Survey research design is the plan for a study that collects standardized answers from a sample of people and uses them to describe or explain a larger population. It settles five things before any question is written: what you will estimate, who you will ask, how you will select them, how you will reach them, and how you will analyze what comes back1, 4.
Most survey research is observational: it records what people report without assigning an intervention. It can describe a population or examine relationships, but those results alone do not establish causation. A survey experiment adds random assignment to test a specific change, such as question wording.
What is survey research design? Definition, purpose and when to use it
The standard survey research definition: the use of a standardized questionnaire or interview to collect data about people, meaning what they do, think, know or prefer4. The design is the set of decisions that make those answers usable as evidence. Fowler puts it simply: the goal is to produce statistics about a population by asking a sample, and the design is everything that makes the sample and the questions fit that goal3.
In a methods chapter the same plan appears as survey study design, survey design in research, or survey design methodology; all three mean the plan described here. It helps to keep three words apart. The questionnaire is the instrument. The survey is the whole study around it. Survey methodology is the body of research on how to do surveys well, organized around the sources of error a design has to control: coverage, sampling, nonresponse, measurement and processing1. A good design names each of those risks and does something about it.
The purpose of a survey is to put a number on something only people can tell you, for a group larger than the people you asked. Use a survey design when the thing you need is best known by the people themselves (satisfaction, intentions, opinions, private behavior), when you need a number for a defined group rather than a story from a few people, and when you can reach a sample that resembles that group. Do not use one when the answer is already recorded somewhere (sales, logs, attendance), when you need to prove cause, or when the people who will answer are not the people you need to describe. For those, look at correlational research for association questions and an experiment for causal ones.
Is survey research quantitative or qualitative?
Quantitative, in almost every case. A survey turns answers into counts, percentages, means and correlations, and it is analyzed with statistics. Creswell classes surveys as a quantitative design alongside experiments9. A survey becomes mixed methods when open-ended questions are coded into themes and reported beside the numbers, and it is qualitative only when the whole instrument is open questions analyzed for meaning, which is closer to a structured interview study than to a survey.
The reason the question is asked so often is that Likert items look like opinions rather than measurements. They are still numbers. If you need to describe how many, how much or how often, you are running a quantitative survey; if you want to understand why in people's own words, add two or three open-ended questions and treat their coding as a separate, qualitative step.
Types of survey research design
Survey designs are named on three separate axes, and a methods chapter usually needs one word from each: the purpose (what the numbers are for), the time structure (when and how often you measure), and the mode (how people answer). "A descriptive, cross-sectional, online survey" is a complete design statement.
| Design type | What it answers | Example | Main limit |
|---|---|---|---|
| Descriptive | How many, how much, how often. A level or a share for a defined group. | What share of customers would recommend us this quarter | Says nothing about why |
| Analytical (correlational) | Whether two things go together, and how strongly. | Do employees who rate their manager higher also intend to stay longer | Association is not cause |
| Exploratory | What the issues, categories and vocabulary are, before you can measure them. | Open questions to a small sample ahead of a full study | Not generalizable; leads to a second survey |
| Cross-sectional | A snapshot: one measurement at one point in time. | Annual employee survey | Cannot separate age effects from period effects |
| Repeated cross-sectional (trend) | Change in the population, with a fresh sample each wave. | Monthly brand tracker | Change can be sampling noise unless waves are matched |
| Longitudinal (panel or cohort) | Change within the same people, and what preceded it. | The same new hires at 30, 90 and 180 days | Attrition, conditioning, cost |
| Survey experiment | Whether a wording, message or concept causes a different answer. | Half see price A, half see price B, compare intent | Only the manipulated element is causal |
| Mode: online, phone, mail, in person, mixed | Who you can reach and how they answer. | Email-to-web with a phone follow-up for non-responders | Mode changes answers; keep wording identical across modes |
Cross-sectional
- One wave, one sample, one date range: the cross-sectional survey research design most theses and pulse surveys use.
- Cheapest and fastest; the default for a thesis or a pulse survey.
- Can compare groups (by age, region, plan) but not track individuals.
- Write it as "a cross-sectional survey of N [population] between [dates]".
Longitudinal
- Two or more waves, the same people each time.
- Shows who changed and what came first; the only survey design that gets near sequence.
- Needs identifiers, consent to recontact and a plan for the 20 to 40 percent who drop out between waves.
- If you only need the population trend, a repeated cross-section is simpler.
Two naming notes for the methods chapter. "Descriptive survey research design" and "cross-sectional" are not rivals; most student surveys are both. And a survey that compares groups is still non-experimental unless you assigned people to those groups, so call it analytical or correlational, not quasi-experimental9.
Which survey research design do you need?
Answer three questions and the chooser names the design, explains it in one paragraph, and writes a methods sentence you can paste and complete. It runs in your browser and sends nothing anywhere.
Descriptive cross-sectional online survey
One sample, measured once, reported as levels and shares. It answers how many and how much for the population you sampled from, and nothing about change or cause.
The sentence follows the design vocabulary in Creswell and Groves et al.9, 1 Add the sampling method from step 3 and the sample-size justification from step 4 and the design paragraph is done.
Survey research methods: how to design a survey study in 8 steps
The order matters. Each step narrows the next, and every step you skip comes back as a limitation you have to write up later. Where a step has its own guide on this site, the link goes there; this page keeps the decision, not the detail.
Write the decision and the population in one sentence
"Among active customers in the US in Q2, estimate the share who would recommend us, overall and by plan." That sentence fixes the target population, the sampling frame you will need, the unit of analysis and the estimates you are promising. If you cannot write it, you are not ready to write questions.
Choose the design type
Descriptive or analytical; cross-sectional, repeated or longitudinal; single mode or mixed. Use the chooser above and the types table. A snapshot needs one wave. A trend needs identical wording across waves. A "who changed" question needs the same people, consent to recontact and a retention plan.
Select the sampling approach and name the bias you accept
Probability sampling (every member of the frame has a known, non-zero chance) is what makes a margin of error meaningful. Convenience, volunteer and opt-in samples are fine for product feedback and pulse surveys, but the claim shrinks to the people you reached. The nine methods are compared in sampling in research. Whatever you choose, list the top three ways your sample could differ from the population (coverage, self-selection, nonresponse) and one check for each, such as comparing early with late responders or your sample's mix with a known benchmark1, 5.
Set the sample size, then work back to invites
For a proportion at 95 percent confidence and a 5-point margin, about 385 completes covers any large population; 3 points needs about 1,068. Small populations need fewer: 218 for a population of 500, 278 for 1,000, 370 for 10,000. Subgroups you want to compare each need their own floor, usually 100 or more. Then: invites needed = completes / (eligibility rate x response rate x completion rate). For 300 completes at 90 percent eligible, a 25 percent response rate and 80 percent completion, that is 1,667 invites. The sample size calculator does the arithmetic and the margin of error.
Two figures from our own data help here. Surveys of three or more questions complete at about 81 percent once someone has started, so 0.8 is a fair default for the last term. And a survey collects 62.8 percent of everything it will ever get on day one and 86 percent within a week, so plan the field period around the send, not the calendar.
Choose the survey method (mode) and the contact plan
The survey method, or mode, decides who you can reach, how they answer and what it costs. The main survey methods are online, phone, mail, in person and mixed mode. Online is fast and cheap and handles branching; phone and in-person reach groups without email and let an interviewer clarify, at interviewer-effect and cost; mail reaches low-digital groups slowly; mixed mode raises coverage but shifts answers unless wording and layout stay identical2. Plan the contacts as part of the design: Dillman's tailored design uses several timed contacts (pre-notice, invitation, reminder, final), each written differently2. On SuperSurvey 73.9 percent of all responses arrive on a phone, so an online design is a mobile design whether you planned it or not.
Build the instrument to match the analysis
Define each construct in plain words, then choose the format the analysis needs: categories for a percentage, a rating or agreement scale for a mean or top-box, one idea per item, easy and relevant questions first, sensitive items late8, 7. Where a validated instrument exists for your construct, use it and cite it. The writing is its own job: see questionnaire design for the seven-step procedure, how to write survey questions for wording, and Likert scales versus rating scales for the choice between them. Two structural decisions are worth making here because the benchmark below prices them: make questions required (it costs 0.4 points of completion and returns 13 percent more answers), and never end on an email address or an open text box.
Pretest, pilot, then launch
Pretest with a handful of people from the target group, thinking aloud, to catch wording and logic problems; then pilot with a small slice of the real list to check completion time, drop-off points and whether screeners and quotas behave3, 5. Build the pilot in a free online survey maker so it runs on the same skip logic, required settings and phone layout as the real send. Fix the invitation before the reminder: state the purpose, the time it takes, the closing date, and only the privacy promise you can keep. Anonymous means you collect no identifiers; confidential means you hold them and restrict access. See survey privacy for the controls, and if the topic could put anyone at risk, report only aggregates with small cells suppressed.
Plan the analysis and the quality rules before fieldwork
Write the cleaning rules first (duplicates, impossible completion times against the pilot median, straightlining on grids), the primary outcomes, the subgroup cuts with their minimum sizes, the weighting variables if any, and the significance rule you will apply1. Report the response rate using AAPOR's standard definitions so the number means the same thing to a reader elsewhere6. Then the report separates what you found from how you found it: population, frame, mode, field dates, completes, weighting, limitations. Statistical significance and response bias cover the two interpretation traps; quantitative data analysis covers the rest.
Steps 7 and 8 are where most survey designs fail in practice, not in the questionnaire. A pretest costs a day; a cleaning rule invented after you have seen the results costs your credibility.
What 2.27 million responses say about survey design
To examine how design choices relate to completion, we analyzed records from the SuperSurvey survey maker between September 2023 and August 2026: where people stopped, what they skipped, and how long different question types took. These are observational comparisons; they can inform a pilot but do not prove that changing one setting will produce the same result in every survey.
1. People do not quit in the middle
Each line follows 100 people who opened a survey of that length; the height at question 7 is the share still answering by then. Surveys are averaged with equal weight, so no single large survey carries the curve.
Question one costs six to ten points. Question two costs four, question three two and a half, and after that the line is almost flat for as long as the survey runs. Then the final question takes another eight to eleven points in a single step. Everything in between costs about one point a question, whether the survey has five questions or thirty.
A response is only recorded once someone interacts, so people who opened the survey and closed it at once are not in the denominator; the real loss at the door is larger. 6.7 percent of starters never answer anything. The flattening after question three holds in every length band.
2. The last question is skipped, not abandoned
What you end on changes how much of the whole survey you get back. Measured inside matching question-count bands against surveys that end on a multiple choice:
Ending on a rating scale returns 4.7 points more of your survey than ending on a multiple choice. Ending on an email address returns 15.2 points less. That is a nineteen-point spread from one decision that takes two seconds to change.
n = 106 surveys ending on a rating scale and 90 ending on an email, against 2,504 ending on a multiple choice. The email row is the thinnest number in this section; treat its size as approximate and its direction as solid.
3. Required questions cost almost nothing, with one exception
The most common piece of survey advice is to keep questions optional so people do not walk away. Matched inside question-count bands, surveys with every question required complete at 81.2 percent (899 surveys) against 80.8 percent with every question optional (4,365 surveys). That +0.4 points is a null. Requiring answers returns 13 percent more answers per person who starts: 88 percent of questions answered against 78 percent. Yet 69.5 percent of survey creators make everything optional.
4. Every question type, priced
How often each type is answered once a respondent reaches it, and how long the median respondent spends on it. These rest on millions of asks, so they are not affected by which audience a type is typically sent to.
| Question type | Answered when reached | Median time | Times asked |
|---|---|---|---|
| Rating scale (1 to 10) | 96.2% | 5.0s | 159,545 |
| Net Promoter Score | 95.0% | 4.0s | 3,013 |
| Ranking | 95.2% | 25.0s | 25,878 |
| Star rating | 93.7% | 4.6s | 57,468 |
| Drop-down list | 91.7% | 6.3s | 106,256 |
| Checkboxes | 88.9% | 8.1s | 624,893 |
| Multiple choice | 86.0% | 4.9s | 2,301,606 |
| Short text box | 79.4% | 10.6s | 379,279 |
| Matrix / grid | 77.4% | 19.9s | 23,579 |
| Long text box | 69.7% | 16.2s | 267,430 |
| Email address | 62.2% | 8.4s | 21,475 |
A rating scale is answered by 96.2 percent of the people who see it, in five seconds. A ranking question is answered nearly as often but takes 25 seconds, and surveys containing one finish 7.2 points worse. Long text is the other tax: 69.7 percent answered and 5.8 points off completion for the survey that carries it. This is the empirical case behind the textbook advice to use closed items for measurement and open items sparingly.
Only 163 surveys with a ranking question cleared the 20-response floor, so treat the 7.2-point figure as a rough size. Question type also partly stands in for who the survey was sent to: ranking questions live in market research aimed at strangers, rating scales in feedback forms aimed at customers.
5. Length costs less than you have been told
Blue bars are completion. The orange line is what you take home: answers collected per person who starts.
Going from five questions to twenty costs about 8.4 points of completion and returns 3.5 times the data. There is no length at which the total stops rising. The median survey has 5 questions and is over in 71 seconds; even a survey of 20 to 29 questions is done in under 4 minutes at the median. Below ten questions, phones complete 83.7 percent against 83.1 percent on desktop; above fifteen questions the gap opens to two to five points. Length is a mobile problem before it is a length problem.
Long surveys are disproportionately run by organizations with a captive audience such as an employer, a university or a school. A thirty-question survey sent to strangers should expect worse than this. What belongs to the survey rather than its audience is the shape: a steady cost of about one point per question through the middle, in every length band.
6. Responses arrive in a burst, and half of surveys never get one
Cumulative share of a survey's responses, counted from the day its first response arrived, for 4,718 surveys with at least 20 responses and a full 90 days in the field. The median survey has 75.8 percent of everything it will ever collect within three days and 86 percent within a week; leaving it open for a month adds 9.7 points. A survey is a launch, not a slow burn. And 52.0 percent of all surveys built never receive a single response, which makes the send, not the build, the hard part of a survey design.
This is how people share surveys, in one burst of links, not a law of nature. A survey embedded on a busy page keeps collecting for months. "Never received a response" includes drafts and practice surveys; it is not a failure rate.
What made no difference
Measured the same way, matched inside question-count bands: a custom theme and brand colors, minus 1.1 points; auto-advancing to the next question, minus 0.6; randomizing question order, plus 1.2 on only 123 surveys; a progress bar, plus 2.7, the largest cosmetic lever and still smaller than every structural one above. The levers you can see move completion by nought to three points. The ones you cannot see, what you end on, what you require and which question type you reach for, move it by five to forty-four.
How this was measured, and what it does not measure
The corpus
Every survey created on SuperSurvey between 16 September 2023 and 25 August 2026, read from production records rather than sampled: 122,887 surveys built by 58,621 creators, of which 58,938 collected at least one response, for 2,274,520 responses in total.
What counts as a real survey
Everything above is measured on surveys with at least three questions and at least three responses: 25,112 surveys carrying 1,180,070 started sessions and 998,864 completions. Any per-survey rate additionally requires 20 started sessions, leaving 6,878 surveys. The question-by-question reconstruction reads 558,927 individual response records across 15,181 surveys. 12.8 percent of the surveys were built on a paid plan and 85 percent by someone with an account; restricting to account holders moves nothing on this page.
How the rates are computed
Every rate is a per-survey average, unweighted: each survey counts once no matter how many responses it collected, because two surveys in this corpus collected over 100,000 responses each and a response-weighted average would be those two customers' numbers. Feature comparisons are length-adjusted: computed inside question-count bands (1 to 4, 5 to 7, 8 to 10, 11 to 14, 15 to 19, 20 to 29, 30+) and re-weighted by the feature side's band mix, so a difference can never be an artefact of one side being shorter. Drop-off is reconstructed from each individual response record, not from a stored counter. Times per question come from per-page timing on the response; the median is reported because the mean is dominated by people who leave a tab open.
What this does not measure
- Invitation response rate. A record exists only once someone opens the survey and interacts with it, so every drop-off number here is conservative.
- Answer quality. We can see that someone typed 14 words; we cannot see whether they meant them.
- Causation. This is observational. Where a result could plausibly be the audience rather than the design, it is flagged in place.
- Your survey. These are averages over tens of thousands of surveys sent to very different people: defaults, not a forecast.
Corpus closed 25 August 2026. Figures generated 2 September 2026.
Advantages and disadvantages of survey research
What a survey design does well
- Reaches large samples cheaply, especially online, so estimates come with a usable margin of error.
- Standardized questions make answers comparable across people, groups and waves.
- Measures things no log records: attitudes, intentions, satisfaction, private behavior.
- Flexible: descriptive, correlational and trend questions from one instrument.
- Fast. Most of the responses you will get arrive within three days of the send.
Where it fails
- Self-report. People misremember, round up, and answer to look good; wording and order shift answers8.
- Nonresponse. The people who answer differ from those who do not, and a low response rate does not prove bias any more than a high one rules it out1.
- Coverage. Your list is not the population; anyone not on it has no chance of being asked.
- Cause. A survey shows association; only a survey experiment with random assignment shows cause, and only for the element you manipulated.
- Depth. Closed items cannot follow up. Open items can, at a 30-point cost in answer rate.
Examples of survey research design
Three survey design examples written the way they would appear in a methods section, each with the decision it serves. The first is worked through the whole worksheet below.
Subscription churn (practitioner, descriptive plus analytical, cross-sectional)
Decision: which churn drivers to fix first next quarter. Design: a cross-sectional survey of active subscribers in the US and Canada, email-to-web, mobile-first, two reminders, with quotas by plan tier and region so each tier reaches at least 200 completes. Primary measures: renewal intent (0 to 10), feature and support satisfaction (1 to 5), reasons for dissatisfaction (multi-select), one open improvement prompt placed mid-survey rather than last. Analysis: top-box by tier, drivers of renewal intent by correlation, early versus late responder comparison as the bias check.
Undergraduate thesis (student, descriptive, cross-sectional)
Decision: describe study habits and their relation to self-reported grades among third-year students at one university. Design: "This study used a descriptive, cross-sectional survey design. A convenience sample of 312 third-year students completed an online questionnaire between 3 and 17 March. Because the sample was not drawn at random from the student register, results describe the respondents rather than the year group, and no margin of error is reported." That last sentence is the one supervisors look for.
Employee pulse (HR, repeated cross-sectional trend)
Decision: whether the manager-training program moved team climate. Design: the same eight items to all staff every quarter, anonymous, fresh sample each wave by design (whoever answers), reported at department level only where at least five people responded. Wording frozen after wave one so movement means change, not rewording. An analytical add-on compares departments whose managers completed the training with those whose did not, reported as an association because managers were not assigned at random.
| Decision | Which churn drivers should we fix first in the next quarter? |
|---|---|
| Target population and frame | Active subscribers in the last 60 days, US and Canada; frame = billing email list |
| Design type | Descriptive plus analytical, cross-sectional, repeated quarterly as a trend |
| Sampling and quotas | Census of the frame with quotas by plan tier and region; minimum 200 completes per tier |
| Sample size and invites | 600 completes; at 95% eligible, 25% response and 80% completion that is about 3,158 invites |
| Mode and contacts | Email-to-web; invitation, reminder at day 3, final at day 7; closes day 10 |
| Measures | Renewal intent 0 to 10; feature and support satisfaction 1 to 5; reasons multi-select; one open prompt mid-survey; tier and tenure from the account, not asked |
| Quality rules | Pilot to 2% of the list; remove duplicates and completions under a third of the pilot median; all items required except the open prompt |
| Bias checks | Sample mix against the customer base by tier and region; early versus late responders; results with and without weights |
| Reporting | One-page summary plus a methods box: frame, field dates, completes, AAPOR response rate, weighting, limitations |
To build the questionnaire itself, start from a survey template for the nearest use case and change the measures to your constructs; a tailored instrument beats a borrowed one, but a borrowed one beats a blank page.
Frequently asked questions
What type of research design is a survey?
A survey is a non-experimental quantitative design. Depending on purpose it is classed as descriptive (levels and shares) or correlational (associations between measures), and depending on timing as cross-sectional or longitudinal. It becomes experimental only when you randomly assign respondents to different versions of a question or stimulus, and then only that comparison is causal.
What is a descriptive survey research design?
A design whose purpose is to describe a population on the measured variables: how many, how much, how often, and how that differs across groups. It makes no claim about why. Most one-off questionnaires, and most student surveys, are descriptive and cross-sectional at the same time, and the methods chapter should say both words.
What is the difference between cross-sectional and longitudinal survey design?
Cross-sectional measures once. Longitudinal measures the same people two or more times, which is the only way to see who changed and in what order. A repeated cross-sectional (trend) design sits between them: it measures more than once but with a fresh sample each wave, so it tracks the population, not individuals. If you only need the trend, choose that; it avoids attrition and recontact consent.
How many respondents does a survey need?
For a share reported with a 5-point margin at 95 percent confidence, about 385 completes for any large population, and fewer for small ones (218 for a population of 500). If you will compare subgroups, each subgroup needs its own floor, usually 100 or more. For a convenience sample there is no margin of error to defend, so size it by the smallest group you need to describe and say so. The sample size calculator on this site does the arithmetic.
What is survey research design in psychology and in market research?
The same design, different vocabulary and different frames. In psychology a survey usually measures constructs with validated multi-item scales, reports reliability, and is called a correlational design when it relates constructs to each other. In market research it measures awareness, preference and intent for a brand or product, samples from customer lists or opt-in panels, and reports shares and top-box scores, often as a repeated tracker. Both are non-experimental unless a stimulus is randomized.
How do I write the survey research design section of a methods chapter?
Four sentences, in this order: the design (purpose plus time structure plus mode), the population and sampling method, the sample size with its justification, and the instrument with its source or development process. Then the field dates, the response rate by a stated definition, and the limitations that follow from the sampling method. The design chooser on this page writes the first sentence; the sampling and sample-size guides supply the second and third.
What is the difference between survey design and a questionnaire?
The questionnaire is one component: the items and their answer options. Survey design is the whole plan around it: population, sample, mode, timing, measures, analysis and reporting. A good questionnaire inside a poor design still produces numbers that describe the wrong people.
Does a survey need an attention check?
Not by default. For open links with incentives or paid panels, one attention check helps flag low-effort responses. For a trusted internal list it mostly annoys people and creates exclusions you then have to justify. If you use one, write the rule (what fails, how many failures are allowed) before fieldwork, and report how many records it removed.
How this guide was put together, and what it does not cover
The design vocabulary and the eight steps follow the survey methodology literature, mainly Groves and colleagues' total survey error framework, Dillman's tailored design method, Fowler's Survey Research Methods, and AAPOR's best-practice and standard-definitions documents, each cited where used. Nothing is drawn from a vendor's framework. The sample-size figures are the standard proportion formula at 95 percent confidence with the finite population correction applied; the calculator page shows the arithmetic.
The benchmark section is our own analysis of SuperSurvey production data, described in full in the method note above. Its limits are stated there: it cannot see invitation response rates or answer quality, it is observational, and it averages over very different audiences. Where a competitor's figure could not be traced to a primary source it was left out; in particular we do not repeat the commonly quoted vendor statistics on survey accuracy or trust.
This page covers the design. It does not cover how to select respondents, how many to select, or how to write the items; those are separate jobs covered in sampling, sample size and questionnaire design.
Disclosure. SuperSurvey makes survey software, so we have a commercial interest in people running surveys. Nothing on this page is for sale or gated, and none of the advice depends on using our tool. The design chooser runs in your browser and sends nothing anywhere. The benchmark exists because we are the only ones who can see this data, and publishing it, including the findings that cut against common advice, seemed more useful than keeping it.
References
- Groves, R. M., Fowler, F. J., Couper, M. P., Lepkowski, J. M., Singer, E., & Tourangeau, R. (2009). Survey Methodology (2nd ed.). Wiley. wiley.com
- Dillman, D. A., Smyth, J. D., & Christian, L. M. (2014). Internet, Phone, Mail, and Mixed-Mode Surveys: The Tailored Design Method (4th ed.). Wiley. wiley.com
- Fowler, F. J., Jr. (2014). Survey Research Methods (5th ed.). SAGE. methods.sagepub.com
- Ponto, J. (2015). Understanding and evaluating survey research. Journal of the Advanced Practitioner in Oncology, 6(2), 168-171. pmc.ncbi.nlm.nih.gov
- American Association for Public Opinion Research. Best Practices for Survey Research. aapor.org
- American Association for Public Opinion Research (2023). Standard Definitions: Final Dispositions of Case Codes and Outcome Rates for Surveys (10th ed.). aapor.org
- Pew Research Center. Writing Survey Questions. pewresearch.org
- Krosnick, J. A., & Presser, S. (2010). Question and questionnaire design. In P. V. Marsden & J. D. Wright (Eds.), Handbook of Survey Research (2nd ed.). Emerald.
- Creswell, J. W., & Creswell, J. D. (2018). Research Design: Qualitative, Quantitative, and Mixed Methods Approaches (5th ed.). SAGE.
- SuperSurvey (2026). The SuperSurvey Response Benchmark: 2,274,520 responses to 25,112 surveys, 16 September 2023 to 25 August 2026. Method note on this page.