Did you know? A class of 24 students can submit 26 online survey forms without producing 26 independent opinions. Two extra entries might be duplicate submissions, but they might also be outsiders, a testing response or an administrative error. The size of the spreadsheet alone cannot tell you what happened. The difference matters whenever a class uses survey numbers to decide what to change.
A duplicate survey response is an additional submission from the same intended respondent to a survey designed to count each respondent once. To identify duplicates responsibly, first establish how the survey defines a respondent and whether it has a reliable, privacy-conscious identifier. Then document a consistent rule for handling confirmed repeats. Matching answers alone do not establish that two forms came from the same person.
The core aim is to count evidence correctly while protecting people from unsupported accusations. This is a fictional classroom exercise, not a report on any real school, student, platform or data breach. No actual student information is required to practise the method.
The first diagnostic: submissions are not necessarily people
Present a learner with three statements: “24 students were invited”; “26 forms were received”; and “17 forms selected the bus option.” Ask them what each number describes. The first counts invitations, the second counts forms, and the third counts forms with one answer. None independently proves how many distinct people answered. An invitation count does not even prove participation: some students might not respond, and some might submit twice.
Ask the student to finish this sentence before dividing: “The denominator should count _____.” If the research question is “What proportion of participating students prefer the bus?”, the blank is distinct eligible respondents with a valid answer to that question, not simply the number of form records. If the survey deliberately allows repeated entries over days, by contrast, the unit may be a daily response rather than a person. The survey design determines the denominator.
A worked example with verified one-time classroom codes
Imagine a made-up voluntary classroom poll about the preferred method of travelling to a hypothetical excursion. A teacher issues 24 one-time random response codes, one for each eligible participant. Assume for this exercise that every code is issued once, codes are not tied to names in the analysis copy, all 24 students submit at least once, and the intended analysis counts one response per student.
- Total submitted forms: 26.
- Distinct valid issued codes in the export: 24.
- Two different codes each appear twice; all other codes appear once. Thus there are two extra forms beyond one submission per code.
- Of the 26 submitted forms, 17 select “bus.” Both repeated entries also select “bus” and match the first entries for those codes.
- After retaining the pre-specified first valid submission for each code, the data contain 24 distinct responses, 15 selecting bus.
The naive form-based percentage is 17 ÷ 26 × 100 = 65.38% (to two decimal places). The respondent-based percentage is 15 ÷ 24 × 100 = 62.5%. Both calculations are arithmetically correct; only the second answers the stated one-response-per-student question under the exercise assumptions. Check it backwards: 62.5% of 24 is 15. The difference is about 2.88 percentage points, not 2.88% of students.
The calculation does not tell us whether the same child deliberately submitted again. A browser refresh, accidental resubmission or platform behaviour could explain the repeats. The codes establish duplicate records under the setup, not motive. If the two entries for one code disagree, the analyst must follow a rule agreed before inspecting preferences—for example, first completed valid response or a verified correction procedure—and explain that choice. Quietly keeping whichever answer supports the desired result would be unfair.
What identical answers do—and do not—prove
Suppose two form rows both say “bus” and both take 42 seconds to complete. They could still be two different people. Even identical full answer patterns cannot prove identity: classmates may share preferences and read questions at similar speeds. Conversely, the same person might submit two very different answers after changing their mind. Without a permitted, reliable respondent identifier, identity may be unknown.
Avoid using names, email addresses, IP addresses or device fingerprints without proper permission and a legitimate educational need. If a school has a required privacy and retention policy, follow it. A random code reduces how much personal information appears in an analysis sheet, but it is not automatically fully anonymous if someone keeps a code-to-student lookup. Prefer aggregate reporting and minimal data.
How to prevent repeats instead of trying to guess afterwards
- State the survey rule plainly: one response per participant, unless the exercise is a repeated diary study.
- Use access controls or single-use codes where the school’s approved survey platform and privacy policy permit them.
- Test the form with a few dummy entries before opening it to students; remove test submissions from the final analysis.
- Define what counts as a valid complete submission and how approved corrections will be handled.
- Keep an unaltered original export and a separate cleaned analysis copy with a short record of every exclusion.
- Report the number invited, the number of submitted forms, the number of distinct eligible responses and the number of unresolved records separately.
This is especially important with optional surveys. Issuing 24 codes does not justify writing “24 students responded” if only 19 codes came back. Likewise, a survey with no identifiers may not permit reliable deduplication at all. The correct response to that limitation is to report uncertainty or redesign the next survey—not to delete apparently repetitive answers until the total looks plausible.
Guided practice: the denominator changes when the question changes
Try a second fictional poll. Eighteen valid codes were issued; 15 distinct codes returned responses; there are 16 submitted forms because one code appears twice. Of the 16 forms, 10 select the library as a preferred study location, and the confirmed extra form also selects the library. The retained first responses therefore include nine library choices among 15 participating students.
If the question is “What percentage of valid participating students selected the library?”, the answer is 9 ÷ 15 × 100 = 60%. If the question is “What percentage of forms selected the library?”, the answer is 10 ÷ 16 × 100 = 62.5%. If the question is “What percentage of all 18 invited students returned a valid response?”, it is 15 ÷ 18 × 100 = 83.33%. Three different denominators answer three different questions. The learner should be able to label each before calculating.
Independent challenge: what if no identifiers exist?
A different fictional survey records 31 forms but has no issued codes, response identity or verified participation roster. Six rows contain the same combination of selections. Can we subtract five rows? No. Identical answer combinations are not evidence that those forms belong to one person. Even with 30 students on a class list, the number of unique respondents cannot be reconstructed just from the number of forms. Students should write a careful limitation such as: “We received 31 forms but cannot verify that each came from a different eligible respondent.”
A useful extension is to propose a new survey design that can count one response per participant while keeping names out of the report. Have students explain what the new design can prove, what it still cannot prove and whether the identifiers would remain linkable to individuals.
Common misconceptions to repair
- “More forms than pupils means everyone answered and some answered twice.” Not necessarily: outsiders, missing responders and trial entries remain possible.
- “Identical answers prove a duplicate.” They do not establish respondent identity.
- “Removing repeats always leaves the true opinion of the group.” That depends on code validity, response rules, missing answers and survey coverage.
- “A spreadsheet’s COUNT function counts people.” It counts records or nonblank values unless the dataset is set up to establish unique respondents.
- “One decimal answer is enough.” A percentage needs a stated numerator, denominator and description of whom it represents.
A simple parent-and-teacher practice routine
A parent can draw eight numbered pretend invitation cards and ask a child to place ten pretend response slips into a box, with two matching codes. Let the child distinguish slips from participants, calculate the share who chose an option and explain whether any unmatched cards matter. Keep personal information out of the exercise. A teacher can make two nearly identical sheets—raw forms and one-response-per-code—and ask groups to explain why their percentages differ.
Progress is observable when learners can name the analysis unit, state the counting rule before looking at preferences, retain raw data, justify each exclusion and express uncertainty where identity cannot be established. An error log can record whether a mistake came from choosing a form denominator, miscounting distinct codes or treating an unverified guess as a fact.
A necessary limitation: a class poll is not automatically representative
Even a perfectly deduplicated classroom poll may not represent the whole school. Nonrespondents can differ from respondents, and a voluntary opinion question can be affected by wording or context. For sound survey design, see the American Association for Public Opinion Research’s survey best practices, which address objectives, population, sampling, nonresponse and responsible reporting. Its guidance is a general research resource, not a claim that a school poll meets professional survey standards.
Frequently asked questions
Can I use a spreadsheet’s Remove Duplicates button?
Only after establishing what uniquely identifies a response and what the retention rule is. Removing rows because all visible answers match may erase valid opinions. Preserve the original data first.
What if a pupil needs to correct an earlier answer?
Create an explicit correction process, preferably through the approved survey system. Decide whether a corrected response replaces the original; do not silently keep whichever version supports a preferred outcome.
Should the teacher publish the list of repeated codes?
Usually the educational report needs counts and methods, not individual identifiers. Follow school policy and use a privacy-preserving aggregate.
Does 62.5% mean 62.5% of the whole class?
Only if all 24 eligible pupils in the invented example submitted a valid response and the question referred to that group. In a voluntary survey with nonresponse, use “of valid respondents” instead.
Keep learning: better questions, better evidence
The related Core Aim lesson on multiple-response survey percentages explains why valid category percentages may total more than 100%. The lesson on non-overlapping survey answer choices tackles a different source of error. Together, they show that careful questions, verified response counts and honest interpretation matter more than a polished chart.
For diagnosis-first learning and gradual independent checking, visit the eduKate Learning System and our Clementi Secondary 1 Mathematics learning reference. A reliable result begins when a learner knows exactly what was counted and can defend that count without guessing about another person.