Four- vs Five-Point Class Feedback Surveys

Did you know that two classes can both report an average satisfaction score near the top of a survey, yet those numbers may not be directly comparable? Imagine one class answered a question with four options and another used five. One questionnaire forced a choice between agreement and disagreement; the other allowed a neutral answer. A comparison that ignores this difference can turn a simple school feedback exercise into a misleading conclusion.

A four-point agreement scale might use 1 = strongly disagree, 2 = disagree, 3 = agree and 4 = strongly agree. A five-point agreement scale can add a middle option: 1 = strongly disagree, 2 = disagree, 3 = neither agree nor disagree, 4 = agree, 5 = strongly agree. The immediate answer is that raw means such as 3.0 out of 4 and 4.0 out of 5 are not interchangeable. Even transforming them to a common numerical range cannot prove that the respondents interpreted the options or questions in the same way.

Start with a learner diagnostic instead of a formula

Give a student two invented school feedback results: Class A averages 3.0 on a four-point scale; Class B averages 4.0 on a five-point scale. Ask which class was more satisfied. A pupil may answer B because four is larger than three. Another may compare 3 ÷ 4 with 4 ÷ 5 and confidently report 75% versus 80%. A third may correctly hesitate but be unable to say why.

The most useful diagnosis is a question: “What exactly did the number 3 mean in each questionnaire?” On the four-point version, 3 meant agree. On the five-point version, 3 meant neither agree nor disagree. The same digit occupies a different labelled response category. Once the learner sees that, the issue becomes one of evidence and meaning, not merely arithmetic.

The purpose of a feedback survey is to learn something about people’s reported views within a particular question and response system. The purpose is not to manufacture a scoreboard. Good comparisons keep the exact question wording, choices, sampling method and reporting basis visible so that a reader can judge what the figures actually support.

Worked example 1: two class averages and the trap in the middle

Every number in this example is fictional and illustrative. The respondents are not real pupils, and the results are not evidence about any actual Singapore school. Assume four independent completed responses in each hypothetical group. Each questionnaire asks “The explanation helped me understand today’s lesson”, but the response options differ.

Class A, four-point scale: the four coded answers are 2, 3, 3, 4. Their sum is 12; the arithmetic mean is 12 ÷ 4 = 3.0, on a 1–4 scale.
Class B, five-point scale: the four coded answers are 3, 4, 4, 5. Their sum is 16; the arithmetic mean is 16 ÷ 4 = 4.0, on a 1–5 scale.
Read the labels: A’s two scores of 3 mean agree; B’s score of 3 means neither agree nor disagree.
Count favourable replies within each instrument: A has three replies in its two agreement categories, 3 ÷ 4 = 75%. B also has three in its agreement categories, 3 ÷ 4 = 75%.
Safe description: in these particular invented responses, three of four respondents in each group selected an explicitly favourable category. That does not establish equal satisfaction across the groups or prove that the scales measure attitudes identically.

Counting labelled agreement responses is more transparent here than simply ranking the coded averages. Yet the same 75% is not a universal bridge between scales. A learner who was unsure on a forced-choice four-point form may have selected “agree” or “disagree”; on the five-point form they might have selected the neutral option. The availability of neutral can change the response distribution. The respondent groups, wording, survey administration and possible nonresponse also matter.

There is another important distinction. A Likert-type response is an ordered category, not automatically a measured physical quantity with equal-sized intervals between labels. Taking an arithmetic mean of category codes is a commonly used descriptive convenience; it depends on assumptions about what those codes represent. Showing the full distribution and the labelled options prevents the mean from hiding what the students actually selected.

Why dividing by the maximum creates another problem

Students sometimes attempt to repair the mismatch by treating each class average as a percentage of the maximum score. For Class A, 3 ÷ 4 = 0.75 or 75%. For Class B, 4 ÷ 5 = 0.80 or 80%. The arithmetic is correct, but the interpretation is easily wrong. Both response scales start at 1, not 0. Moreover, a number such as “80% of possible satisfaction” is not a defensible measure of how many learners agreed.

If the sole goal is to express the position of a coded score between its numerical endpoints, a transparent linear transformation is (score − minimum) ÷ (maximum − minimum). For Class A’s mean, (3 − 1) ÷ (4 − 1) = 2 ÷ 3 = 0.6667, about 66.7% of the coded numerical span. For Class B, (4 − 1) ÷ (5 − 1) = 3 ÷ 4 = 0.75, or 75% of its numerical span.

This endpoint transformation is mathematically well defined, but it does not make the two questionnaires psychometrically equivalent. It does not insert a neutral response where none existed; it does not make agreement categories equally spaced; it does not prove that a score in one survey expresses the same attitude as a transformed score in another. Never label 66.7% or 75% from this operation as “the percentage of students satisfied”. The earlier three of four explicitly agreeing calculation answers that different question.

Worked example 2: the category that disappears

Suppose a different fictional five-point classroom survey receives these five coded responses: 3, 3, 4, 5, 2. Under the labels stated above, two pupils selected neutral, two selected an agreement option, and one selected disagreement. That is 2/5 = 40% neutral, 2/5 = 40% explicitly favourable, and 1/5 = 20% explicitly unfavourable. The categories together account for all five responses: 40% + 40% + 20% = 100%.

Now imagine we replace the questionnaire with the four-point version. We cannot conclude that its two previously neutral pupils will divide evenly between agreement and disagreement. They could choose different sides, skip the item, or respond differently because of the changed context. Without a properly designed comparison study, the four-point outcomes are unknown. The honest response to a missing neutral category is not to split its 40% in half and announce what would have happened.

This example also distinguishes no neutral choice from no neutral opinions. Removing an option from a form cannot establish that a person stopped feeling unsure. Survey architecture affects which views become visible in the data.

What original survey research tells us to check

The Pew Research Center’s guidance on writing survey questions explains that question wording, the order of choices and response options can affect answers, and that maintaining wording matters when tracking changes over time. That is relevant to school questionnaires, even though a short classroom poll does not automatically meet the standards or purposes of a professional population survey.

Pew’s analysis of a “don’t know” response option also illustrates why adding or omitting a way to express uncertainty can change the distribution of responses. “Don’t know” is not identical in meaning to “neither agree nor disagree”, so the example should not be imported as though the labels were equivalent. Its transferable lesson is to inspect which kinds of answers the instrument permits.

A school survey should therefore carry a small method note: the question text as shown, all response labels, the count of people asked, the count who answered, any missing responses, the date and method of collection. Comparing charts with those details hidden makes an apparent improvement or decline difficult to evaluate.

How to compare responsibly when the scales differ

First identify the shared question: are both surveys asking about the same lesson, experience and time period? If wording or context differs, state it.
Show the actual labels: do not write “score 3” without specifying whether it means neutral, agreement or something else.
Report response counts: give numerators and denominators, not just a rounded percentage. For example, “3 of 4 selected agree or strongly agree.”
Separate missing from neutral: a blank response and an explicit neutral response are not the same observation.
Present the two distributions side by side: make scale differences visible rather than silently renaming columns.
Qualify the interpretation: a category-count comparison can be descriptive, but it is not proof that attitudes are equivalent or that a change in teaching caused a change in answers.

When longitudinal comparison is important, the strongest practical approach is often to keep a stable, well-designed question and response set across survey rounds. If a change is necessary, document it clearly. More advanced methods can investigate comparability, but a quick spreadsheet rescaling is not such a validation study.

Guided practice before independent judgement

Level 1: label the categories correctly

Give a pupil two printed response lists, one four-point and one five-point, and ask them to circle every option that explicitly says “agree”. The immediate goal is not a mean; it is accurate reading. On our stated four-point form, agreement categories are 3 and 4. On our stated five-point form, they are 4 and 5. The pupil must explain why “3” is not portable across the two forms.

Level 2: calculate and check a new set

Use an invented four-point list 1, 2, 3, 4, 4. There are three explicitly favourable responses of five, so 3 ÷ 5 = 60%. The coded mean is (1 + 2 + 3 + 4 + 4) ÷ 5 = 14 ÷ 5 = 2.8, but that mean is not 60% and should not be labelled as such. Ask the pupil to verify the favourable count by reading the word labels rather than relying only on the digits.

Level 3: identify an unanswerable claim

Another class changes from a five-point survey in one term to a four-point version in the next. Its reported average rises from 3.6 to 3.7. Can we confidently say satisfaction increased? No. The numbers come from instruments with different response structures. We would need more information about wording, participants, response distributions, measurement comparability and sampling before drawing such a conclusion.

Level 4: explain uncertainty to a real audience

Ask the learner to draft a two-sentence note for a parent newsletter: “The two surveys offered different response choices, so their averages should not be read as a direct trend. We have reported the separate distributions and will use a consistent question next time to support future comparisons.” That is a useful communication outcome, not simply a calculation.

Common mistakes and why they mislead

“Four is larger than three, so Class B is happier.” The numbers use different scales and may label different attitudes.
“Divide both means by their maximum and the problem disappears.” This ignores the nonzero minimum and the meanings of the categories.
“A common 0–100 transformation proves fairness.” Numeric endpoint alignment does not establish the same measurement meaning.
“Neutral means the student was absent.” Neutral is an explicit answer; no response is missing data.
“A four-point scale has no uncertainty.” It has no explicit neutral choice, which is not the same as removing uncertainty from respondents.
“Two equally sized favourable percentages prove equivalent teaching.” They are descriptive proportions, not controlled evidence of the cause of attitudes.
“More decimal places make the comparison precise.” Extra decimals cannot repair a changed questionnaire or a small, unrepresentative sample.

What students, teachers and parents can do next

Students can practise describing a result before calculating. Write “three of four respondents chose an agreement category” and only then compute 75%. If the labels differ, give them in a caption. This builds the habit of treating words and numbers as one evidence system rather than two separate school subjects.

Teachers can keep copies of each questionnaire and build a one-page reporting template containing question, labels, respondent count, missing-answer count and time period. If a form changes, annotate the chart instead of silently joining the old and new numbers. For formative feedback, a qualitative comment or follow-up conversation may reveal problems that a coded average hides.

Parents reading a school project can ask a constructive question: “Were the same choices offered to both groups?” Encourage the child to show the form, explain how a neutral reply was handled and say what the graph cannot reveal. A correct limitation is a sign of careful reasoning, not a weakness in the project.

A useful progress check is to give a learner four small survey summaries. Record how many times they identify the response scale, choose an appropriate numerator and denominator, spot a missing neutral option and avoid an unsupported trend claim. Re-test with unfamiliar wording a week later; improvement means the method transfers beyond the original example.

Limitations of a short classroom survey

Small self-selected groups do not reliably represent a whole school. The teaching style, timing of questions, anonymity, who responded, item order and meaning of agreement can all affect answers. A change between two response sets might reflect real learning experiences, differences in who answered, response-option effects or a combination. The survey alone may not distinguish these explanations.

For especially sensitive feedback, protect student privacy and follow school policies. Do not publish identifiable answers or imply that invented classroom examples are real outcomes. A survey question also cannot replace observing student work, checking understanding and talking to learners. The purpose is useful evidence, not an unearned guarantee of improvement.

Frequently asked questions

Is a five-point survey automatically better than a four-point one?

No. The choice depends on what you need to learn, the audience and the interpretation plan. Providing a neutral category may capture uncertainty, but the questionnaire must still use clear, relevant wording and a sensible response structure.

Can we compare the percentage who selected an agreement option?

You can describe each instrument’s explicitly favourable responses with the correct category definitions and denominator. However, the resulting proportions are not automatically equivalent across differently designed questions or groups, and a difference is not necessarily a change in underlying satisfaction.

What if a respondent skipped the question?

Report missing responses separately. If five people were invited and four answered the item, the denominator for a completed-response category proportion is normally four, provided you state that basis. Do not quietly count the blank as neutral.

Should I avoid all averages for survey scales?

Not necessarily. Averages can be a useful descriptive summary with appropriate assumptions and labels. They should not conceal the original response distribution or be interpreted as interval-level physical measurements without justification.

Can I convert the old scale to the new one in a spreadsheet?

You may mathematically map coded endpoints, but the mapping does not recreate how people would actually have answered if they had seen different options. When a decision depends on comparability, document the limitation and consider collecting consistently designed data.

The core aim is evidence you can explain

A useful school survey is one whose meaning survives a careful reader’s questions. If pupils learn to ask what each choice means, who answered, which denominator was used and what changed between questionnaires, they are practising a skill larger than percentages. They are learning to keep conclusions proportional to evidence—and to improve the next question rather than merely decorate the last chart.

The Core Aim of Education | How to Compare Class Feedback When Surveys Use Four-Point and Five-Point Scales