Questionnaire Design: A Guide to Standard Questionnaires for Theses and Research (With Sample Questions)
The standard structure of a questionnaire, choosing a scale, validity (face, content with Lawshe's CVR table, construct) and reliability (Cronbach's alpha with a worked example), sample size, a complete 15-item sample, running it online, and exporting to SPSS.
A questionnaire is a measuring instrument, not a list of questions. The difference between the two is the difference between a scale and "guessing someone's weight": a standard questionnaire measures what it claims to measure (validity), and if it measures again it gives the same number (reliability). This guide is written for the student building the questionnaire for their thesis, and for the researcher or HR specialist who wants results that will hold up in front of a demanding reviewer. Everything comes with worked numbers and tables, and at the end there's a complete 15-item sample.
How does a questionnaire differ from a survey?
| Survey | Research questionnaire | |
|---|---|---|
| Purpose | Describing a group's opinion for a decision | Measuring a construct (latent variable) to test a hypothesis |
| Unit of analysis | Each question on its own | The sum of several items that make up one construct |
| Requirement | Clarity and neutrality | + reported validity and reliability |
| Output | Percentages and charts | Construct scores, correlation, regression |
If your goal is an operational decision (customer satisfaction, choosing between options), read the guide to creating an online survey; this article is about the research instrument.
The standard structure of a questionnaire
- Introduction and informed consent page. Who the researcher is and their university, the purpose in two sentences, the approximate time, a guarantee of confidentiality or anonymity, the voluntary nature of participation, and a way to get in touch. Reviewers read this page.
- Response instructions. "There are no right or wrong answers; mark your first impression." Plus the meaning of each point on the scale.
- Demographic questions. Only those used in the analysis (age as a range, gender, education, tenure). Some researchers place these at the end so their sensitivity doesn't reduce response to the main section; both practices are common.
- Construct sections. Each construct (say, "job satisfaction") gets one section with a heading, 4 to 8 items, and the same scale throughout. The items of each dimension sit together.
- A closing open-ended question (optional) and thanks.
An important rule: each item is one short declarative statement with one concept ("I am satisfied with my salary"), not a question, not two concepts, not a double negative.
Standard or researcher-made questionnaire?
Standard means an instrument that has already been built, validated and published, and that you use with a citation, such as the Minnesota Satisfaction Questionnaire (MSQ), Allen and Meyer's organizational commitment scale, SERVQUAL for service quality, or the Maslach Burnout Inventory. The advantage: it comes with reported validity and reliability, and reviewers raise fewer objections. The condition: a validated translation and a re-test of reliability in your own sample (because the instrument was built in a different population). Where to find a trustworthy version, and how to spot a fake "standard" file: what a standard questionnaire is.
Researcher-made is for when no ready instrument exists for your construct. Then you have to walk the full path: conceptual and operational definition, an item pool, expert judgment (content validity), a pilot with 30 people, calculating alpha, revising, and running it.
Choosing the scale
| Scale | Form | Use | Permitted analysis |
|---|---|---|---|
| 5-point Likert | Strongly disagree … strongly agree | Attitude, satisfaction; the most common | Ordinal; the sum of items is approximately interval |
| 7-point Likert | The same with two more points | When fine differences matter and the audience is educated | As above; more variance (5 or 7?) |
| Semantic differential | Two opposite adjectives at the ends (slow … fast) | Brand image, evaluation | Interval-like |
| Frequency | Never / rarely / sometimes / often / always | Behavior | Ordinal |
| Yes/No | Dichotomous | Facts | Nominal |
| Numeric 0 to 10 | NPS and intensity | Loyalty, pain | Interval-like |
Three notes: (1) use one scale throughout a construct; (2) label every point; (3) if you have a reverse-coded item ("I am tired of my job" in a satisfaction construct), reverse its score before summing (1↔5, 2↔4). A reverse-coded item is useful for catching careless respondents, but don't put more than one or two in each construct.
Validity: are you measuring what you claim to measure?
Face validity
A few people from the target population (not your colleagues) read the items and say whether they seem clear and relevant. It's cheap, and it's the first filter for ambiguous sentences.
Content validity with Lawshe's CVR
Give each item to a panel of experts (usually 8 to 15 people) and ask them to rate it "essential / useful but not essential / not necessary." For each item:
CVR = (n_e − N/2) / (N/2)where n_e is the number of experts who said "essential" and N is the total number of experts. An item whose CVR falls below the minimum in Lawshe's table is dropped or rewritten:
| Number of experts | Minimum CVR | Number of experts | Minimum CVR |
|---|---|---|---|
| 5 | 0.99 | 12 | 0.56 |
| 6 | 0.99 | 13 | 0.54 |
| 7 | 0.99 | 14 | 0.51 |
| 8 | 0.75 | 15 | 0.49 |
| 9 | 0.78 | 20 | 0.42 |
| 10 | 0.62 | 25 | 0.37 |
| 11 | 0.59 | 30 | 0.33 |
Example: 10 experts; for the item "I am satisfied with the welfare facilities," 8 said "essential": CVR = (8 − 5) / 5 = 0.6, which is below 0.62, so the item is rewritten. Lawshe called the mean CVR of the remaining items the content validity index (CVI); the item-level CVI based on relevance ratings from 1 to 4 (keep items with an I-CVI above 0.79) is a different index. Both, with a "minimum expert votes" table, are worked through in questionnaire validity and reliability.
Construct validity
For a master's thesis this is most often reported with factor analysis (exploratory or confirmatory): the items of one construct should load on one factor (factor loading above 0.4 or 0.5) and KMO should be above 0.7. If an item lands on another factor, it's in the wrong place.
Reliability: if you measure again, do you get the same number?
Cronbach's alpha
The most common index of internal consistency. It's calculated separately for each construct:
α = (k / (k − 1)) × (1 − Σ s²_i / s²_total)k is the number of items, s²_i the variance of each item and s²_total the variance of the construct's total score. The usual interpretation:
| Alpha | Interpretation | Next step |
|---|---|---|
| Above 0.9 | Excellent (you may have redundant items) | The questionnaire can be shortened |
| 0.8 to 0.9 | Good | — |
| 0.7 to 0.8 | Acceptable | The conventional minimum for a thesis |
| 0.6 to 0.7 | Weak | Find the problem item with "alpha if item deleted" |
| Below 0.6 | Unacceptable | Redesign the construct or the items |
Worked example: a construct with 5 items, item variances of 1.1, 0.9, 1.3, 1.0 and 1.2 (sum 5.5), and a total-score variance of 16.0: α = (5/4) × (1 − 5.5/16) = 1.25 × 0.656 ≈ 0.82, which is good. If alpha is low, the "Cronbach's alpha if item deleted" column in SPSS tells you which item's removal gives the biggest improvement; a reverse-coded item you forgot to flip is the first suspect.
Test-retest reliability
Give the same questionnaire to 20 to 30 of the same people 2 to 4 weeks later and compute the correlation between the two rounds; above 0.7 is acceptable. It suits stable constructs (personality) better than state-like ones (daily mood).
The pilot: a step you must not skip
Run the questionnaire with 30 people from the target population, time it, compute alpha, and ask a few of them which items were confusing. Most reviewer objections are prevented at this very stage.
Sample size: Cochran or Morgan?
For a known population, the Morgan table (Krejcie and Morgan) gives roughly the same result as Cochran's formula with p = 0.5 and a 5% margin of error (population 300 → 169, 1,000 → 278, 10,000 → 370). For an unlimited population, 384 to 385 people. If you're doing factor analysis, also respect the rule of thumb of "5 to 10 people per item"; a 40-item questionnaire needs at least 200 responses. Get the exact number from the Cochran calculator and state in your methodology chapter which parameters you used. If your hypotheses are about correlation or regression, Cochran alone isn't enough; power analysis and worked examples are in how to determine sample size.
State your sampling method too, and honestly: "convenience" is not the same as "stratified random," and the reviewer knows the difference. For random sampling, build a list of individuals and send personal invitations to the selected sample so that each person answers once.
A complete sample: job satisfaction questionnaire (15 items)
Introduction: "Researcher: …, master's student in Management, University of …. This questionnaire has been designed for a thesis on the relationship between job satisfaction and turnover intention. Responses are completely anonymous and will be reported only in aggregate statistical form. It takes about 4 minutes to complete. Thank you for your participation."
Section A — General information: gender; age range (under 30 / 30–39 / 40–49 / 50 and over); education; years of service in this organization (as a range).
Section B — Job satisfaction (5-point scale: strongly disagree = 1 to strongly agree = 5)
| # | Dimension | Item |
|---|---|---|
| 1 | Nature of the work | My work is meaningful and worthwhile to me. |
| 2 | Nature of the work | I am satisfied with the variety of tasks I perform. |
| 3 | Nature of the work | My work matches my abilities. |
| 4 | Pay and benefits | My salary is commensurate with the work I do. |
| 5 | Pay and benefits | The organization's non-cash benefits are satisfactory to me. |
| 6 | Pay and benefits | Compared with similar organizations, my pay is fair. |
| 7 | Supervisor | My supervisor listens to my opinions. |
| 8 | Supervisor | I receive clear feedback from my supervisor about my performance. |
| 9 | Supervisor | My supervisor treats all team members fairly. |
| 10 | Coworkers | My coworkers help me when I need it. |
| 11 | Coworkers | The atmosphere of cooperation in our unit is friendly. |
| 12 | Promotion | I have opportunities for career advancement in this organization. |
| 13 | Promotion | The promotion criteria are transparent and fair. |
| 14 | Overall | All in all, I am satisfied with my current job. |
| 15 | Overall (reverse-coded) | I often think about leaving this organization. |
Open-ended question (optional): "If one thing in your work environment could change to increase your satisfaction, what would it be?"
This structure, five dimensions with 2 to 3 items each, one overall item and one reverse-coded item, is the skeleton you see in most standard questionnaires. For real research, bring each dimension up to 4 to 6 items so that alpha stabilizes.
Running it online: what to enable in the tool
- Real anonymity. Turn on anonymous mode so that the IP address and device details are not stored, and that statement is visible on the form; you can cite this in your ethics section.
- One response per person. If the respondents are known (the employees of one organization), send single-use invitations; if the link is public, turn on "one response per mobile number." Duplicate responses inflate reliability on paper and invalidate the result.
- Make construct items required. Construct items should be required so that missing data doesn't ruin the factor analysis; demographics can stay optional.
- A matrix for each dimension. Put the items of each dimension into one matrix question with a shared scale; on mobile it automatically becomes cards, and the respondent doesn't read the scale 15 times. At most 6 rows per matrix.
- Attention-check questions. One reverse-coded item per construct, and in the report, look separately at "straight-line" responses (the same column in every row) and "very fast" ones, and explain their removal in your methods report.
- Deadline and cap. Set a closing date and a response cap so that collection stays within the period you report.
Porsino's online questionnaire builder has all of this, and building and publishing are free; the full procedure, from the informed-consent page to the SPSS export, is in running your thesis questionnaire online. If you're starting from a ready-made standard questionnaire, several common ones (job satisfaction, customer satisfaction, course evaluation) can be copied from the templates.
Exporting to SPSS and Excel: code it from day one
- A short variable name with no spaces for every item (set the field key to JS1, JS2, … from the start so you get the same columns in the Excel export).
- Numeric option values: for scales, define the option value as 1 to 5, not text; then the export goes straight into SPSS.
- Reverse-coded items: flip them in SPSS with Recode (6 − x for a 5-point scale) and say so in your report.
- Construct score = the mean (not the sum) of that dimension's items, so that constructs with different numbers of items remain comparable.
- Keep one anonymous ID column (a tracking code or row number) so that deletions can be traced.
If keys and values are set correctly from the start, the direct SPSS (.sav) export carries the variable names, codes and labels. The step-by-step analysis that follows (reverse-coding, alpha, dimension scores and a t-test against 3) is in analyzing a Likert questionnaire in Excel and SPSS.
The demanding reviewer's checklist
- The conceptual and operational definition of each construct is written, and the items trace back to it.
- The source of the standard questionnaire, or the construction steps of the researcher-made one, is reported.
- Face and content validity (CVR/CVI with the number of experts) are included.
- Cronbach's alpha for each construct separately, in both the pilot and the main sample.
- Sample size with the formula and explicit parameters, and an honest sampling method.
- The response rate, the number of discarded questionnaires and the reason.
- Ethical considerations: informed consent, anonymity, voluntary participation.
- The full questionnaire in the appendix.
Common student mistakes
- Translating a foreign questionnaire on their own, without back-translation and without reporting reliability in the Iranian sample.
- One alpha for the whole questionnaire instead of one per construct.
- Forgetting to reverse-code, then being surprised by an alpha of 0.4.
- Double-barreled items ("My salary and benefits are adequate").
- The 60-item questionnaire where everyone ticks the same column after item 30.
- A public link in Telegram groups with no duplicate control, followed by a claim of "random sampling."
- Not removing the 40-second responses.
Frequently asked questions
What is the minimum Cronbach's alpha for a thesis?
0.7 is conventional; between 0.6 and 0.7 is sometimes accepted in exploratory research if you explain it. Above 0.95 is a sign of redundant items.
5-point or 7-point Likert?
5 for the general population; 7 for an educated audience and when you want more variance. Pick one within a single questionnaire.
How many items per construct?
At least 3 (for confirmatory factor analysis), usually 4 to 6. More than 8 items are usually redundant.
Demographic questions first or last?
If they're sensitive (income, marital status), last; if they're needed for conditional logic (for example, only permanent employees see Section B), first.
Can I run the questionnaire on paper and online at the same time?
Yes, but say so in your methods chapter and control for the effect of the administration mode; the simplest approach is to enter the paper responses into the same online form (kiosk mode) so you have a single database.
How do I export for SPSS?
In the Responses tab, choose SPSS (.sav) from the Export menu; variable names are built from the question keys and the codes and labels are ready. An Excel export with one row per response and one column per item is there too; if you defined the option values as numbers, it goes into SPSS unchanged.
To get started, open Porsino's online questionnaire builder or copy a sample questionnaire; before distributing, calculate your sample size, and to understand the common biases in writing items, read this short article.