Did you know a class survey can rise from 60% agreement to 75% agreement without proving that any individual pupil changed their mind? The percentages are accurate, yet the story a writer attaches to them can be much too confident. The missing question is wonderfully simple: did the same people answer both times?
Two repeated surveys measure responses at two times; if different pupils answer each round, the results are comparisons of two respondent groups, not a record of how particular pupils changed their views. Calculate each percentage with its own denominator, check question wording and respondent mix, and describe the observed group-level difference. Do not claim that an individual changed their opinion unless their before-and-after responses were actually linked and analysed appropriately.
A diagnostic question before any arithmetic
Show the learner the following fictional statement: “In January, 12 of 20 respondents supported a longer library opening period. In March, 15 of 20 respondents supported it. Therefore, three pupils changed their minds.” Ask whether the conclusion is certain. A pupil who answers yes has noticed the net difference in supportive responses but has not checked who was interviewed. Let the student identify the information missing from the description.
Even if the exact same twenty pupils had answered twice, the total number of Yes responses alone would not prove that exactly three people switched. For example, five people could switch from No to Yes while two switch from Yes to No. The net change would still be three. If the respondents are different, we have even less information about individual changes. That is the conceptual problem this lesson repairs.
The distinction: snapshot comparison versus paired change
A cross-sectional survey describes a group of respondents at a particular point in time. Conducting a similar survey later can show a difference between two snapshots, but the individual respondents may differ. A longitudinal or panel-style design follows the same people across time, subject to consent, loss to follow-up and careful matching. The US Centers for Disease Control and Prevention describes these cross-sectional and longitudinal distinctions in its National Center for Health Statistics survey overview.
Neither design is automatically worthless or perfect. A well-designed repeated cross-sectional survey can track change in a defined population. A matched panel can examine who changed answers, but the repeated questioning itself can sometimes affect responses, and those who drop out may differ from those who remain. The useful habit for school projects is to name what the chosen design can—and cannot—show.
An original worked example: a school reading-corner suggestion
First define the question and the groups
Illustrative invented data, not actual school survey findings: a fictional class asks “Would you support making the reading corner available at lunchtime?” In Round 1, twenty pupils respond, twelve Yes and eight No. In Round 2, twenty pupils respond, fifteen Yes and five No. The same wording and response choices are used, but only ten pupils answered in both rounds. We do not have a matched response record for those ten. There are no blank responses in this teaching example.
Calculate what is genuinely supported
Round 1 agreement = 12 ÷ 20 × 100 = 60%. Round 2 agreement = 15 ÷ 20 × 100 = 75%. The difference is 75% − 60% = 15 percentage points. This is a change in the observed share of Yes answers among each round’s respondents. It is not proof that exactly fifteen percent of a fixed group changed its mind.
Check the counts separately: Round 1 has 12 + 8 = 20 responses; Round 2 has 15 + 5 = 20. The arithmetic is correct. What is not justified is the statement “Three pupils were persuaded.” The identities and individual before-after answers needed to prove that are missing.
Explain the denominator and rate distinction
The word percentage points describes an arithmetic difference between percentages. A relative percentage increase in the agreement share would be (75 − 60) ÷ 60 × 100 = 25%. These are different ways to describe the same two percentages, and neither establishes what individual pupils did. Use the first when explaining that the observed agreement share rose by fifteen percentage points; if reporting relative change, name the 60% baseline explicitly.
Why the same twenty respondents would still need paired information
Imagine a separate hypothetical version in which precisely the same twenty pupils answered both rounds, but the teacher only kept the two total counts: twelve Yes in Round 1 and fifteen Yes in Round 2. One possible pattern is that three initial No voters became Yes voters, with everyone else unchanged. A different possible pattern is that five No voters became Yes voters while two Yes voters became No voters. Both produce the same totals. Only matched responses can distinguish these patterns.
A small paired illustration makes the point concrete. Among eight fictional pupils: two answer Yes both times, two switch Yes to No, three switch No to Yes, and one answers No both times. The first survey has four Yes (2 + 2) and the second has five Yes (2 + 3), so the net increase is one, even though five individuals changed their answers. An aggregate net change is not the same as a headcount of switches.
The five questions to ask before claiming attitudes changed
- Who was eligible? Was the survey for one class, one year level or a wider group? Do not generalise a convenience sample to a whole school without adequate grounds.
- Who actually responded? Record the count and relevant group composition at each round. The number may match even while the people differ.
- Was the wording identical? Changing the question, answer scale, survey mode or context can change what a response means.
- Is this a snapshot or matched follow-up? A snapshot comparison permits statements about percentages among each round’s respondents; properly linked paired data permit more specific individual-change analysis.
- What uncertainties remain? Nonresponse, self-selection, changing class membership, repeated-question effects and small numbers limit stronger interpretations.
What a careful report can say instead
Use wording such as: “The share supporting lunchtime access among respondents was 60% in Round 1 (12 of 20) and 75% in Round 2 (15 of 20), a difference of 15 percentage points. Because different pupils responded and individual answers were not linked, this does not establish how many pupils changed their views.” That is more informative than a dramatic but unsupported claim about persuasion.
Do not pretend that collecting personally identifying information is always necessary. A small classroom project can avoid tracking individuals and simply state that it compares anonymous snapshots. If longitudinal tracking is genuinely required and authorised, minimise personal data, use appropriate permissions and safeguards, and recognise that pseudonymous identifiers may still permit linkage to people. Never promise impossible anonymity in a very small group.
How better survey design changes what you can infer
If the research question is “What proportion of a given year group supports the reading-corner idea this term?”, representative snapshots may be adequate. If the question is “Which pupils changed their minds after reading more information?”, a properly designed matched before-and-after study is needed. But even with matching, a pre/post design without a suitable comparison cannot automatically establish that the intervening information caused the change. Other events, discussion and repeated questioning can matter.
Pew Research Center explains in its survey-consistency discussion why repeating the same methodology helps make trend comparisons more credible, while still not guaranteeing that the result is accurate. For school learners, the transferable lesson is methodological honesty: ask whether apparent change could reflect differences in who answered or how the question was asked.
Practice in three increasingly independent stages
Guided: read the percentage but limit the claim
Two lunchtime surveys ask the same question. In the first, 9 of 15 pupils answer Yes (60%). In the second, 14 of 20 answer Yes (70%). The student calculates each percentage using its own denominator and reports a 10-percentage-point difference. Ask whether the extra five Yes responses proves five individual mind changes. It does not: the response counts differ, and the participating pupils have not been matched.
Paired: separate a population claim from an individual claim
Partner A writes a plausible sentence: “More respondents favoured the idea in the second survey.” Partner B checks whether “more” means a higher count, a higher percentage, or both. Then they rewrite “The whole school became more supportive” as a bounded statement about the sampled respondents. Swap roles and challenge whether equal sample sizes imply identical people. They do not.
Independent: construct two possible histories
A fictional matched set of ten pupils has six Yes before and seven Yes after. Give two different plausible switch patterns. Pattern 1: one No-to-Yes switch and no Yes-to-No switches. Pattern 2: two No-to-Yes switches and one Yes-to-No switch. Both yield seven Yes at the end. A learner who can produce two consistent histories understands why marginal totals alone cannot reveal all transitions.
For an extension, ask the student to sketch a four-cell paired table with Yes→Yes, Yes→No, No→Yes and No→No. Check that the four cells add to the number of matched respondents and reproduce the before and after Yes totals. The additional structure is not mere presentation: it contains the evidence required to describe individual transitions.
How parents and teachers can observe progress
Start with a diagnostic sentence asserting that “three pupils changed their mind” from a 12-to-15 Yes count. After teaching, present a new scenario with unequal response counts or partial overlap and see whether the learner spontaneously asks who answered and what records exist. That question is evidence of real statistical literacy. Merely calculating two correct percentages is not enough.
A teacher can assess three separate skills: numerical accuracy, appropriate scope of the conclusion, and recognition of missing design information. Parents can practise with harmless examples such as choosing a class trip activity, while respecting pupil privacy. Keep fictional scenarios clearly labelled as fictional and do not publish small identifiable student response data.
Limits and common misconceptions
A higher second percentage can reflect genuine change in the eligible population, changes in who responded, random sampling variation, changes in wording or several influences together. With only two tiny samples, it may be impossible to choose between explanations. Likewise, a matched survey may have missing follow-ups; analysing only those who returned can favour pupils whose experiences differ from those absent. Strong statistical claims require design and analysis beyond two displayed percentages.
There is also a distinction between more supporting responses and more support in the underlying population. The first describes your observed data; the second is an inference about people not necessarily observed. Students should practise saying what their data establish before arguing about possible causes.
Frequently asked questions
If the sample size is equal in both rounds, are we following the same pupils?
No. Two groups of twenty can contain wholly different people, partially overlapping people or exactly the same people. The survey record must tell you which case applies.
If agreement increases by three people, did exactly three change their mind?
Not necessarily. That number is a net difference in Yes counts. More pupils may have switched in opposite directions, and if the respondents differ there may be no linked individual change record at all.
Should a class always collect names to match surveys?
No. Collect personal data only when justified, properly authorised and protected. If the goal is a group snapshot, an unlinked survey can be more appropriate, with honest limits clearly described.
Does 60% to 75% mean a 15% increase?
It means a rise of 15 percentage points. Relative to the original 60% share, the increase is 25%. Name the measure you mean rather than using “percent” ambiguously.
Where to deepen the learning
Compare this lesson with the existing guide to surveys with unequal response counts and the explanation of how respondent group mix changes a headline percentage. Families will also find structured resources at the eduKate Education Hub. The unchanged Secondary 1 Mathematics tutorial reference demonstrates how to move from diagnostic questions to checked independent explanations.
The aim is not to make learners afraid of percentages. It is to make them curious about what those percentages can actually tell us. Once pupils ask who answered, how questions were asked, and what comparison was possible, their conclusions become both more modest and more powerful.