How to Conduct a Survey: Steps & Examples

What a survey is — and when to use one

A survey is a systematic method of gathering self-reported information from a defined group of respondents using a standardised instrument (the questionnaire). Because every respondent answers the same questions in the same form, the resulting data are comparable and can be aggregated, counted, cross-tabulated and tested statistically. Surveys most commonly produce quantitative data, but open-ended questions can also yield qualitative material.

Methodologists such as Saunders, Lewis and Thornhill position the survey as a research strategy typically associated with a deductive approach: you start with a theory or hypothesis, operationalise it into measurable questions, and collect data to test it. Choose a survey when you want to:

• Describe the characteristics, attitudes, opinions or behaviours of a population (e.g., how satisfied final-year students are with academic support).
• Measure the prevalence of something (how many, how often, how much).
• Examine relationships or associations between variables (does workload predict stress?).
• Compare groups (do part-time and full-time students differ in engagement?).
• Collect data from a large, geographically dispersed sample at relatively low cost.

A survey is a weaker choice when you need deep, contextual understanding of why people think or behave as they do — there, interviews, focus groups or ethnographic work serve better. If you are still deciding between fixed and open-ended designs, our guides to the quantitative research questionnaire and the qualitative research questionnaire set out the trade-offs.

How to conduct a survey: the step-by-step process

The eight steps below run in sequence, but expect to loop back — piloting often sends you back to redrafting questions, and analysis sometimes exposes objectives you never properly operationalised.

Step 1 — Define your objectives and research questions

Begin with the decision the survey must inform, then translate it into specific, answerable research questions and (for explanatory work) hypotheses. Every question on your final instrument should earn its place by mapping back to an objective. A useful discipline is to write a one-line analysis plan for each item: “This question measures X; I will report it as Y; it answers research question Z.” If you cannot complete that sentence, cut the question. Vague aims (“understand the student experience”) produce sprawling, unfocused questionnaires; sharp aims (“estimate satisfaction with feedback turnaround and test whether it varies by year of study”) produce tight ones.

Step 2 — Define the population and choose a sampling method and size

Specify your target population (the entire group you want conclusions about) and your sampling frame (the accessible list from which you actually draw respondents). The gap between the two is a common source of bias. Then choose a sampling technique — probability methods (simple random, systematic, stratified, cluster) support statistical generalisation, whereas non-probability methods (convenience, quota, purposive, snowball) are quicker but limit how far you can generalise. Our dedicated guide to sampling methods of research compares each in detail.

Sample size is driven by the precision you need (margin of error), the confidence level, the variability in the population, and — for hypothesis tests — the effect size you want to detect and your desired statistical power. As a rough orientation, many descriptive student surveys aim for several hundred completed responses; for inferential work, a formal power calculation is preferable to a rule of thumb. Always over-recruit, because non-response will erode your achieved sample.

A practical worked illustration helps. If you are estimating a single proportion (say, the percentage of students satisfied with feedback) and want a 95% confidence level with a ±5% margin of error, the classic formula returns a required sample of roughly 384 respondents for a very large population — a figure you can then adjust downward with a finite-population correction when your population is small. If you anticipate a 40% completion rate, you would need to invite around 960 students to land that achieved sample. Building this arithmetic into your plan early prevents the common disaster of discovering, after data collection has closed, that your sample is too small to support the comparisons your research questions demand.

Step 3 — Choose a survey mode

The administration mode shapes cost, reach, response rate and data quality. The table below compares the four main modes.

Mixed-mode designs (e.g., email invitation with a phone follow-up) can lift response rates and reduce coverage bias, at the cost of added complexity and possible mode effects.

Step 4 — Design the questionnaire

This is where most surveys are won or lost. Work through four design decisions.

Question types. Use closed questions — dichotomous (yes/no), multiple choice, ranking, and rating scales — for clean, codeable quantitative data, and a small number of open-ended questions for nuance you cannot anticipate. Match the response options to the variable’s measurement properties you intend to analyse.

Scales. Likert-type scales (e.g., “Strongly disagree” to “Strongly agree”) are the staple for attitudes. Decide deliberately on the number of points (5- and 7-point scales are common), whether to include a neutral midpoint, and whether to offer “Don’t know” / “Not applicable.” Keep scale direction and labelling consistent throughout.

Wording. Use plain, unambiguous language; one idea per item. Avoid the four classic faults:

• Leading questions that signal a “correct” answer (“How excellent was your tutor’s support?”).
• Double-barrelled questions that ask two things at once (“Was the feedback timely and useful?”) — split them.
• Loaded or emotive terms and unexplained jargon or acronyms.
• Recall-heavy or assumptive questions (“How many lectures did you miss last year?” assumes the respondent missed some).

Sequencing. Open with easy, engaging, non-threatening questions to build momentum; group related items together; move sensitive or demographic questions towards the end; and use a logical funnel from general to specific. Apply skip logic so respondents only see relevant items. For a fuller treatment of instrument construction, see our guide to methods of data collection.

Two further design habits separate amateur instruments from professional ones. First, write a codebook as you draft, assigning each question a variable name and recording how each response will be coded numerically. This forces you to think about analysis at the design stage and makes the eventual dataset far easier to clean. Second, keep the instrument honest about length: every additional item costs you respondents through fatigue and drop-out, so prune ruthlessly to the questions your objectives genuinely require. A focused twelve-question survey that 41% of your sample completes will almost always beat a forty-question survey that 12% abandon halfway through. Finally, mind your response-option design — make categories mutually exclusive and collectively exhaustive, anchor and label scale points consistently, and decide deliberately whether “Don’t know” is a legitimate substantive answer or simply an escape hatch that inflates missing data.

Step 5 — Pilot test

Never field a questionnaire untested. Run a pilot with a small group resembling your target population (often 10–30 respondents) to check that questions are understood as intended, that the flow and skip logic work, that completion time is acceptable, and that the response options capture real answers. Cognitive interviewing — asking pilot respondents to think aloud as they answer — surfaces misreadings you would otherwise never see. Use the pilot to refine wording and to check provisional reliability (for example, the internal consistency of multi-item scales) before committing to full data collection.

Step 6 — Distribute and maximise the response rate

A technically perfect survey is worthless if too few people respond, because non-response bias distorts results when non-responders differ systematically from responders. Evidence-based tactics to lift response rates include:

• A clear, credible invitation explaining purpose, sponsor, anonymity and how long it takes.
• Pre-notification and one to three polite reminders to non-responders.
• Keeping the instrument as short as the objectives allow.
• Modest, appropriate incentives where ethics permit.
• Mobile-friendly formatting and a simple, distraction-free layout.
• Sensible timing (avoid exam weeks or holidays for student samples).

Throughout, observe research ethics: informed consent, voluntary participation, confidentiality or anonymity, and — in a university — ethics-committee approval before you collect anything.

Step 7 — Clean and analyse the data

Before analysis, clean the dataset: remove duplicates, drop speeders and straight-liners, handle missing values, recode reverse-scored items, and check that variables fall within valid ranges. Then analyse in line with the plan from Step 1: descriptive statistics (frequencies, means, standard deviations, cross-tabulations) to summarise, and inferential statistics (t-tests, chi-square, ANOVA, correlation, regression) to test relationships and differences. Code and thematically summarise any open-ended responses. If the statistics are beyond your comfort zone, our statistical analysis service can help you specify and run the right tests.

Step 8 — Report

Write up transparently so others could replicate your work. Report the population and sampling method, the achieved sample and response rate, the instrument and its reliability/validity evidence, the analysis methods, and the findings — using tables and charts for clarity. State limitations honestly (coverage, non-response, self-report) and connect results back to your research questions and the wider literature.

Reliability, validity and common survey errors

Good surveys are judged on two pillars. Reliability is consistency — would the instrument yield the same results on repeated use? Multi-item scales should show acceptable internal consistency, and questions should be interpreted the same way by all respondents. Validity is accuracy — does the instrument measure what it claims to? Content validity comes from grounding items in theory and the literature; construct validity from items behaving as the underlying concept predicts. Our guide to reliability and validity explains how to evidence both.

Several recurrent errors threaten survey quality — anticipate and design against each:

• Sampling bias. When the sample is not representative of the population — often from convenience sampling or a frame that omits part of the population — findings cannot be safely generalised. Use probability sampling and a complete frame where possible.
• Non-response bias. When those who do not respond differ systematically from those who do. Minimise with reminders, brevity and incentives; assess by comparing responders’ demographics to the population.
• Social-desirability bias. When respondents answer to look good rather than truthfully, especially on sensitive topics. Reduce with anonymity, neutral wording and self-administered (online/postal) modes.
• Acquiescence and straight-lining. A tendency to agree or to tick the same column throughout. Counter with a mix of positively and negatively worded items and attention checks.
• Question-order and framing effects. Earlier items can prime later answers; small wording changes shift responses. Pilot alternative orders and keep wording neutral.

Address these systematically and your survey will produce data you can defend in a viva. The survey is only one strand of a robust methodology — see how it fits the wider design in our overview of data-collection methods.

Related methodology guides

• Pilot Study
• Mixed Methods Research

Frequently Asked Questions