Did you know two columns in the same research table can both be correct while using a different number of people? A class project might say 75% of pupils enjoyed a reading activity and 80% found its resource helpful. It is tempting to conclude that more pupils liked the resource. But what if the first question was answered by 24 pupils and the second by only 20? The denominator changes what those percentages can honestly tell us.
The immediate answer: the symbol n in a research table usually indicates a count of observations or participants used for the particular statistic, but its exact meaning must be checked in the table legend and methods. Different columns can have different n values because some questions were skipped, records were missing, a question applied only to a subgroup, or an analysis used different eligibility rules. Always pair a percentage, mean or other statistic with the actual denominator that produced it. Do not quietly apply the whole-study sample size to every cell, or interpret different respondent groups as though they were identical.
- Read the title, column labels, footnotes, study population and the definition of n.
- Record how many pupils were invited and how many supplied valid answers to each question.
- Calculate each percentage using its own number of valid responses.
- Separate ‘percentage among respondents’ from ‘percentage of everyone invited’.
- Look for missing, not applicable, excluded and skipped values; they do not all mean ‘no’.
- Do not subtract two percentages as though they measured the same pupils without checking comparability.
- Write conclusions with both the numerator and denominator, then identify what is not known.
Why a table is more than a collection of numbers
A research table compresses a story of data collection into a small area. It may contain a study population, a narrower eligible population, a set of responses to each question and an even narrower set of complete records for an analysis. Columns show summaries of those records, not necessarily summaries of precisely the same individuals.
Suppose 30 pupils are invited to give feedback. All 30 received the survey, 27 opened it, 24 answered Question A and 20 answered Question B. ‘30 invited’, ‘27 opened’, ‘24 with A answers’ and ‘20 with B answers’ are four different counts. Each can be valid in its proper place. A careful reader must establish which count belongs to which claim.
This is a general principle of transparent reporting, not an obscure statistical trick. The STROBE initiative, which supports reporting of observational studies, provides checklists on explaining participant numbers and data availability. Its reproduced checklist explicitly asks researchers to indicate how many participant records are missing for each variable of interest. Our classroom survey below is an invented teaching dataset, not a reported research study.
Worked example: thirty invitations, two different denominators
Here is the complete fictional dataset summary. Thirty pupils were invited to answer two yes/no questions. For Question A, 24 provided valid answers: 18 yes and 6 no; the other 6 did not provide a valid answer. For Question B, 20 provided valid answers: 16 yes and 4 no; the other 10 did not provide a valid answer. These counts have been chosen for an illustrative calculation and do not describe real pupils.
Question A: enjoyed the reading activity
The percentage saying yes among the 24 valid Question A respondents is 18 ÷ 24 × 100% = 75%. Check the arithmetic: 24 × 0.75 = 18. The number 24 is the denominator; 30 would answer a different question.
Question B: found the resource helpful
The percentage saying yes among the 20 valid Question B respondents is 16 ÷ 20 × 100% = 80%. Check: 20 × 0.80 = 16. Therefore 80% is a truthful summary of valid Question B respondents, but it is not automatically 80% of all 30 invited pupils.
Can we say more pupils liked Question B?
Not from the two percentages. The recorded number of yes answers is 18 for A and 16 for B, although the percentage among respondents is higher for B. It would be false to write ‘more pupils selected yes for B’ because 16 is smaller than 18. It would also be misleading to treat B’s 80% and A’s 75% as a controlled comparison for precisely the same pupils: we have not established which pupils answered both.
If a report simply states ‘A: 75%; B: 80%’ without the n values, it hides an important difference. A clearer report is: ‘18 of 24 valid respondents (75%) answered yes to A; 16 of 20 (80%) answered yes to B.’ Such reporting makes the evidence inspectable.
A second denominator: what about everyone who was invited?
Among all 30 invited pupils, the confirmed yes count for A is 18 out of 30, or 60% of invitees; for B, 16 out of 30, or approximately 53.3% of invitees. These figures mean ‘at least these many were recorded as yes’. They do not mean the non-respondents were no; their opinions are unknown.
What are the mathematical extremes if every missing answer were actually yes? A would have 18 + 6 = 24 yes answers, or 80% of invitees. B would have 16 + 10 = 26 yes answers, about 86.7% of invitees. Thus the data alone allow a range of possible full-class yes proportions: A from 60% to 80% and B from about 53.3% to 86.7%. These are simple worst-case logical bounds, not confidence intervals or estimates that missing pupils hold a particular opinion.
This exercise is useful because it exposes how much depends on missing responses. The observed percentages of 75% and 80% are not automatically unbiased estimates of the entire class. The size and reason for nonresponse matter.
Where different n values come from
- An item is blank: the respondent leaves a question unanswered.
- The form includes skip logic: only a defined subgroup was asked the follow-up question.
- A response is invalid or outside an allowed range and is excluded according to the stated rules.
- A pupil joins the study after the first activity or leaves before the final measure.
- A laboratory measurement fails for one sample, even though other measurements remain valid.
- A statistic needs several fields at once; a record missing one required field may be excluded from that analysis.
- The author deliberately reports a smaller subgroup; the analysis denominator then depends on group eligibility.
These explanations are not interchangeable. Missing information can arise for different reasons, and a particular table may not tell you which one occurred until you read the methods or footnotes. Never fill an empty cell with a zero or a ‘no’ by default.
A diagnostic question: what does n count here?
Show a learner the statement ‘Mean weekly reading hours = 3.5, n = 20’. Ask what the 20 refers to. It is not the mean of 20 hours; it indicates 20 records contributing to that statistic if the table notes define n that way. Next ask whether 3.5 hours describes every pupil. No: it is a group summary. Finally ask whether we know the total reading hours. Under an ordinary arithmetic mean of 20 numeric responses, yes: 3.5 × 20 = 70 pupil-hours summed across those 20 respondents. We cannot claim anything about ten other invited pupils from that calculation.
Another diagnostic is to show 16/20 = 80% and ask, ‘Does that prove 24 of the 30 invited pupils said yes?’ It does not. The numerator is 16, not 24; multiplying the observed response rate by the larger invitation count invents answers that were not provided.
A method to read any such table in seven checks
- Check 1 — name the population: all invited, eligible, enrolled or sampled individuals.
- Check 2 — locate the table’s n notation and read every footnote.
- Check 3 — separate counts used for each variable from the overall study total.
- Check 4 — calculate one percentage back from its numerator and denominator to detect a mismatch.
- Check 5 — ask why a column lost observations: missing data, subgroup, filter or another rule.
- Check 6 — compare like with like, preferably paired records or clearly defined groups when making direct claims.
- Check 7 — restate the result in ordinary language with uncertainty and no invented responses.
Progressive practice: from guided counting to independent judgement
Practice 1: one clear denominator
A toy table says 12 yes, 3 no and 5 unanswered out of 20 invited. There are 15 valid answers, so the yes share of valid answers is 12 ÷ 15 = 80%. The confirmed yes share of all invited is 12 ÷ 20 = 60%. A learner should be able to say why both percentages are correct but answer different questions.
Practice 2: two questions
Now the same invented class receives another item with 9 yes, 1 no and 10 unanswered. Among the 10 valid replies, 9 ÷ 10 = 90%. The recorded yes count is lower than the first question’s 12. Ask learners to write two sentences: one comparing percentages among valid respondents, and another declining to conclude that more of the whole class liked the second item.
Practice 3: independent table repair
Show a student a report that says ‘Reading enjoyment = 80%, resource helpfulness = 90%, n = 20’ and ask them to rewrite it using the counts from Practices 1 and 2. A defensible answer names each denominator separately: ‘12/15 (80%) for enjoyment; 9/10 (90%) for helpfulness; 20 pupils invited.’ The original use of one n was ambiguous.
Practice 4: cautious research language
Ask learners to critique: ‘The new resource improved pupil enjoyment by ten percent.’ The statement is unsupported. We have two questions with different denominators, not matched before-and-after results or an experiment showing a causal effect. The apparent difference is ten percentage points among respondents to different questions. A causal improvement claim would need appropriate study design and evidence.
Why even a beautifully formatted table can mislead
Clear fonts, coloured bars and exact decimals can make a weak conclusion look settled. The real test is whether the numbers mean what the sentence claims. If a reported survey percentage changes because one denominator excludes absent pupils, the arithmetic may be flawless and the interpretation still wrong. A reliable reader treats denominators as part of the meaning, not housekeeping to hide in small print.
A related trap is taking an average of two percentages without accounting for their denominators. Here (75% + 80%)/2 = 77.5% is merely the unweighted average of those two question percentages. It is not the proportion of unique pupils who said yes to either question, because the same pupil may have answered both. Even (18 + 16)/(24 + 20) = 34/44 ≈ 77.3% pools question-responses, not unique individuals; it only answers a specifically defined item-response question and should not be called ‘percentage of pupils satisfied’.
What parents and teachers can observe
For a classroom investigation, teachers can keep an anonymous count of invitations and valid responses to each item, without displaying identifying pupil records. Parents can ask children, ‘Who exactly is in that percentage?’ instead of only asking whether they used a calculator. Learners can maintain a short evidence ledger that records the question, numerator, denominator, missing count and a one-sentence limitation. These are observable habits and are easier to improve than a vague promise to ‘be careful’.
One sign of progress is when a learner spontaneously writes ‘18 out of 24 valid replies’ instead of ‘75% of everyone’. Another is when they notice that a table cannot support a before-and-after causal claim. A third is the ability to solve a new example without prompts and explain the difference between an unanswered and negative response.
Limitations: not every study is a class survey
Large studies may use weighting, nonresponse adjustments, imputation, complex sampling or preplanned subgroup analyses. Some statistics use a different definition of sample size, such as numbers of observations rather than distinct people. A summary table alone cannot tell you whether missing data are harmless or biased. Readers of a professional study should inspect the study design, denominators, methods, uncertainty and limitations before interpreting results. This lesson teaches the basic logic and does not replace a full statistical analysis.
Frequently asked questions
Does n always mean the number of pupils?
No. It commonly represents the number of relevant observations or participants, but sometimes counts measurements, events or other units. Follow the author’s definition.
If a question was skipped by design, should I mark it no?
No. ‘Not asked’ or ‘not applicable’ is different from an observed no answer. Follow the stated questionnaire rules.
Can I compare 75% with 80% at all?
You can state that the percentages among the respective valid respondents differ by five percentage points. But a claim about the same people, the whole class or a causal effect needs additional evidence.
Is a larger n always better?
More valid observations may improve precision, but quality, selection, study design and measurement matter. A large biased sample does not automatically solve bias.
Should every table cell show n?
Readers need an unambiguous way to identify the denominator of each statistic. A clearly labelled shared n may be enough when it genuinely applies; varying n values should be made transparent.
Continue learning with verified guides
Explore multiple-answer survey percentages and how to interpret before-and-after school studies. The sampling and survey-bias lesson provides subject practice. For the larger learning framework, see the Education Hub, Core Aim of Education hub, and our immutable Clementi Mathematics reference on explanation, guided practice and independent checking.
The most useful habit
Before believing any percentage, ask a wonderfully ordinary question: ‘Out of how many?’ Then ask why that number was chosen. The aim of education is not to decorate an answer with more digits; it is to make the evidence, reasoning and uncertainty clear enough that another person can check them.