Questionnaire design: how to word questions, choose scales and set the order

1. The problem: precise data that answers a different question

A bank asks its customers "How satisfied are you with our services and fees?" and gets a healthy average score. What it cannot tell is whether people are happy with the services, the fees or both, because the question asked for two answers in one. A city hall asks "Do you agree that the central park urgently needs renovating?" and gets 85% agreement, a figure the opposition immediately disputes, and rightly so: the question pointed to its own answer. In both cases the sample may have been flawless. The error happened at the stage of questionnaire design, and it shows up nowhere in the tables and cannot be fixed by weighting or analysis.

The questionnaire is the measuring instrument of any quantitative research. If the instrument is crooked, every interview adds the same error, and a large sample only makes the wrong answer look more certain. This article walks through the stages of building a good questionnaire, with the rules that matter most in practice.

2. Questionnaire design starts with decisions and analysis tables

A good questionnaire does not begin with questions but with the decisions it has to support. The information needs come out of the brief: what we need to know, about whom, and how precisely. For each need, you then sketch the analysis tables you will show in the report, often called dummy tables: for example, "purchase intention by region and age group". Every question in the questionnaire must feed at least one table, and every table must have the questions it needs.

The exercise has two effects. It removes the "interesting" questions that serve no decision and only lengthen the interview. And it exposes the gaps: if a planned table has no source, a question is missing. In questionnaire design, this two-way check takes a few hours and can save a repeated wave of fieldwork. A clear brief makes this stage much easier and cuts several rounds of revision; see how to write a research brief.

3. Question types: open, closed and scales

Any piece of information can be asked for in several forms, and choosing the form is part of questionnaire design.

• Open questions let respondents answer in their own words. They are useful for reasons, suggestions and spontaneous associations, or as the first question on a topic. They cost more to code, and the quality of the answer depends on the interviewer and on how articulate the respondent is.
• Dichotomous questions have two options, such as yes or no. They are simple but force a choice even when the respondent has no position.
• Multiple-choice questions offer a list of answers, with one answer or several allowed. The list must be exhaustive and its options must not overlap, usually with an "other, please specify" option.
• Rating scales measure intensity: agreement, satisfaction, importance, likelihood. They underpin most attitude indicators.

In practice most questionnaires combine types: an open question for the spontaneous reaction, followed by a closed one for comparable measurement. Order matters here too: if the list of options is read out before the open question, spontaneous answers will simply repeat the list.

Closed questions are best built after a qualitative stage, or after a pilot with open questions, which yields the real answer options in the respondents' own language rather than the company's.

4. Question wording: the rules that matter

Question wording is where most errors are made, because the author knows what they meant to ask and can no longer see what they actually wrote. The basic rules are simple and broken constantly.

Define the issue clearly

Who, what, when, where. "How often do you buy dairy products?" leaves open whether the respondent means themselves or the household, which products, and over what period. "In the last 7 days, how many times did you personally buy milk or yoghurt?" leaves no room for interpretation.

Use ordinary, unambiguous words

Company jargon ("digital channels", "portfolio", "touchpoint") is not the respondent's vocabulary. Vague frequency words such as "often", "sometimes" or "regularly" mean different things to different people; where possible, ask for actual frequencies or defined ranges.

One idea per question

The double-barrelled question, like the bank example above, is the most common wording error. It is easy to spot by the "and" or "or" that joins two things to be rated.

Do not lead the respondent

Questions that contain an opinion ("Do you agree that..."), invoke an authority ("Doctors recommend... Do you...?") or present only one side of the alternative shift the answers. The correct version presents both directions: "Some people say..., others say.... Which view is closer to your own?"

Avoid hidden assumptions and needless effort

"What do you like most about our app?" assumes the respondent uses the app and likes something about it. Filter questions come first. And questions that require arithmetic ("what share of your income do you spend on...") produce rough guesses; it is better to ask for the components and let the analyst do the calculation.

Allow for memory and social desirability

The reference period should match how often the behaviour occurs: a week for everyday shopping, a year for rare purchases. On sensitive topics such as income, alcohol use or voting, respondents tend to give the socially acceptable answer. Neutral wording helps, as do ranges instead of exact amounts, self-completion on a tablet or online, and an explicit assurance of confidentiality.

5. Response scales: decisions, not habits

Different response scales for the same concept produce results that cannot be compared. Four decisions need to be made deliberately.

• Number of points. Five or seven points are enough for most purposes; on the telephone, where the respondent cannot see the scale, short scales with every point labelled in words work better.
• Balance. A balanced scale has as many positive options as negative ones. A scale running "excellent, very good, good, fair, poor" has four favourable points and one unfavourable, and inflates the results.
• Neutral midpoint. An odd number of points allows a middle position. If many respondents are expected to be genuinely neutral, keep the midpoint; if not, an even scale forces a direction, at the risk of creating opinions where none exist.
• The "don't know" option. It reduces made-up answers on topics the respondent knows nothing about, but gives an easy exit to those who do not want to think. Include it where lack of knowledge is plausible.

For indicators that will be tracked over time, the scale chosen in the first wave becomes binding for every later wave. One standardised scale, with its strengths and limits, is discussed in our article on customer satisfaction and NPS.

6. Question order and the structure of the questionnaire

The answer to a question is influenced by the questions that came before it. A well-built questionnaire follows a predictable structure:

• The introduction: who is conducting the study, its general purpose, how long it takes, confidentiality and the voluntary nature of taking part.
• Screening questions, which establish whether the person belongs to the target population.
• Simple, neutral opening questions that build trust.
• The body of the questionnaire, from general to specific: unaided awareness before aided, the overall rating before individual attributes. If you ask about prices first and overall satisfaction afterwards, the overall score will be coloured by prices.
• Sensitive and difficult questions towards the end, once rapport with the interviewer has formed.
• Demographics at the end, except those needed for quotas or screening.

Long lists of options and attributes are rotated or randomised to control position bias: options at the start or the end of a list are chosen more often. Skip logic ("if no, go to question 15") is programmed into the interviewing software, in CAPI and CATI or online, and every possible route is checked before fieldwork.

7. The collection method changes the questionnaire

The same questionnaire does not behave the same way face to face, by telephone and online. On the phone, the respondent cannot see the options: lists get shorter, scales get simpler and interviews rarely run beyond 15-20 minutes. Online, the respondent reads alone: instructions must be complete, large question grids are tiring, and on a mobile phone wide columns become unreadable. Face to face, you can use show cards and longer interviews, but the interviewer effect grows on sensitive questions. More on each method on our CAWI and online panel page and in our articles on face-to-face and telephone interviewing.

In Moldova, most national studies run in both Romanian and Russian. Translation is not a technical step: a phrase that is neutral in one language can be loaded in the other, and a scale can carry different shades of meaning. Good practice is translation followed by independent review and a separate test of each version. Doing the questionnaire design in both languages in parallel, by researchers fluent in both, reduces the risk that the translated version is only an approximation of the original.

8. Question types compared

9. What to check in the questionnaire the agency proposes

Before you approve it, read the questionnaire out loud as a respondent would, then go through the list below. The next step, and a mandatory one, is testing it on real people, described in our article on questionnaire pretesting and cognitive interviews.

• Every question maps to an objective and to a planned analysis table.
• No question asks for two ratings at once.
• No question suggests the answer or assumes a behaviour that has not been screened.
• Reference periods are defined and suit how often the behaviour occurs.
• Scales are balanced and consistent throughout.
• Option lists are complete, do not overlap, and are rotated.
• The order runs from general to specific, with sensitive questions towards the end.
• The estimated length suits the collection method.
• The Romanian and Russian versions have been reviewed independently.
• A pretest is planned before fieldwork.

Good questionnaire design takes time, and that time is repaid several times over in data quality. Question wording and the choice of response scales decide what the final report will be able to say, long before the first interview. If you are preparing a study and want your instrument reviewed by researchers with field experience, our quantitative research team can help.

See also: Quantitative research CAPI and CATI CAWI and online panel Questionnaire pretesting and cognitive interviews How to write a research brief Customer satisfaction and NPS

The Perspicuus research team · 1 February 2026