Writing questions is the easy part. The hard part is making sure the questions add up to valid measures of the concepts in your research question, that respondents understand them the way you intend, and that enough people answer.
This post walks through the process from constructs to items, the question types and scales, wording rules, order and layout, piloting, response rates and reporting.
Start From Constructs, Not Questions
- List your constructs. Every concept in your research question or conceptual framework that you need to measure.
- Define each one. A one-sentence definition from the literature.
- Look for an existing scale. Validated scales exist for most common constructs in psychology, management, education and health. Using one makes your results comparable and your validity argument far easier. Cite it and report its original reliability.
- Write items only where no scale exists, and plan to test their reliability yourself.
- Map every item to a construct and an analysis. If you cannot say which hypothesis an item serves, delete it.
Question Types
| Type | Use for | Analysis |
|---|---|---|
| Dichotomous (yes/no) | Simple facts and screening | Frequencies, chi-square |
| Multiple choice | Categories such as role or sector | Frequencies, group comparisons |
| Likert scale | Attitudes and perceptions, usually as multi-item scales | Scale scores, correlation, regression |
| Rating (0-10) | Single evaluations such as satisfaction | Means, comparisons |
| Ranking | Relative priorities | Mean ranks, non-parametric tests |
| Numeric open | Age, hours, counts | Continuous statistics |
| Text open | Reasons and examples, used sparingly | Coding, thematic analysis |
The statistical test you can run depends on the type of answer you collect, so choose the type with the analysis in mind.
Likert Scales
- Use five or seven points, with every point labelled.
- Keep the direction consistent: disagree on the left, agree on the right.
- Include a neutral midpoint unless you have a reason to force a choice.
- Measure each construct with three to six items, not one. Single items cannot be checked for reliability.
- Report internal consistency for each scale, usually Cronbach's alpha. See reliability and validity.
Turn survey results into a written thesis
ThesisAI drafts full academic documents with methodology, results and discussion chapters, with every citation checked against the paper it came from.
Începe să scriiWording Rules
Double-barrelled: "My manager is supportive and communicates clearly." Two questions in one; split them.
Leading: "How much has the excellent new system improved your work?" Remove the evaluation.
Vague time frame: "Do you exercise regularly?" Ask "On how many days in the last 7 days did you exercise for 30 minutes or more?"
Jargon: "How would you rate your organisational citizenship behaviour?" Respondents do not know your constructs.
Double negative: "I do not disagree that remote work is not productive." Rewrite plainly.
Better: "In the last month, my manager gave me useful feedback on my work." (Strongly disagree ... Strongly agree)
Further rules: one idea per question, short sentences, answer options that are exhaustive and do not overlap, an "other" or "not applicable" option where needed, and sensitive questions phrased neutrally.
Order and Layout
- A short introduction: purpose, time needed, anonymity, consent.
- Easy, engaging questions first.
- Group questions by topic, with a short heading for each section.
- Sensitive and demographic questions near the end.
- A thank-you and contact details for questions.
Keep it under 15 minutes. Completion rates fall sharply after that, and the last answers become less careful.
Piloting
Test the questionnaire on five to ten people from your target group before launch. Ask them to think aloud as they answer, time them, and check the exported data. Piloting catches misunderstood questions, broken skip logic and answer options that nobody uses. Report the pilot and what changed in your methodology chapter.
Response Rates and Sampling
- Calculate the sample size you need before launch. See sampling methods.
- Send a personalised invitation and two reminders a few days apart.
- Ask a gatekeeper the respondents trust to share the link.
- Report the response rate and discuss who is likely to be missing.
FAQs About Questionnaire Design
How many questions should a research questionnaire have?
As many as your constructs require and no more. Most student questionnaires work best at 20 to 40 items, which takes about 10 minutes.
Can I use an existing questionnaire?
Yes, and it is usually better. Cite the original authors, check whether permission is needed, and report any changes you made to the wording.
Is a Likert scale qualitative or quantitative?
Quantitative. Likert items are ordinal, and multi-item scale scores are usually treated as continuous in analysis.
Do I need ethics approval for an online survey?
Almost always, if you collect data from people. See ethical considerations in research.
Start from your constructs, borrow validated scales where they exist, write every item with its analysis in mind, and pilot before launch. A short, focused questionnaire that measures what it claims to will produce a stronger thesis than a long one that tries to measure everything.