By the Wavefield Research team · Published Oct 14, 2026
A survey methodology is the complete plan behind a survey: who the study describes, how respondents are selected, the channel they answer through, the instrument, the fielding rules, the quality checks, and the analysis plan. Seven decisions, and a study is only as defensible as the weakest of them.
Most writing on survey methodology in research stops at defining the term. This guide specifies it: each of the seven parts, the step order they run in, and a methodology statement you can adapt for your own report.
The question “is a survey a method or a methodology?” ranks on page one of Google because the word gets used three ways. A survey is a research method: collecting data by asking people structured questions. A survey method, in the narrow sense, often means the mode: online, telephone, mail. And survey methodology is the study-level plan that wraps around all of it. In universities it is also the name of the discipline that studies how design choices change results.
The practical reading: methods are ingredients, methodology is the recipe. A questionnaire is not a methodology; neither is a panel, a sampling technique, or a statistics package. The methodology is the document that says how those pieces combine for this study and why. Government statistical agencies formalize exactly this: the U.S. Census Bureau's Statistical Quality Standards specify requirements for each stage, from frame development through release.
The discipline's organizing idea is worth stealing even for a small study: total survey error. Every stage of the plan contributes its own error: coverage error from the frame, sampling error from the draw, nonresponse error from who declines, measurement error from the questions themselves. The total is what reaches your report. Spending the entire budget shrinking one source while ignoring the others is the classic failure: a sample of 5,000 answering a leading question measures the wrong thing very precisely. The seven-part spec below is, in effect, a checklist of error sources.
Treat the methodology as a spec sheet with seven sections. Each answers one question a skeptical reader will ask, and each has a deep guide on this site where the craft details live.
Who the results describe, and what decision they feed
The population definition is the claim your study makes before a single question is asked: U.S. adults, IA-01 registered voters, customers active in the last 90 days. Every later choice gets judged against it, and a methodology that skips it leaves nothing to judge against.
How people get into the study
The sampling section names the frame (the list or source respondents come from) and the selection rule applied to it: random from a roster, quota cells on a panel, every kth customer. It also owns the honest labeling: a quota sample described as random is a methodology error, not a wording slip. Deep dive: all 9 sampling methods.
The channel respondents answer through
Mode (online link, SMS, telephone, face-to-face, mail) shapes who can respond and how they answer. Sensitive questions get franker answers self-administered; older populations are undercovered online. The methodology states the mode and why it fits the population. Deep dive: data collection methods compared.
What is asked, in what order, with what logic
The instrument section covers the questionnaire itself: question wording and types, screening criteria, rotation and randomization, skip logic, and length. It is where most non-sampling error enters, so a serious methodology documents the version fielded, not the draft that went into review. Deep dive: the bad-question catalogue.
When and how data collection actually runs
Fielding covers dates, soft launch, invitations and reminders, and any incentive. Field dates belong in every report: opinion measured the week before a product recall is a different fact from opinion measured the week after. A soft launch of the first few dozen completes is the built-in pilot that catches instrument problems while they are still cheap.
Which responses count
The quality section states the rules before fielding: speeder thresholds, straight-lining, duplicate and device checks, attention checks, and how exclusions are documented. Rules chosen after seeing the data tend to bend toward the answer the analyst wanted, which is why this section gets written first. Deep dive: screening questions guide.
How raw responses become reported numbers
The analysis section names the weighting targets and method (raking to census age, gender, region), the base for each reported percentage, significance testing conventions, and how precision is described. Stating it up front keeps the analysis from quietly bending toward the preferred answer. Deep dive: survey weighting explained.
The spec says what you decided; the process says when each decision gets executed. Same methodology, viewed as a timeline. The order matters, because every step locks in before the next one starts.
Every report needs a methodology paragraph, and there is a professional consensus on what belongs in it: AAPOR's Disclosure Standards list what any published survey result should reveal: who sponsored it, the population, the sampling frame and method, mode, field dates, sample sizes, weighting, and how precision is characterized. Here is that list as a fill-in statement:
Two phrasings in that template do real work. “Quota-balanced on X” is the defensible claim for an unweighted panel study: it says what was controlled without implying weighting that never happened. And the margin-of-error sentence states the convention instead of borrowing probability-sample authority, a distinction the sampling methods guide unpacks in full. A reader who sees both phrasings knows the author understands the machinery; a reader who sees “±3%” on a convenience sample knows the opposite.
Filled in, the statement reads like this: “This survey was conducted October 6 to 10, 2026 among 500 U.S. adults who purchased coffee in the past month, recruited via an online research panel using quota sampling on age, gender, and region. Interviews were completed online and took a median of six minutes. Results are quota-balanced on the named variables and reported unweighted. Twelve responses were excluded under pre-stated speeding and duplication rules. A probability sample of this size would carry a margin of error of approximately ±4.4 points; as a non-probability sample, this figure is reported by convention.” A few sentences cover every disclosure element without overselling anything, and a reader can judge the study from them alone.
Types of survey methodology sort along two axes, and naming your position on each is most of the typing work. The time axis: a cross-sectional survey measures one moment; a tracker repeats the instrument on fresh samples each wave; a longitudinal panel re-interviews the same people. The purpose axis: descriptive methodology estimates how common things are; analytical methodology tests how answers relate across groups, which raises the bar on sample sizes per subgroup and on stating comparisons in advance.
Mode families (online survey methodology, telephone, mail, mixed-mode) are variants of part 3 rather than separate methodologies. The spec structure is identical in every case; only the entries change. That is the advantage of thinking in the seven-part frame: a political tracker, a customer satisfaction program, and an academic study differ in content, not in kind.
Tracking methodology adds one discipline the single-shot study never faces: between waves, the instrument must stay frozen. Reworded questions, reordered options, or a mid-stream mode switch each break comparability, and the trend line quietly starts measuring the methodology change instead of the world. The working rule: changes happen at wave boundaries, get documented in the methodology statement, and any revised question trends only from its revision forward. Sampling and quotas, by contrast, should be re-run identically each wave, same cells and same source, so that movement between waves comes from real change in the population rather than from a change in sourcing.
A survey is a method: a way of collecting data by asking people structured questions. The survey methodology is the full plan wrapped around that method: population, sampling, mode, instrument, fielding, quality rules, and analysis. The word also names an academic discipline that studies how survey design choices affect accuracy, which is why the term feels overloaded.
A poll is a short survey, usually a handful of questions on one topic, fielded fast and reported as toplines, like election polls. A survey can be any length and typically supports deeper analysis: crosstabs, drivers, segments. The methodology requirements are identical; a two-question poll with a stated sampling method beats a 40-question survey without one.
Define the decision and population; design the sample; draft the instrument; pilot with a soft launch; field to plan; clean and weight; analyze and report with full disclosure. The spec (what you decided) and the process (when you did it) are two views of the same methodology.
Who was surveyed (population and achieved sample size), how they were selected (frame and sampling method, named honestly), the mode, the field dates, response or completion information, quality exclusions, the weighting approach, and how precision is described. That list tracks the professional disclosure standards. If a report omits most of it, treat its numbers accordingly.
Descriptive research (how many, how satisfied, who prefers what), analytical research (how answers relate across groups), and tracking research (how measures move over waves). Surveys support qualitative goals only partly: open-ended questions collect qualitative material, but depth interviews and focus groups remain separate methods.
The phrase usually means sampling for a cross-sectional survey: one snapshot of a population at a single point in time, as opposed to following the same people longitudinally. The sampling method itself is whatever the methodology specifies; cross-sectional describes the time design, not the selection rule.
Quota cells, soft launch, quality flags, and raking weights are built into Wavefield. Describe your study and the AI drafts the questionnaire and the plan around it, ready for you to review.
Start a study