Step 1: Start from your research questions
List the research questions or hypotheses. For each one, write down the variables you need and how you will analyse them. If a question does not feed an analysis, drop it. Shorter surveys get better completion and better data.
Step 2: Reuse validated scales where you can
For concepts such as satisfaction, stress or self-efficacy, a published scale with known reliability is usually better than inventing your own. Check permission terms and keep the wording and response options as published unless you plan to re-test the scale.
Step 3: Write clear questions
- Ask one thing per question. “How satisfied are you with the price and quality?” is two questions.
- Avoid leading wording such as “Don’t you agree that…”.
- Use plain language and define technical terms.
- Give a clear time frame: “in the last 7 days”, not “recently”.
- Make response options exhaustive and mutually exclusive, and add “Other” or “Prefer not to say” where sensible.
Step 4: Choose response scales
Likert-type scales usually have 5 or 7 points with clear labels. Keep direction consistent, and mark any reverse-coded items for later recoding. Use numeric inputs when you need exact values such as age, and use bands only when precision is not needed.
Step 5: Order and layout
Start with easy, engaging questions and place sensitive questions later. Group related questions and keep skip logic simple. Test how the survey looks on a phone, since many respondents will use one.
Step 6: Plan your sample
Define the population, the way you will reach people and the sample size you need. For estimating a proportion, you can use our free sample size calculator. For regression or PLS-SEM, see the sample size guide. State clearly who your results represent.
Step 7: Pilot and revise
Test with a small group from your target audience. Ask what was unclear, time the survey and check that the data export correctly. Fix problems before launch; after launch, changes make earlier responses incomparable.
Step 8: Prepare for analysis
Create a codebook with variable names, labels and coding before data arrive. After collection, clean and check the data, then analyse. Our survey data analysis and Likert scale analysis services cover the next steps.
Bad questions and better versions
- Double-barrelled: “How satisfied are you with the price and quality?” → Ask two questions, one about price and one about quality.
- Leading: “Don’t you agree that the training was excellent?” → “How would you rate the training?”
- Vague time frame: “Do you often exercise?” → “In the past 7 days, on how many days did you do at least 30 minutes of exercise?”
- Jargon: “Do you have access to WASH facilities?” → “Does your household have a toilet and a place to wash hands with soap?”
- Overlapping options: “18–25, 25–35” → “18–24, 25–34, 35–44”.
Response scales that work
- Label every point, or at least both ends, so everyone reads the scale the same way.
- Keep the direction consistent, for example always from negative to positive.
- Include “Don’t know” or “Not applicable” when they are real answers, and decide in advance how you will treat them in analysis.
- Avoid long lists of identical agree/disagree items, which encourage straight-lining.
Choosing a sampling method
- Simple random sampling: everyone has an equal chance. Needs a complete list of the population.
- Stratified sampling: random samples within groups such as region, so each group is represented.
- Cluster sampling: randomly choose clusters such as villages or schools, then survey within them. Practical in the field, but needs a larger sample.
- Convenience or purposive sampling: easy to run but cannot support claims about a whole population. Say so in your limitations.
A sample size example
To estimate a proportion with a 95% confidence level and a 5% margin of error, the usual formula is n = z² × p × (1 − p) ÷ e², using p = 0.5 when you have no prior estimate. That gives 1.96² × 0.25 ÷ 0.05² ≈ 384. If your population is small, for example 2,000 people, a finite population correction reduces this to about 322. If you expect 20% of people not to respond, divide by 0.8 to get about 403 invitations. Our sample size calculator does this for you. Note that this formula applies to estimating a proportion; comparing groups or running regression needs a different calculation.
Pilot checklist
- Can respondents explain each question in their own words?
- Are any questions skipped or answered with “Other” unexpectedly often?
- How long does it take, and does the survey work on a phone?
- Does the exported data match your codebook?
- For multi-item scales, is reliability reasonable? Our Cronbach’s alpha calculator can check it.
Ethics and consent
State the purpose of the study, who is running it, that taking part is voluntary and how the data will be stored and protected. Ask only for personal information you need. If your institution requires ethics approval, get it before the pilot, not after.
Frequently asked questions
How long should a questionnaire be?
As short as possible while answering your research questions. For most online surveys, aim for under 10 minutes.
5-point or 7-point scale?
Both work. Five points are simpler for respondents and seven can add sensitivity. Clear labels and consistency matter more.
How do I check if my questions are clear?
Pilot the survey with a few people from your target group and ask them to explain what each question means to them.
What sample size do I need?
It depends on your analysis, expected effect and desired precision. Use a sample size calculator and allow for non-response.
Can I translate an existing questionnaire?
Yes, but have a qualified translator do it and test it with a small group. Formal back-translation is recommended for validated scales.
Related tools and services
Want an expert to check or run your analysis?
Send us your question and data description. We reply with a clear plan and a quote.
Discuss your project on WhatsApp