally does not guarantee every other variable is balanced.

Changed question: Try sample size100.

Worked example 14: Raw disproportionate share

Question: A sample has30 group A and10 group B. In A,24 support; in B,2 support. Find overall raw sample share.

Solution or interpretation: Support total26 of40, so65%.

What it shows: Group A contributes three-quarters of the sample.

Wrong inference: Raw65% may not match population composition.

Changed question: Find each within-group support share.

Worked example 15: Stratum-specific percentages

Question: Using A24/30 and B2/10 from prior example, calculate separate support percentages.

Solution or interpretation: Group A80%, group B20%.

What it shows: These are different preferences in the measured groups.

Wrong inference: The rates do not themselves reveal full population share without its mix.

Changed question: What is the difference in percentage points?

Worked example 16: Population-composition weighted illustration

Question: Population has40% group A and60% group B, and estimated within-group support rates80%,20%. Compute weighted overall estimate.

Solution or interpretation: 0.4×0.8+0.6×0.2=0.32+0.12=0.44=44%.

What it shows: This differs from raw sample65% because the sample mix differs.

Wrong inference: Estimate relies on within-group sample suitability.

Changed question: Explain when this adjustment might still be unreliable.

Worked example 17: A school response frame

Question: A survey target is all800 students, but the contact list contains only300 after-school-club members. What bias risk arises?

Solution or interpretation: Students outside clubs have no inclusion opportunity from that frame; whole-school results may be biased.

What it shows: The frame covers only part of the population.

Wrong inference: Random selection within the club alone does not fix the missing students.

Changed question: Describe one more appropriate frame.

Worked example 18: Same percentage, different counts

Question: Two samples have60% support, with n=20 and n=200. Find positive counts.

Solution or interpretation: In first:12; in second:120.

What it shows: Both share60% but observed counts differ by factor ten.

Wrong inference: Equal percentages do not mean equal absolute totals.

Changed question: Discuss why sample size can affect uncertainty.

Worked example 19: Different percentages and counts

Question: Sample A has50%, n=50; sample B has25%, n=200. Which has more supporters?

Solution or interpretation: A25, B50, so B has more despite smaller proportion.

What it shows: Count=fraction×total.

Wrong inference: Do not compare percentages alone when asked for number.

Changed question: Compare relative shares instead.

Worked example 20: Missing selected invitees

Question: A survey sends100 invitations; seventy-two submit valid forms. Find number not completing and response rate.

Solution or interpretation: 28 do not complete; response rate72%.

What it shows: Counts sum100.

Wrong inference: The result tells nothing about which option those respondents selected.

Changed question: What if six submitted forms were unusable?

Worked example 21: Usable response distinction

Question: Eighty invitations generate60 returns, but five forms lack the key answer. What is usable response count for that question and usable rate?

Solution or interpretation: Usable count55; usable rate55/80=68.75%. The return rate60/80=75% is different.

What it shows: The denominator of a category percentage should match available answers to that question.

Wrong inference: Do not silently treat incomplete records as favouring an option.

Changed question: If33 of55 favour A, find its share.

Worked example 22: Respondent support from usable forms

Question: Among55 usable answers,33 favour A. Find result percentage.

Solution or interpretation: 33/55=60% of usable respondents.

What it shows: The count and total are explicitly identified.

Wrong inference: This does not establish60% of all invited or population.

Changed question: Name the 25 unclassified invitees.

Worked example 23: A sampling-frame correction

Question: A school list excludes part-time programme pupils. A selection procedure is otherwise perfectly random. Is school-wide representation ensured?

Solution or interpretation: No. Undercoverage in the frame remains even with random selection from the incomplete list.

What it shows: All target population members need appropriate inclusion chance.

Wrong inference: A valid random number generator cannot select missing names.

Changed question: State the necessary scope correction.

Worked example 24: The survey question and recorded variable

Question: A form asks ‘preferred main study location’ with labels home, library, school. Is an arithmetic mean of these names meaningful?

Solution or interpretation: No. These are nominal categories; mode and percentages are meaningful, not an arithmetic mean of arbitrary codes.

What it shows: Numerical labels assigned to categories do not create quantitative meaning.

Wrong inference: Do not report mean location code2 as a real average location.

Changed question: Find the mode if counts are20,12,8.

Worked example 25: A missing category

Question: Forty exclusive responses have counts18,12,5 and unknown other. Find other count and share.

Solution or interpretation: Other=40−18−12−5=5; share5/40=12.5%.

What it shows: All category counts sum40.

Wrong inference: The subtraction assumes one response per person.

Changed question: What if multiple answers are allowed?

Worked example 26: Respondents versus invited percent

Question: An online survey invites80; fifty reply; thirty favour A. Find respondent support and confirmed-invitee share.

Solution or interpretation: Respondent share30/50=60%. Confirmed invitee share30/80=37.5%.

What it shows: They answer two distinct denominator questions.

Wrong inference: Neither alone equals population-wide support.

Changed question: State both results in complete sentences.

Worked example 27: A changing survey location

Question: One researcher surveys only a sports field, another surveys a whole-school random list. Why may responses differ?

Solution or interpretation: Sports-field visitors may differ systematically from those absent; selection location influences who is represented.

What it shows: The question itself could be identical.

Wrong inference: Do not conclude the sports-field responses are false; only that broad inference is limited.

Changed question: Design a simple mixed-location method.

Worked example 28: Anonymity and useful measurement

Question: A hypothetical school revision survey asks for names, phone numbers and preferred topic, though only overall topic shares are needed. Which fields are unnecessary?

Solution or interpretation: Names and phone numbers are not necessary to calculate aggregate topic percentages, absent another legitimate purpose. A safe classroom exercise can collect anonymous preferences.

What it shows: Collect only relevant data under applicable school procedures.

Wrong inference: More personal information does not automatically improve a statistical calculation.

Changed question: Draft a minimal anonymous form.

Worked example 29: A future estimate, not a count

Question: One hundred sampled students yield sixty-two positive responses. Can we say exactly620 of1000 school students are positive?

Solution or interpretation: No. Multiplying0.62×1000=620 is an extrapolated estimate under assumptions, not a measured count.

What it shows: Only62 sampled responses were observed.

Wrong inference: Do not report an estimate as verified school census data.

Changed question: Write a cautious report sentence.

Worked example 30: Sampling one day only

Question: A school surveys after-school visitors only on Friday. What concern arises?

Solution or interpretation: Friday attendance and activities may differ from other days; the time frame can affect which students appear.

What it shows: A broader timeframe could cover more student patterns.

Wrong inference: Do not assume one day’s attendance is representative of the week.

Changed question: Propose data collection across several days.

Worked example 31: Percentages can look precise while the sample is weak

Question: An open poll of twelve responses has nine favouring A. Report the observed percentage and a limitation.

Solution or interpretation: 9/12=75% of poll respondents. It is based on only twelve voluntary responses and may not describe the target population.

What it shows: The exact sample arithmetic does not imply exact population share.

Wrong inference: Do not hide n=12 behind a large-font75% graphic.

Changed question: Draft a report caption.

Ten ways good arithmetic can lead to a bad conclusion

1. The sample percentage is reported as a census

Error: A child finds62/100=62% and writes ‘62% of all1000 students prefer walking.’ This overstates what was observed. The exact result is for the hundred sampled students; population inference is an estimate requiring assumptions.

Corrective action: Write sample claim and population estimate as separate sentences.

Independent evidence: The student labels population N and sample n correctly on a changed question.

2. Response rate and support rate confused

Error: Eighty invited, fifty-six reply, and thirty-five favour A. The learner calls70% support because56/80=70%. That is response rate, while support among respondents is35/56=62.5%.

Corrective action: Make three columns: invited, usable responses, favourable replies.

Independent evidence: The next result uses an explicitly named denominator.

3. A large convenience sample is assumed unbiased

Error: Thousands of self-selected online responses are treated as representative simply because the number is large. Explain that increasing size can reduce some random variability but not automatically repair systematic selection bias.

Corrective action: Contrast an unbiased small frame with a huge one-club voluntary poll.

Independent evidence: The learner identifies the actual selection mechanism before interpreting percentages.

4. The sampling frame misses target members

Error: A whole-school question samples only library visitors. Students who never visit the library are missing from the frame. A random lottery inside the library list does not fix that exclusion.

Corrective action: Compare target population and available contact list.

Independent evidence: An improved method gives all intended students a reasonable chance to be sampled.

5. A leading question is treated as neutral

Error: A survey asks whether a ‘wonderful’ programme should continue, then presents support as unprompted. The value judgement can influence response. Rewrite without praise and provide a balanced response scale.

Corrective action: Identify wording that suggests the desired answer.

Independent evidence: An unfamiliar survey question is revised neutrally.

6. Two issues are combined into one yes/no

Error: A learner reports one yes as support for both later start and longer lessons. Respondents might favour only one. Split the question and interpret each answer separately.

Corrective action: Draft two single-topic items.

Independent evidence: The new instrument measures precisely defined preferences.

7. A multiple-choice respondent is counted twice in an exclusive pie

Error: A student selects both bus and walk for a journey, and both counts appear in a pie allegedly partitioning forty students. That can exceed the total. Define a single main-mode answer or change the denominator to reported selections.

Corrective action: Count category totals and compare with the survey’s response rule.

Independent evidence: The final display has a clearly defined whole.

8. An imbalanced sample is used as the population mix

Error: A sample with75% group A is treated as reflecting a population with40% group A. Raw sample share may be weighted toward A’s different preferences. Explain sample composition before considering a weighted population estimate.

Corrective action: Calculate raw65% and illustrative weighted44% in the supplied example.

Independent evidence: The pupil sees why a correct sample percentage can still mislead as a population estimate.

9. Nonrespondents are implicitly assumed to agree

Error: Forty of sixty respondents support A, while twenty invitees never answered. A report treats all eighty invited as supporters of the respondent percentage. The nonresponses contain no recorded answer.

Corrective action: Show simple invitee bounds based on all nonrespondents disagreeing or agreeing.

Independent evidence: The new report distinguishes known and unknown counts.

10. Precision in a chart replaces honest methodological notes

Error: A polished pie chart has percentages to one decimal place but no survey N, collection method or question text. Arithmetic neatness does not prove validity. Add a clear source, total, wording and relevant method caveat.

Corrective action: Write a concise caption for a toy sample.

Independent evidence: The chart reader can tell who actually responded and what was asked.

Forty-eight quick checks with reasons

Quick check 1: Population1000, sample100: sample fraction

Answer: 10%. Reason: 100/1000.

Reporting habit: state the group and denominator behind the percentage before claiming what the number means. Distinguish a measured response count from an extrapolation to people who were not observed.

Quick check 2: Sample100,62 support: observed share

Answer: 62%. Reason: 62/100.

Quick check 3: Does that prove620 of1000 support?

Answer: No. Reason: Extrapolation estimate only.

Quick check 4: Population1200, sample120

Answer: 10%. Reason: 120/1200.

Quick check 5: Invited80, replies56: response rate

Answer: 70%. Reason: 56/80.

Quick check 6: Invited80, replies56: nonreply count

Answer: 24. Reason: 80−56.

Quick check 7: Of56 replies35 support: share

Answer: 62.5%. Reason: 35/56.

Quick check 8: Of56 replies35 support: not-support count

Answer: 21. Reason: 56−35.

Quick check 9: Invited80, reply60, support40: respondent share

Answer: 66⅔%. Reason: 40/60.

Quick check 10: Same invitees min confirmed support fraction

Answer: 50%. Reason: 40/80.

Quick check 11: Same invitees max possible support fraction

Answer: 75%. Reason: 60/80.

Quick check 12: Census of a 40-person class if all40 reply?

Answer: Yes. Reason: Whole defined class.

Quick check 13: Census of an800-person school if only40 reply?

Answer: No. Reason: Small subset.

Quick check 14: Library visitors only: random school sample?

Answer: No. Reason: Frame limited.

Quick check 15: Voluntary online poll unbiased by size alone?

Answer: No. Reason: Self-selection.

Quick check 16: Group A40%, group B60%: sample50 split

Answer: 20 and30. Reason: Proportional.

Quick check 17: Group A sample30 with24 support: share

Answer: 80%. Reason: 24/30.

Quick check 18: Group B sample10 with2 support: share

Answer: 20%. Reason: 2/10.

Quick check 19: Raw support of24+2 among40

Answer: 65%. Reason: 26/40.

Quick check 20: Population mix40%A60%B weighted estimate

Answer: 44%. Reason: 0.4×0.8+0.6×0.2.

Quick check 21: Response rate72 usable of100

Answer: 72%. Reason: 72/100.

Quick check 22: Missing invitations after72 returns of100

Answer: 28. Reason: 100−72.

Quick check 23: 80 invited,60 returned: return rate

Answer: 75%. Reason: 60/80.

Quick check 24: 80 invited,55 usable: usable rate

Answer: 68.75%. Reason: 55/80.

Quick check 25: 33 favourable among55 usable

Answer: 60%. Reason: 33/55.

Quick check 26: Sample n20,60% support: count

Answer: 12. Reason: 0.6×20.

Quick check 27: Sample n200,60% support: count

Answer: 120. Reason: 0.6×200.

Quick check 28: Sample n50,50% support: count

Answer: 25. Reason: Half.

Quick check 29: Sample n200,25% support: count

Answer: 50. Reason: Quarter.

Quick check 30: 40 exclusive counts18,12,5,x: x

Answer: 5. Reason: Total40.

Quick check 31: Remaining5 of40 percentage

Answer: 12.5%. Reason: 5/40.

Quick check 32: Invited80, reply50, favour30: respondent share

Answer: 60%. Reason: 30/50.

Quick check 33: Same confirmed favour among invitees

Answer: 37.5%. Reason: 30/80.

Quick check 34: Poll12 replies9 favourable

Answer: 75% of respondents. Reason: 9/12.

Quick check 35: Does75% of12 prove75% entire population?

Answer: No. Reason: Selection uncertainty.

Quick check 36: If one student selects two categories, exclusive pie valid?

Answer: No. Reason: Categories overlap.

Quick check 37: Multiple-response categories may sum above100%?

Answer: Yes. Reason: Several selections per person.

Quick check 38: Does random selection eliminate all nonresponse bias?

Answer: No. Reason: Respondents may differ.

Quick check 39: Can two equal percent results have different counts?

Answer: Yes. Reason: Different sample N.

Quick check 40: Is support/usable replies same as response rate?

Answer: No. Reason: Different denominator.

Quick check 41: A leading adjective in survey: possible bias?

Answer: Yes. Reason: Wording influence.

Quick check 42: A double-barrelled yes/no is unambiguous?

Answer: No. Reason: Two decisions.

Quick check 43: Stratified 40/60 sample of100

Answer: 40 and60. Reason: Proportional.

Quick check 44: Does stratification alone guarantee unbiased within strata?

Answer: No. Reason: Selection still matters.

Quick check 45: Anonymity is useful when names unnecessary?

Answer: Yes. Reason: Data minimisation.

Quick check 46: Which group does a schoolwide claim describe?

Answer: Population. Reason: Target group.

Quick check 47: Which group does35 of56 describe?

Answer: Respondents. Reason: Observed sample.

Quick check 48: Is pie-chart title ‘School opinions’ enough without N?

Answer: Not ideal. Reason: Missing context.

Quick check 49: Does exact percentage mean exact population fact?

Answer: No. Reason: Unobserved members.

Four weeks of useful statistical reasoning

Week 1: separate population, sample and respondent counts

Use simple fictional school counts, calculate sample fractions and response rates, and label exactly whose views each fraction describes. Check whether a class census differs from a whole-school sample.

Independent progress check: A new question receives the correct denominator and a complete statement of the observed group.

Week 2: compare sampling methods and frames

Contrast a random choice from a complete student list with a club-only convenience sample and an open poll. Write one possible systematic exclusion for each weak design. Avoid unsupported claims that a larger number automatically fixes the problem.

Independent progress check: The pupil identifies the selection method before trusting a percentage.

Week 3: survey wording, response and categories

Rewrite leading and double-barrelled questions, distinguish one-answer and multi-answer categories, and calculate nonresponse counts and simple arithmetic bounds for invitees.

Independent progress check: The learner writes a fairer question and an honest response summary with N.

Week 4: interpret reports with appropriate caution

Use mixed tables containing sample composition, group-specific rates and pie-chart results. Treat weighting and extrapolation as extensions, explaining assumptions. Finish with a short method note that a parent could verify independently.

Independent progress check: The final report separates observed facts, calculations, estimates and limitations.

A three-pupil 90-minute data lesson

The original eduKateSG Clementi Mathematics reference describes three-pupil small-group tutorials lasting ninety minutes. A teaching sequence might begin with ten minutes of denominator diagnosis, fifteen minutes distinguishing sample and population, twenty minutes of guided response-rate questions, twenty-five minutes comparing survey designs and revised question wording, ten minutes reading graphs and ten minutes writing a cautious summary. Class availability is not implied.

One learner may calculate35/80 when a question asks among respondents, another might label an open poll as random, and a third may see a high percentage without noticing only twelve people answered. All three can benefit from different feedback. A good small-group tutorial makes each pupil’s first interpretation visible and tests a fresh scenario without prompts.

Questions parents can ask about a chart or survey

Parent question	What it tests
Who belongs to the population?	A clearly defined target group.
Who was selected and from what list?	Sampling frame and method.
How many people actually responded?	The observed denominator.
Is this a response rate or opinion share?	Correct fraction meaning.
Could some students have been excluded?	Undercoverage and convenience bias.
Is the wording leading?	Measurement fairness.
What data are we missing?	Nonresponse uncertainty.
Is that conclusion a sample fact or population estimate?	Appropriate confidence and explanation.

Frequently asked questions

1. What is the difference between a population and a sample?

The population is the full group the study aims to describe. A sample is a smaller selection measured to learn about it. A survey of one class may be a census of that class but only a sample of the whole school.

2. What makes a sample representative?

Its selection and resulting composition should meaningfully reflect the target population for the question. Random selection from an appropriate frame can help, but nonresponse, undercoverage and chance differences remain. No simple headcount alone guarantees representativeness.

3. Does a bigger sample always remove survey bias?

No. A larger sample may reduce some random variation under a sound sampling design, but many responses from an unrepresentative frame can preserve systematic selection bias.

4. What is a response rate?

The number of usable answers received divided by the number invited or eligible under the stated definition, expressed as a percentage. Distinguish invited, returned and usable forms when some responses are incomplete.

5. Why is a voluntary online poll different from random sampling?

People choose whether to see or answer the poll, so participation chances may differ by interest, availability and access. Its results describe respondents, not automatically the entire population.

6. What is a leading question?

A question that suggests or emotionally favours an answer, such as calling a proposal ‘excellent’ before asking for support. Neutral wording reduces this particular influence, though it cannot repair every other bias.

7. What is nonresponse bias?

It can occur when the people who fail to answer differ systematically from those who do on the variable being measured. Good random invitation selection does not ensure that every invited person replies.

8. Why does the denominator matter so much?

A percentage among respondents uses their usable response total, while response rate uses invitations, and a sampling fraction uses population size. Mixing denominators changes the meaning of the result even when the arithmetic is correct.

9. What is stratified sampling?

A method that divides a target population into relevant groups and selects a sample within each, sometimes in proportion to their population sizes. It may improve coverage of chosen groups but does not automatically solve bias within those groups.

10. Can a sample percentage be used to estimate a school total?

Under suitable sampling assumptions it can inform an estimate, but it is not an exact count of people not observed. State the observed count, sample size and uncertainty rather than implying a census.

11. Should school surveys collect names and private details?

Only when there is a legitimate reason and applicable rules permit it. Many Mathematics exercises need only anonymous aggregate preferences. Avoid collecting unnecessary identifying or sensitive information.

12. Is sampling bias a Secondary 2 G2 examination chapter?

Data interpretation and statistics are curriculum-connected. Formal survey design, weighting and advanced inference may be enrichment rather than tested content in a particular class. Match practice to the student’s current school scope.

13. How can parents help check a chart?

Ask who the survey was meant to describe, who actually replied, the question wording, the denominator used and which conclusions are facts rather than estimates.

14. Does the lesson guarantee an outcome?

No. It develops statistical literacy and mathematical communication, while school results and subject-level decisions depend on broader assessment and readiness.

One practical conclusion

Did you know? A percentage can be completely correct yet answer the wrong question if its sample and denominator are unclear. Strong data literacy begins with the population, the selection method, the actual replies and the wording. Teaching a child to distinguish what was observed from what is only estimated is a valuable habit far beyond one Mathematics topic.

The original Clementi three-pupil Mathematics tutorial reference remains unchanged. For current class arrangements, contact eduKate Singapore directly. No grade or subject-level result can be guaranteed.

Explore the other three Secondary 2 G2 Mathematics guides in this batch

This batch connects exact fractional equations, irregular polygon angle reasoning, reliable statistical samples and percentage changes. These are different skills that share one central habit: define the mathematical quantity, use the correct relation and check the answer in its original context. Choose the guide that matches a current school problem and attempt a changed example after reading it.

Secondary 2 G2 Mathematics Tutor Bukit Timah | Choosing Reliable Samples and Detecting Survey Bias