Survey Design Basics for Student Research Projects

A badly worded question produces clean-looking data that means nothing. Here is how to write items, choose scales, order questions and pilot properly before you collect anything.

Surveys look like the easy option. They are not. A questionnaire is a measuring instrument, and a badly built one produces data that looks perfectly analysable and means nothing. Worse, you cannot fix it afterwards, because by the time you see the problem the responses are already in. Everything that matters in a survey study happens before you send it.

The Question Is the Instrument

Every statistical test you run assumes your questions measured what you think they measured. If an item is ambiguous, respondents answer different questions and you average the results as though they answered one. The resulting mean is arithmetically correct and substantively meaningless.

This is why survey methodologists spend most of their effort on wording rather than analysis. Pew Research Center, which runs some of the most heavily tested public survey instruments in the world, treats question wording and order as the central design problem rather than a detail to settle at the end.

Question Types and What They Cost

Type	Gives you	Costs you
Closed, single choice	Clean categorical data, easy analysis	Forces a choice that may not fit
Closed, multiple choice	Realistic where several apply	Harder to analyse, cannot sum percentages
Likert scale	Attitude strength, summable into scales	Acquiescence bias, ordinal debate
Ranking	Relative priority	Cognitively demanding beyond about five items
Numeric entry	Precise continuous data	Invites implausible values without validation
Open text	Unanticipated answers, participant voice	Slow to analyse, low completion

Open questions are the ones students overestimate. They feel richer, but response rates on them are low, answers are often a few words, and coding them properly takes far longer than expected. Use two or three deliberately, not a dozen hopefully.

Rules for Writing Items

Most bad items fail in one of a small number of recognisable ways.

Problem	Bad item	Better
Double-barrelled	The library is quiet and well stocked	Split into two items
Leading	How helpful did you find the excellent new portal?	How helpful did you find the new portal?
Loaded	Do you support wasteful spending on X?	Do you support increased spending on X?
Assumes behaviour	How often do you use the study spaces?	Filter first: do you use them at all?
Vague quantifier	Do you attend regularly?	How many of the last ten sessions did you attend?
Double negative	Should the policy not be withdrawn?	Should the policy be withdrawn?
Jargon	Rate the efficacy of formative assessment scaffolding	Use words your respondents actually use
Memory beyond recall	How many hours did you study last term?	Ask about last week, or offer bands

The vague quantifier problem deserves particular attention because it is so common and so invisible. "Regularly" means weekly to one respondent and termly to another, and you have no way of knowing which. Any question whose answer depends on the respondent's private definition of a word is producing noise.

Choosing Response Scales

Number of points. Five and seven are standard. Below five loses discrimination, above nine adds little beyond apparent precision.
Midpoint or not. An odd number allows genuine neutrality. An even number forces a direction, which is defensible only when you are confident neutrality is not a real position.
Label every point. Fully labelled scales are more reliable than scales labelled only at the ends, because respondents interpret unlabelled middle points inconsistently.
Balance the scale. Equal numbers of positive and negative options, with symmetrical wording.
Separate "don't know" from neutral. They are different answers, and merging them destroys information.
Keep direction consistent across the questionnaire, or respondents will answer on autopilot and you will silently reverse-code errors into your data.

Order Effects Are Real

Question order changes answers. A general satisfaction question placed after a series of specific complaints will score lower than the same question placed first, because the earlier items have primed what the respondent is thinking about.

Practical consequences. Put general questions before specific ones on the same topic. Put sensitive items later, once some trust is established, and demographics at the end unless you need them for filtering. Group related items so the respondent is not switching context repeatedly. And where order could plausibly bias a key measure, randomise the block and say so in your methods.

Length and Attrition

Length	Realistic expectation
Under 5 minutes	Best completion, suitable for most student projects
5 to 10 minutes	Workable with a motivated population
10 to 20 minutes	Noticeable dropout, needs a strong reason to participate
Over 20 minutes	Substantial attrition and straight-lining near the end

Every item should earn its place by mapping to a research question. Build the mapping table before you build the questionnaire, and delete anything that maps to nothing. Questions included because they are interesting are the reason surveys become too long.

Pilot Before You Launch

Piloting is the step most student projects skip and the one that saves the project. It need not be elaborate. Five to ten people from a similar population, asked to complete it and then talk you through what they thought each question meant, will surface most problems.

Ask them specifically: was any question unclear, did any answer options not fit, was anything uncomfortable, and how long did it take. Then fix the instrument. Pilot data are not usually included in the final analysis, and if you change items after piloting they cannot be, so say what changed in your methods.

Sample and Response Rate

Two numbers matter and students often report neither. The sample size should come from a power calculation for your planned analysis, not from a round figure. The response rate should be reported, because a survey of 200 people drawn from a list of 4,000 is a very different object from a survey of 200 drawn from a list of 220.

Non-response is not random. People with strong views and people with time respond more. You cannot eliminate this, but you can report the response rate, compare your respondents with the known population on any characteristic you have, and discuss the likely direction of the bias.

Ethics for Surveys

Information and consent at the start, with a clear statement that participation is voluntary.
Anonymity or confidentiality, and be precise about which you are offering. Anonymous means you cannot identify respondents at all, which also means you cannot honour a withdrawal request after submission. Say so.
A withdrawal point. If responses are anonymous, state that withdrawal is only possible before submitting.
Sensitive topics need signposting to support, and often need questions to be optional.
Data handling. Where the data are stored, for how long, and who can access them. Free survey platforms may store data outside your jurisdiction, which your ethics committee will ask about.

Common Mistakes and Fixes

Mistake	Fix
Writing the questionnaire before the research questions	Map every item to a research question first
No pilot	Pilot with five to ten people. It is the highest-return hour available
Compulsory answers on every item	Forcing answers on sensitive items is an ethics problem and inflates missing-data workarounds
Response rate not reported	Report it, with the denominator
Reverse-coded items not reversed before analysis	Check your coding before running anything
Using a validated scale but changing the wording	Changing items breaks the validation. Use it intact or justify the change
Analysis plan decided after seeing the data	Pre-specify it. Choosing tests to suit results is a integrity problem

Where Ethical Support Fits

Survey design is a technical skill and there is nothing improper about getting help with it. Supervisors comment on draft questionnaires, statistics advisers will tell you whether your planned analysis matches your scale types, and ethics committees will flag consent wording before you launch rather than after.

Piloting is itself a form of legitimate help, since asking people to critique your instrument is the method working as designed. What stays yours is the research question, the interpretation of the results and the honest reporting of what the data can support.

Frequently Asked Questions

How many questions should a student survey have?

Few enough to complete in under ten minutes, which usually means 15 to 30 items depending on complexity. Every item should map to a research question.

Can I use a published questionnaire?

Often yes, and it is usually a better choice than writing your own, since validated instruments come with established reliability. Check the licence, cite the source, and do not alter the items.

Is a Likert item ordinal or continuous?

A single item is ordinal. A summed multi-item scale is commonly treated as continuous, and that is widely accepted. State which you chose and why.

What response rate is acceptable?

There is no threshold, and student surveys often run low. What matters is that you report it and discuss non-response bias rather than leaving the reader to guess.

Should I offer an incentive?

It raises response rates but needs ethics approval, since a large incentive can be seen as inducement. Prize draws are common and usually acceptable.

Can I distribute my survey on social media?

You can, but recognise what it produces. The resulting sample is self-selected and its boundaries are unknowable, so describe the recruitment route precisely and bound your claims accordingly.

Your Next Step Today

Build the mapping table before touching the questionnaire: one column for each research question, one row for each proposed item. Any item with no research question beside it comes out, and any research question with no item is a gap you would otherwise have discovered after collecting data.

Trusted Sources

Pew Research Center, Writing Survey Questions, on wording, order and response options. Accessed 11 August 2026.
University of Southern California, Quantitative Methods. Accessed 11 August 2026.
University of Southern California, The Methodology. Accessed 11 August 2026.
Andrade C, Sample Size and its Importance in Research. Open access. Accessed 11 August 2026.
Tavakol M and Dennick R, Making sense of Cronbach's alpha, on scale reliability. Accessed 11 August 2026.

Survey and ethics requirements are set by your institution. Where this guide and your department's guidance differ, follow your department.