A questionnaire looks simple: a list of questions, sent to people, answered and analysed. In practice, a badly designed questionnaire produces data that cannot answer the research question, however many responses arrive. Ambiguous wording, leading questions, and missing answer options quietly corrupt results long before analysis begins.
This guide explains how to design a research questionnaire that collects reliable, analysable answers. It covers defining what to measure, choosing question types, writing neutral questions, ordering and layout, scales, pilot testing, and honest reporting. It builds on our guides to writing a methodology section and organising and storing research data, and it pairs with our guide to choosing the right statistical test.
Every question in a questionnaire must trace back to a research objective. Before writing any question, list the concepts the study must measure: knowledge, attitudes, practices, or behaviours, for example. Then list the indicators for each concept: the observable things that show it. A question that cannot be linked to an indicator does not belong in the questionnaire, however interesting it seems.
This discipline also controls length. Long questionnaires suffer from fatigue: respondents rush, skip questions, or abandon the survey halfway, and the later questions end up with the worst data. A questionnaire that takes more than about fifteen minutes needs a strong justification, and most research questionnaires work better when they are shorter than the author first planned.
Write the analysis plan before the questionnaire, not after. Decide which comparisons and tests each question will feed, so that the answer format matches the analysis. A question designed for counting needs fixed options; a question designed for exploration needs open text. Mixing these up at the design stage creates data that cannot be analysed as intended.
Also decide who counts as a valid respondent before writing questions. If the study targets practising veterinarians, for example, the questionnaire needs screening questions that confirm eligibility, so that answers from others do not dilute the results. Screening items placed at the start save analysis effort later and keep the sample clean.
Closed questions offer fixed answer options and produce data that are easy to count and compare. They work well when the possible answers are known in advance, such as the type of housing, the vaccination status of a herd, or the frequency of a practice. Their weakness is that they force respondents into the author's categories, which may miss the respondent's real answer.
Open questions let respondents answer in their own words and are valuable when the range of answers is unknown, or when reasons and explanations matter. Their cost is analysis time: open answers must be read, coded, and grouped before they can be counted, and long open sections increase fatigue. Use open questions sparingly, usually one or two per questionnaire, where depth matters most.
Scale questions sit between the two: they offer ordered response options, such as agreement levels, for an attitude or perception. They are the workhorse of survey research because they produce numerical data that can be summarised and compared, while still letting respondents express degrees of opinion. The next sections explain how to write each type well.
Clarity starts with simple language. Use short sentences, everyday words, and one idea per question. Double-barrelled questions, such as "Do you vaccinate and deworm your animals regularly?", cannot be answered honestly by someone who does one but not the other, and the resulting data are meaningless. Split such questions into two.
Neutrality is equally important. Leading questions push respondents towards an answer: "Do you agree that good farmers vaccinate their animals?" invites agreement, not opinion. Remove praise, blame, and assumptions from the wording, and give every position a fair chance. Where a question touches a sensitive topic, neutral wording also improves honesty, because respondents feel less judged.
Check every question for hidden assumptions. "How often do you visit the veterinary clinic?" assumes the respondent visits one at all; a respondent who never visits has no honest answer. Add a filter or a "never" option wherever an assumption could exclude someone. A question that some respondents cannot answer truthfully is a design error, not a respondent error.
Watch for wording that assumes knowledge the respondent may not have. Asking about a specific regulation, drug, or programme by name will confuse respondents who have never heard of it, and their guesses will pollute the data. Where specialist knowledge cannot be assumed, add a brief explanation or a "not aware of this" option.
Order affects answers. Start with easy, non-sensitive questions that build confidence: background facts the respondent knows without thinking. Place sensitive or difficult questions later, once the respondent is engaged, and group related questions together so the survey feels logical rather than random.
Avoid order effects where the answer to one question changes the answer to the next. If an early question mentions a problem, later questions about that problem may be answered differently than they would have been alone. Where order effects are a real risk, consider rotating the order of blocks between respondents, and mention this in the methodology section.
Layout matters too, especially on paper and on phones. Keep one question per screen or block, use consistent formatting, and make the response options easy to tap or tick. In online surveys, show progress so respondents know how much remains; a survey that feels endless loses people, while one that shows progress keeps them going to the end.
The Likert scale is the most common survey scale: a statement with ordered agreement options, typically from "strongly disagree" to "strongly agree". Five options are standard; seven give finer distinctions but demand more from respondents. Whichever length is chosen, use it consistently across the questionnaire so respondents learn the pattern once and reuse it.
Every scale needs balanced options: equal numbers of positive and negative positions around a neutral middle, where a neutral position genuinely makes sense. An unbalanced scale, with more positive than negative options, pushes responses in one direction and biases the results. Label every point, not only the ends, because respondents interpret unlabelled middle points differently.
Other scales suit other purposes. Frequency scales ("never" to "always") fit behaviours; importance scales fit priorities; visual scales suit respondents with limited literacy. Choose the scale that matches the concept being measured, and avoid borrowing a scale from another study without checking that it fits this population and this question.
Be consistent about direction as well. If "strongly agree" sits on the right for most items, keep it there throughout, because flipping direction mid-questionnaire confuses respondents and creates errors. If a reversed item is used to catch careless answering, flag it clearly in the analysis plan so it is recoded before scoring.
No questionnaire should be launched without a pilot test on people similar to the target respondents. A pilot of even ten to twenty people will reveal ambiguous wording, missing options, questions that take too long, and technical problems in an online survey. These are exactly the problems that are cheap to fix before launch and expensive afterwards.
During the pilot, ask respondents to think aloud as they answer, or interview them afterwards about any question they found confusing. Watch for questions that everyone answers the same way, which may be too obvious or too leading, and for questions that are always skipped, which may be unclear or sensitive. Record completion times, since a pilot that takes twice as long as planned needs cutting.
Revise after the pilot and, if the changes are substantial, pilot again briefly. A questionnaire that changed significantly is effectively a new instrument, and launching it untested repeats the original risk. Report the pilot in the methodology section: who was included, what changed, and why. Reviewers treat a piloted questionnaire as far more credible than an untested one.
Keep the pilot data separate from the main study data. Pilot respondents have already seen the questions, so including their answers would double-count rehearsed responses. If pilot participants are later invited into the main study, note this in the methodology section so reviewers can judge whether it matters.
A good questionnaire given to the wrong people still produces the wrong data. Define the target population clearly, choose a sampling approach that can reach it, and record how respondents were invited. Convenience samples, such as sharing a link in one group, are common but limited, and their limits must be acknowledged when the results are reported.
Response rate matters because non-respondents often differ from respondents. A survey with a low response rate may describe only the most interested or most available people, not the whole population. Report the response rate honestly, describe any reminders sent, and discuss who is likely missing. A modest response rate with honest discussion is more credible than a high rate claimed without evidence.
Ethics apply to surveys as to any research with human participants. Explain the study's purpose, assure confidentiality, make participation voluntary, and obtain consent before the first question. For studies involving farmers, patients, or students, these steps are not optional extras; they are part of the design, and our guide to research ethics and consent explains the requirements in detail.
Analysis should follow the plan made before launch. Summarise closed questions with counts and percentages, always showing the number of respondents each percentage is based on, since a percentage without its denominator cannot be checked. For scale questions, report the distribution of responses, not only the average, because an average can hide a split between two opposed groups.
Handle missing data openly. Report how many respondents skipped each question, and do not silently drop incomplete questionnaires without saying so. If open answers were coded, describe the coding process and, where possible, have a second person check a sample of the coding. These details are what separate a trustworthy survey report from a casual one.
Finally, report limitations plainly, as our guide to writing the limitations section explains. State who was not reached, what the response rate was, and what the questionnaire could not measure. A survey that acknowledges its limits gives readers a reason to trust the findings it does claim.
| Question type | Produces | Best when | Main risk |
|---|---|---|---|
| Closed (fixed options) | Counts and percentages | Answers are known in advance | Forcing respondents into wrong categories |
| Open (own words) | Text to be coded | Reasons and new ideas matter | Slow analysis and respondent fatigue |
| Likert scale | Ordered numerical scores | Attitudes or perceptions are measured | Unbalanced or unlabelled points bias answers |
| Frequency scale | Ordered frequency scores | Behaviours and practices are measured | Vague anchors like "sometimes" mean different things |
| Ranking question | Order of preference | Priorities must be compared | Long lists confuse respondents |
| Filter question | Routing to relevant items | Some items apply only to a subgroup | Wrong routing loses valid answers |
These authoritative external resources explain the standards and guidance mentioned in this article:
How long should a research questionnaire be?
As short as the research question allows, ideally answerable in about fifteen minutes or less. Longer questionnaires suffer from fatigue, skipped questions, and abandonment, and the later questions end up with the worst data.
What is the difference between open and closed questions?
Closed questions offer fixed options and produce countable data; open questions let respondents answer in their own words and produce text that must be coded. Closed questions are faster to analyse, while open questions capture reasons and ideas the author did not anticipate.
How many points should a Likert scale have?
Five is the standard choice and works for most studies; seven gives finer distinctions but demands more from respondents. Use the same length consistently, label every point, and keep the scale balanced around a neutral middle.
Do I really need a pilot test?
Yes. A pilot with even ten to twenty people similar to the target respondents reveals ambiguous wording, missing options, and technical problems that are cheap to fix before launch and expensive afterwards. Report the pilot in the methodology section.
What is a good response rate?
There is no single standard; it depends on the population and the method. What matters more than the number is honest reporting: state the rate, describe reminders, and discuss who is likely missing from the sample.
Can I reuse a questionnaire from another study?
Yes, if it fits this population and research question, and with permission where required. Check that the wording, scale, and concepts still fit; a borrowed questionnaire that does not fit is worse than a short new one designed for the study.