A class poll tells you how the recorded respondents answered. It does not automatically tell you what the whole school thinks. The gap is not repaired by a colourful chart, a larger percentage or more replies from the same friendship group. Before making a school-wide claim, identify the population, the route by which people entered the poll, the people who did not answer and the exact question they received.
This guide follows one school-project failure from an impressive headline back to the evidence. All students, schools, response counts and calculations in the examples are fictional. The aim is to help learners repair a conclusion, not to dismiss the voices they have already collected.
The percentage is right. The headline is not.
Alicia’s group wants the library to offer a Friday makerspace session. They ask classmates whether they support the proposal. Eighteen of twenty-four respondents say yes. The calculation is correct: eighteen divided by twenty-four is 0.75, or seventy-five percent.
The presentation announces, “Seventy-five percent of our school wants a makerspace.” Nothing went wrong on the calculator. The error occurred in the noun after the percentage. Twenty-four respondents from one class became the entire school.
Perhaps the school-wide percentage really is seventy-five. Perhaps it is higher or lower. The poll, as described, does not settle that question. Agreement between a convenient sample and the population would be fortunate, not something established by the method.
A better first sentence is, “Eighteen of the twenty-four classmates who answered supported the proposal.” This preserves the result while removing the unsupported leap. The group can still explain why it wants a wider consultation. Correcting the headline does not erase the work; it makes the work usable.
Keep four groups separate
The population is the group the project wants to describe. The sampling frame is the list or practical route from which people can be selected. The selected or invited group contains those approached. The respondents are those who actually provide usable answers. These groups can overlap without being identical.
For a school-wide project, the population might be all current students. A class register covers only one class. An invitation sent through a club chat reaches only people with access to that chat. The final responses may come from a smaller, self-selected group again.
AAPOR’s survey-research guidance explains the role of the sampling frame and distinguishes probability-based selection from opt-in recruitment. For this project, the practical question is simple: what route allowed each student to be heard?
Draw that route before calculating. “School → one class → invitation → completed response” immediately exposes the point at which most of the school disappears.
Missing replies are not votes against the idea
Suppose Alicia’s class contains thirty students, all of whom were invited. Twenty-four answer; eighteen support the proposal and six oppose it. Six students do not answer.
The support proportion among respondents is eighteen out of twenty-four, or seventy-five percent. The known supporters form eighteen out of thirty, or sixty percent of the class. Those statements use different denominators and answer different questions.
Calling the remaining twelve students opponents would be wrong. Six explicitly opposed the proposal; six supplied no answer. Silence might reflect absence, a missed message, lack of interest, uncertainty or another reason. The recorded data do not identify which.
Keep “no response” separate from “no”, “not sure” and “I prefer not to answer”. These are different observations. A blank cell should not silently become a negative vote just because the spreadsheet needs a number.
This is where language and Mathematics meet. A denominator is not merely the number under a fraction bar. It names the group about which the fraction makes a statement.
What the missing answers could change
Stay with the thirty-student example. Assume each recorded answer is valid, every student has one yes-or-no preference for the same proposal at that time, and only the six missing preferences remain unknown.
If all six missing students oppose the idea, the class has eighteen supporters out of thirty: sixty percent. If all six support it, the class has twenty-four supporters out of thirty: eighty percent. Under these assumptions, the class-wide support proportion lies between sixty and eighty percent.
This is a logical range from missing information, not a statistical confidence interval. It does not assign probabilities to the possibilities. It also says nothing about other classes.
Now change the question to whether a majority of this class supports the proposal. Eighteen known supporters already exceed half of thirty. The majority conclusion survives every possible allocation of the six missing preferences under the stated assumptions. The exact percentage does not.
A stronger learner therefore asks which conclusion the uncertainty affects. Missing data do not automatically make every statement impossible. They limit particular statements in particular ways.
More replies can leave the same people missing
Alicia’s group recruits fifty more responses through the makerspace club. The sample is larger. Yet students already interested in making things may now have even more influence over the result.
That does not prove the new respondents answered dishonestly. It shows why honest answers can still produce an unrepresentative collection. The problem concerns who entered the evidence, not the moral character of the people who spoke.
Statistics Canada’s explanation of non-probability sampling describes convenience and volunteer approaches and the care needed when generalising from them. A student should not attach a conventional random-sample margin of error to a club-chat poll without the required design and assumptions.
The next useful action is not automatically “collect more”. It may be “change the route through which students are invited”. Fifty additional voices from an already well-heard group do not supply the missing experiences of a group that was never reached.
Two groups can distort a combined percentage
Consider a different fictional school with three hundred younger students and three hundred older students. A project receives eighty responses from older students and twenty from younger students. Sixty-four older respondents and eight younger respondents support a proposal.
Within the responding older group, support is sixty-four out of eighty: eighty percent. Within the responding younger group, support is eight out of twenty: forty percent. Pooled together, support is seventy-two out of one hundred: seventy-two percent.
The pooled result gives older respondents four times as much representation as younger respondents, although the school contains equal numbers of each. That imbalance matters because the recorded support rates differ.
If, as an additional assumption, each group’s responding students adequately represented that entire age group, giving the two equal-sized groups equal weight would yield sixty percent: half of eighty plus half of forty.
The phrase “as an additional assumption” is essential. Reweighting cannot establish that the responding younger students resemble the missing younger students. The calculation illustrates composition; it does not certify a convenience poll.
Do not turn weighting into a rescue button
Imagine that all twenty younger respondents belong to the same after-school club. Giving each response more weight still leaves the project dependent on that club’s experiences. A multiplier does not create answers from students who were never represented.
For a school assignment, it may be better to report the groups separately and acknowledge the recruitment limitation than to produce a sophisticated-looking adjusted total. Use formal weighting only with a method you understand and appropriate teacher guidance.
Where adjustment is legitimate, show the original counts, the population information used and the assumptions. Readers should be able to tell which numbers were observed and which were calculated.
The wider statistical machinery belongs in How Surveys and Sampling Work. The narrower job here is to stop a student-project percentage from silently changing the people it describes.
“Random” describes a selection procedure
“We asked random people in the corridor” often means the group did not deliberately choose particular friends. That is not enough to establish a probability sample of the school.
The corridor, time and invitation process still determine who is reachable. Students in another building, on a different timetable or absent that day may have no opportunity to enter the sample.
For a simple random sample from a complete student list, an appropriate random procedure would give each student an equal chance of selection. Other valid probability designs can use unequal but known selection probabilities. Randomisation is not a promise that the realised sample will perfectly match every population characteristic.
A school project should use an approved selection process organised with the teacher. Students should not acquire or circulate private registers themselves. The important learning is to describe the actual selection rule, rather than use “random” as a synonym for casual.
Inviting everybody does not guarantee hearing everybody
Suppose the link is sent to every student. This repairs one part of the earlier problem: the invitation is no longer restricted to one class. However, the final respondents can still differ from the students who do not reply.
An attempted census is not a completed census simply because every invitation was delivered. Equally, a low response count does not reveal the exact direction or size of any bias. The missing students’ views remain unknown unless there is further evidence.
For a practical review, ask which groups had unusually few responses and whether the collection method created avoidable barriers. A brief, accessible alternative response route may help. Follow school arrangements and respect a student’s choice not to participate.
The project should report both reach and participation. “Invited all six hundred students; received one hundred and twenty responses” is more informative than presenting one hundred and twenty as though it were the entire school.
The question can change what a yes means
Compare these invented questions: “Do you support an exciting makerspace that will improve our school?” and “Which use of the library’s Friday session would you prefer: a makerspace, quiet reading, another activity, or no preference?”
The first attaches an attractive judgement and an unestablished benefit to the proposal. The second asks for a choice among uses of the same session. They are not interchangeable measurements.
Pew Research Center’s questionnaire guidance explains why wording, answer options and question order matter. It also recommends asking about one concept at a time. A carefully selected sample cannot repair a question that respondents interpret differently.
Before collecting the final responses, ask a few suitable peers to explain what they think the question asks. Use their feedback to remove ambiguity. This is a comprehension check on the instrument, not an opportunity to coach people towards the group’s preferred answer.
Support, attendance and usefulness are different outcomes
A student may support having a makerspace without intending to attend. Another may want to attend but be unavailable on Fridays. A third may attend once and later decide it is not useful.
Therefore, “supports the idea”, “expects to attend”, “actually attends” and “benefits from attending” are separate claims. A yes-or-no preference question cannot supply all four.
Choose the evidence for the decision. A planning group estimating attendance needs a question about likely attendance under specified conditions, while recognising that intentions are not observed behaviour. A later evaluation of attendance needs attendance records collected appropriately.
This is a useful cross-subject lesson: the outcome definition determines what the number means. It is the same reason a graph of completed worksheets cannot, by itself, establish how much a student learned.
Audit the poll before replacing it
Begin with the proposed conclusion. Underline the group it describes and the outcome it claims. Then reconstruct the collection route: who could receive the question, who was invited, who answered and what happened to incomplete responses.
Next inspect the exact question and options. Could a student reasonably interpret a term in more than one way? Did the question combine two proposals? Was a missing answer counted as opposition?
Finally, recalculate the headline from the recorded counts and name the denominator. This sequence often reveals a repair that does not require discarding everything. The respondents’ views can still be reported accurately.
Only then decide whether the project needs new evidence. Repeating the same recruitment method with a fresh online form may recreate the original weakness while making the activity look like a restart.
A repaired paragraph for the presentation
Here is a defensible version of Alicia’s fictional result: “We invited all thirty students in our class to answer a question about a Friday makerspace. Twenty-four responded. Eighteen supported the proposal and six opposed it. The six missing responses were not counted as either support or opposition. Our respondents came from one class, so we cannot use this poll alone to estimate school-wide support.”
The group can continue: “The result gives us a reason to investigate interest more widely. Before making a school-wide recommendation, we would use a teacher-approved process to reach other classes and clarify whether students support the idea, expect to attend or prefer another use of the session.”
This paragraph separates observation, limitation and next action. It does not pretend that uncertainty makes the project worthless. It makes the proposed next step follow from the exact weakness in the evidence.
Use a disagreement as a diagnostic clue
Another class reports only thirty percent support. Alicia’s group should not immediately decide that the other class conducted a bad poll. The two groups might genuinely have different preferences.
Compare the proposals, wording, timing, invitation process and denominators. Was one poll about an optional lunchtime session and the other about replacing quiet reading? Did one include only students who already used the library?
Only after checking those details should the results be compared. Do not average percentages from differently sized responding groups without deciding what should receive equal weight. Do not merge answers to materially different questions into one total.
A disagreement can reveal a genuine difference, a measurement difference or both. The point is to investigate rather than choose the result that makes the preferred proposal easiest to defend.
Keep the project proportionate and private
A simple classroom investigation does not need a professional polling operation. It does need an honest description of its limits. Follow the assignment’s requirements and the school’s rules for collecting responses.
Ask only for information necessary to the project. Avoid collecting names, contact details or sensitive personal information merely to make the dataset look detailed. Do not publish classmates’ individual responses or identify a student through a very small subgroup.
AAPOR’s best-practice guidance includes respondent protection and transparent reporting. In a school setting, an adult should supervise the process rather than ask children to manage privacy risks alone.
The skill being learned is evidence judgement. It is not extracting an answer from every reluctant classmate or constructing a database that the project does not need.
Three questions that reveal real understanding
A club collects two hundred answers. Is it representative? The number alone cannot tell us. We need the population, recruitment method, relevant coverage and response information. A large convenience sample remains a convenience sample.
Eight of ten respondents prefer option A. Does eighty percent of the class prefer it? Only if the respondents are the whole class, or additional design-based or model-based reasoning supports an estimate for the class. First write the direct observation: eight of ten respondents selected A.
One student raises an access problem that most respondents did not mention. Can it be ignored? No. A frequency estimate and a design requirement are different questions. A minority report may identify a real barrier that deserves investigation even when it cannot establish how widespread that barrier is.
These answers show more than the ability to calculate percentages. They show that the learner can keep a claim attached to the kind of evidence it requires.
Count people separately from answers
Suppose twelve students each answer five questions. The project now contains sixty item responses, but it has not heard from sixty different students. The number of recorded answers and the number of participating people perform different jobs.
Likewise, asking the same twenty students again next week produces another set of observations from those students. It may help investigate whether their responses changed. It does not expand the sample to forty distinct people or bring another class into the evidence.
Before calculating a percentage, decide the unit being counted. Is it a student, a completed questionnaire, an answer to one item or a choice selected within an item? A spreadsheet row is not automatically a new person.
Check the collection method for accidental repeat submissions using a school-approved, privacy-conscious process. Do not invent identities for anonymous entries or delete inconvenient answers merely because they resemble one another. Where duplicates cannot be resolved, report the limitation and avoid pretending the participant count is known more precisely than it is.
This distinction protects the meaning of the denominator. More cells in a spreadsheet do not necessarily mean more voices in the project.
Percentages can exceed one hundred when choices overlap
In another fictional question, twenty respondents may select every activity they would consider attending. Fourteen select making, ten select reading and six select drawing. The corresponding percentages are seventy, fifty and thirty. Together they total one hundred and fifty percent.
That total is not automatically an arithmetic mistake. The categories overlap because each person may select more than one activity. The thirty selected options came from twenty respondents, not thirty different people.
Write “percentage of respondents selecting each activity” beside the figures. Do not turn the results into shares of a single exclusive choice unless the question actually required one choice. A pie chart would be unsuitable for presenting these overlapping respondent percentages as pieces of one whole.
Nor can the group add fourteen and ten and conclude that twenty-four different students chose making or reading. Some students may have selected both. The number of distinct students choosing either activity requires information about that overlap.
The question format therefore belongs in the report. Without it, a reader can misinterpret an accurate calculation. Recording whether people could select one option or several is a small administrative detail with a large effect on the conclusion.
A majority selecting an activity is not necessarily a majority ranking it first; preference strength needs a different question.
Let the percentage travel only as far as the evidence
Class polls are useful for hearing people, practising data collection and finding questions worth exploring. They fail when the result is promoted into a claim about people who had no meaningful route into the evidence.
Before submitting the project, read every sentence containing “students”, “everyone”, “most” or “the school”. Ask whether the recorded evidence actually describes that group. A small change in wording can be a large improvement in intellectual honesty.
Continue through the Mathematics Article Directory for fractions and statistical reasoning. Use the graph-comparison guide when the display itself changes the impression, and the source-checking guide when a project relies on published claims rather than its own poll.
The strongest headline is not the biggest claim the group can make. It is the clearest claim the evidence can carry.