Start from the question, not the questionnaire
List the constructs you need to measure (for example job satisfaction, intention to stay, perceived supervisor support) and the demographic or control variables. Then, for each, find a validated scale from the literature if one exists. Using an established scale saves time and gives you evidence of reliability and validity. Write your own items only for what is not covered, and mark them as new.
| Step | Output |
|---|---|
| List constructs from the framework | A measurement plan |
| Find validated scales | Items with source citations |
| Add demographics and controls | Only those you will use in analysis |
| Draft and order the questionnaire | First version |
| Pilot test | Revisions and timing |
| Collect and clean data | Analysis dataset |
Write questions that people can answer
| Problem | Weak item | Better item |
|---|---|---|
| Double-barreled | My manager is supportive and communicates clearly. | Two items: My manager is supportive. My manager communicates clearly. |
| Leading | Don't you agree that training is excellent? | How would you rate the training you received? |
| Vague | I often attend meetings. | In a typical week, how many team meetings do you attend? |
| Jargon | Rate the efficacy of the onboarding intervention. | How helpful was your first-week induction? |
| Absolute words | I never feel stressed at work. | I usually feel able to manage my workload. |
Use a consistent scale. A five-point or seven-point agreement scale (strongly disagree to strongly agree) is typical for attitudes, with clear labels on each point. Place easy and engaging questions first, sensitive questions later and demographics at the end. Keep the survey short: completion drops sharply beyond 10 to 15 minutes. Include an attention check only if needed, and a clear consent statement at the start.
Pilot test
Test the survey with 5 to 15 people similar to your sample. Ask them to think aloud as they answer, and note where they hesitate, misread or skip. Measure the time taken. Revise unclear items, remove duplicates and check that the logic and skip patterns work. If you use a validated scale, check that your sample understands the wording, especially if translated.
Sample size calculations, worked
For a proportion (for example the share of customers who would repurchase), the required sample size with a given confidence level and margin of error is n = z squared x p x (1 - p) / e squared, where z is 1.96 for 95 percent confidence, p is the expected proportion and e is the margin of error. If you do not know p, use 0.5, which gives the largest (safest) sample.
Sample size for a proportion
95 percent confidence (z = 1.96), p = 0.5, margin of error e = 0.05 (plus or minus 5 points).
n = 1.96 squared x 0.5 x 0.5 / 0.05 squared = 3.8416 x 0.25 / 0.0025 = 384.2, so 385 responses.
Finite population correction (if the whole population is only 2,000): adjusted n = 384.2 / (1 + (384.2 - 1) / 2,000) = 384.2 / 1.1916 = 322.4, so 323 responses.
For a mean (for example average satisfaction on a five-point scale), n = (z x s / E) squared, where s is the standard deviation and E is the margin of error in scale points.
Sample size for a mean
Expected standard deviation s = 1.2, margin of error E = 0.15 scale points, 95 percent confidence.
n = (1.96 x 1.2 / 0.15) squared = (15.68) squared = 245.9, so 246 responses.
Planning for the response rate matters as much as the formula. To receive 385 completed surveys with a 25 percent response rate, you must invite 385 / 0.25 = 1,540 people. Quantitative analyses such as regression have their own needs, often at least 10 to 15 cases per predictor as a rough guide, so check the requirement for your planned analysis.
| Margin of error | Sample needed (large population, 95%, p = 0.5) |
|---|---|
| 10 points | 97 |
| 7 points | 196 |
| 5 points | 385 |
| 3 points | 1,068 |
Bias, response rates and data quality
A large sample does not fix a biased one. Think about who is missing and why.
| Bias | Cause | Reduction |
|---|---|---|
| Coverage | Sampling frame leaves out part of the population | Use the best available list; describe gaps |
| Non-response | People who respond differ from those who do not | Reminders, short survey, compare early and late responders |
| Social desirability | Respondents give acceptable answers | Anonymity, neutral wording |
| Common method bias | All data from one source at one time causes inflated links | Separate sources; vary scales; statistical checks |
| Self-selection | Volunteers are more engaged | Be honest about limits; weight where possible |
Report the response rate, the profile of respondents compared with the population and how you handled missing data. For reliability, report Cronbach's alpha for each scale; values of 0.70 or higher are commonly treated as acceptable, though the threshold depends on the field and the scale length.
Working on this assignment now? Get a price for help with your paper.
Get an instant quoteReliability of a scale, worked
Cronbach's alpha measures how consistently the items of a scale hang together. A convenient form uses the number of items k and the average correlation r between items: standardized alpha = k x r / (1 + (k - 1) x r).
| Items (k) | Average inter-item correlation (r) | Calculation | Alpha |
|---|---|---|---|
| 3 | 0.50 | 3 x 0.5 / (1 + 2 x 0.5) = 1.5 / 2.0 | 0.75 |
| 6 | 0.50 | 6 x 0.5 / (1 + 5 x 0.5) = 3.0 / 3.5 | 0.86 |
| 6 | 0.30 | 6 x 0.3 / (1 + 5 x 0.3) = 1.8 / 2.5 | 0.72 |
Longer scales raise alpha even when items correlate only moderately, so a high alpha on a long scale does not prove the items measure one idea. Check the items and, for larger scales, use factor analysis. Report alpha for your own data, not just the value from the original paper.
Improving the response rate
| Action | Why it helps |
|---|---|
| Personal invitation with a clear purpose | People respond to a request that explains why it matters to them |
| Short survey with a stated time (for example, 8 minutes) | Lower effort and honest expectations |
| Mobile-friendly design | Many respondents answer on phones |
| Two or three reminders at spaced intervals | Many responses arrive after reminders |
| Sponsor or gatekeeper endorsement | Raises trust, if it does not pressure participants |
| Offer a summary of results | Reciprocity; keep within ethics rules |
Compare early and late responders on key variables. If late responders (a rough stand-in for non-responders) look similar to early ones, non-response bias is less likely, which you can report as a limited check.
A model consent statement
Opening screen (hypothetical)
You are invited to take part in a study on how new employees experience their first year at regional banks. The survey takes about 8 minutes. Taking part is voluntary, and you may stop at any time. Your answers are anonymous: no names are collected, and results are reported only for groups of ten or more. Data are stored on a password-protected university system and deleted five years after the study ends. Questions about the study can be sent to the researcher at the contact address below. By selecting Continue, you confirm that you are 18 or over and agree to take part.
Use your institution's template and wording, which will include required contact details and approval numbers. The statement should say what happens to the data, who will see it and how participants can withdraw.
Reporting your survey method
- Describe the population and sampling frame And how participants were contacted.
- Justify the sample size Show the calculation and the response rate.
- Cite scales With their reliability from earlier work and from your data.
- Report the pilot What you changed.
- Address ethics and anonymity Consent, storage and the right to withdraw.
- Discuss limits Bias, response rate and the generalization boundary.
Move on to analysis in our data analysis chapter guide. If you want help with a survey project, you can order MBA dissertation help or business analytics assignment help.