Must the same respondents participate every time?

No; the OECD definition explicitly allows different sample units.

Can a repeated cross-sectional survey measure individual transitions?

Not directly, because it does not link the same individuals across waves.

Does repeating the exact questionnaire guarantee comparability?

No; population coverage, mode, timing, response patterns and interpretation can still change. Comparable wording is helpful but must be supported by the broader survey design and documentation.

AI Glossary / Statistics and probability / Foundation

Repeated survey

Definition

A repeated survey is a sample survey conducted on more than one occasion using essentially the same questionnaire or schedule. The sampled units need not be the same each time. It can therefore include repeated cross-sectional surveys, panel surveys or designs that retain some units while replacing others. Its central purpose is to collect comparable information across occasions, while recognizing that changes in sampling, population composition or measurement can affect the apparent trend.

What it means

Repeating a survey makes it possible to monitor population conditions over time rather than relying on a single snapshot. If each wave samples different people, it can estimate changes in population summaries but does not directly reveal each person's change. A panel follows the same units and can study individual transitions, although attrition introduces its own difficulties. A rotating design combines retained and new units. The phrase repeated survey alone does not identify which of these designs was used.

How it works

Define the target population, reference period, questionnaire and sampling plan for each wave. Preserve comparable concepts and wording where possible, while documenting necessary updates. Calculate estimates with the appropriate weights and design-based uncertainty. To compare waves, account for overlap and covariance when the same units contribute to both estimates. Review changes in response rates, collection mode, coverage and population composition. If a measurement change is unavoidable, a bridge study or other explicit adjustment may help distinguish it from substantive change.

Three practical examples

Example 1: A business association surveys a fresh sample of firms each quarter about AI adoption. Example 2: A household panel repeatedly asks the same participants about digital services. Example 3: A customer survey retains part of its previous sample and recruits additional respondents at each wave. All are repeated surveys when the core measurement remains comparable. Only the panel or retained component directly links a particular unit's responses across occasions.

Think of it this way

Think of taking regular photographs of a changing town square. Fresh photographs can reveal how the crowd as a whole changes, even if different people appear each time. Following named people through the photographs answers a different question about individual movement. In either case, changing the camera angle or the definition of who counts as part of the crowd can create an apparent change that needs explanation.

Worked example

Suppose two independent simple random samples from a large stable target population ask the same yes-or-no question. In the first wave, 160 of 400 respondents answer yes, giving 40%. In the second, 225 of 500 answer yes, giving 45%. The estimated increase is five percentage points. Ignoring a negligible finite-population correction, the estimated variance of the difference is 0.40 times 0.60/400 plus 0.45 times 0.55/500 = 0.001095. Its standard error is approximately 0.03309, or 3.309 percentage points. A normal-approximation 95% interval for the change is about -1.49 to 11.49 percentage points. This calculation does not establish an increase clearly separated from zero. If respondents overlapped, weights were complex or nonresponse required adjustment, the variance calculation would need to reflect those features instead of assuming two independent simple samples.

Illustrative case study

In this fictional case study, Orchard Enterprise tracks small-business use of AI assistants every six months. The first two waves use different samples but the same core questions. A later wave changes from telephone interviews to a web form, and reported adoption rises. Before presenting the rise as a genuine trend, the team examines respondent composition and runs a limited comparison of collection modes. Its report separates directly comparable estimates from the wave affected by the method change and explains the remaining uncertainty.

Common misconceptions and mistakes

Common mistakes include calling every repeated survey a panel, interpreting population-level change as individual change, and ignoring dependence between overlapping samples. Another error is changing a question's wording while presenting the resulting series as perfectly comparable. Larger samples reduce some sampling uncertainty but do not automatically repair nonresponse or coverage problems. Calendar timing, seasonal patterns and reference periods also matter. A measured trend is not by itself evidence that a particular policy or AI product caused it.

Business use

Businesses use repeated surveys to monitor adoption, satisfaction, employee experience and market conditions. A stable core questionnaire helps preserve a useful time series, while clearly labeled optional modules can address new issues. Teams should maintain version histories, sampling documentation and reproducible weighting procedures. AI-generated trend summaries should distinguish percentage changes from percentage-point changes and retain uncertainty. When a survey design changes, decision-makers need to know whether the observed movement could partly reflect the new measurement process.

Check your understanding

What is the change from 40% to 45% in the worked example?

Answer: Five percentage points, which is a 12.5% relative increase from the initial estimate.

Why are the two variance contributions added?

Answer: The example assumes independent samples.

What additional information is needed when samples overlap?

Answer: Their covariance or a design-based method that accounts for the overlap.

Can these two samples identify which respondents changed their answer?

Answer: No.

Related terms

Longitudinal study
Related terms include repeated cross-section
panel survey
rotating panel
survey wave
attrition
sampling weight and comparability over time

Learn next

Learn next how panel and repeated cross-sectional designs answer different questions. Study variance estimation for changes between correlated estimates, then examine mode effects and questionnaire testing. Before drawing a trend chart, build a small comparability record describing every wave's population, reference period, collection method and question version.

Sources

Sources: OECD Glossary of Statistical Terms, Repeated survey, https://www.oecd.org/content/dam/oecd/en/publications/reports/2008/09/oecd-glossary-of-statistical-terms_g1gh9ad7/9789264055087-en.pdf; Eurostat, Handbook on the design and implementation of business surveys, https://ec.europa.eu/eurostat/documents/3859598/5825949/CA-09-97-818-EN.PDF.pdf/f5ee3198-6fc0-4672-96a1-8fdb4a81ca93?t=1414780218000. The two-wave calculation and fictional business-adoption case are original illustrations, not reported survey findings.

Previous: Repeated state Next: Repeatedmeasures designs