Does fixed sample mean fixed sample size?

Not in this survey sense; it refers to retaining the same sampled units.

Is a panel automatically representative?

No. Selection, coverage, attrition and changes in the target population still matter.

Can missing members simply be replaced?

Replacement may support another objective, but the new members do not supply the missing individuals' longitudinal histories.

AI Glossary / Statistics and probability / Intermediate

Fixed sample

Definition

A fixed sample, in the repeated-survey sense used here, consists of the same selected units observed on successive occasions. The units might be people, households, establishments or another clearly defined population element. Keeping their identities consistent makes it possible to examine changes within units. This meaning differs from a fixed sample size, which concerns how many observations are collected before analysis. A survey can maintain the same number of respondents while replacing every respondent, so an unchanged count alone does not establish a fixed sample.

What it means

Repeated observations can answer questions that separate cross-sections cannot answer directly. For example, the proportion of firms using an AI tool may increase while some individual firms stop using it and others begin. Following the same firms reveals these transitions, whereas two overall percentages do not identify who changed. The resulting panel remains tied to its selection rules and target population. An original sample does not automatically remain representative indefinitely as the population changes, respondents leave or new eligible units appear.

How it works

Start with a defined population, sampling frame and selection procedure. Assign stable identifiers and return to the selected units at planned intervals using comparable measurements. Store the time of each observation and distinguish genuine changes from altered questionnaires, definitions or reporting units. Analyze repeated records with methods that account for their dependence, rather than treating every wave as unrelated data. Record missing responses and reasons for loss where feasible. Sampling weights, nonresponse adjustments or additional samples may be needed for particular estimands. A refreshment or rotating component changes the design and should be documented rather than silently presented as an unchanged panel.

Three practical examples

Customer research: follow a selected set of customers to describe changes in their reported experience across several survey waves.
Establishment statistics: revisit sampled businesses to study within-business changes in activity while monitoring closures and nonresponse.
Learning research: measure the same participants before and after a training period to describe individual progress, without assuming that observation alone establishes a treatment effect.

Think of it this way

Think of photographing the same garden plots every month. Matching each plot across photographs reveals its development. Photographing four different plots each month still produces four photographs, but it answers a different question about change.

Worked example

Consider four hypothetical units with initial values [10, 20, 30, 40] and follow-up values [12, 18, 33, 45]. Their individual changes are [2, -2, 3, 5], giving a mean change of two. The initial mean is 25 and the follow-up mean is 27, which confirms the same difference when every unit is observed. Now suppose the fourth unit misses follow-up. The observed follow-up mean becomes 21. Comparing that with the original mean of 25 produces an apparent decrease of four, partly reflecting the changed composition. Among the three matched respondents, the initial mean is 20 and the follow-up mean is 21, so their mean change is one. That matched result describes the retained units; it does not recover the missing unit's change or establish the population change. The example illustrates why both identity matching and missingness matter.

Illustrative case study

Fictional case: Birch Enterprise Observatory recruits a panel of firms for a quarterly technology-use survey. Stable identifiers allow its analysts to separate continued use, adoption and discontinuation. When several firms stop responding, the team reports retention and compares available characteristics of respondents and nonrespondents. It considers an additional sample for newly formed businesses and clearly labels its role. The invented project does not claim that maintaining a panel guarantees representative estimates; its lesson is to align the analysis with the units and periods actually observed.

Common misconceptions and mistakes

Do not interpret fixed as an assurance that participation or measurements never change. Do not confuse repeated observations of the same units with repeated independent samples of equal size. Complete-case analysis can be biased when retention relates to the outcome. Repeated questioning may itself influence responses. A business merger or household split can complicate unit identity. Observing a before-and-after change does not establish that an intervention caused it. Correlation within units must be considered when estimating uncertainty; adding waves does not create the same information as adding independent units.

Business use

Organizations can use fixed samples to understand trajectories, transitions and persistent differences that aggregate snapshots obscure. Define whether the decision concerns the original cohort or the current population, because those targets can diverge over time. Budget for follow-up, identity resolution and consistent measurement. Report retention alongside results and distinguish observed changes from estimates requiring assumptions about missing data. When new entrants matter, evaluate whether supplementation or a different design is needed. Reliable longitudinal evidence depends on maintaining the meaning of the measurements as well as maintaining contact with respondents.

Check your understanding

A survey interviews 200 different people each month. Is its sample fixed in this sense?

Answer: No. Its size is constant, but its membership changes. Following the same 200 selected people would support within-person comparisons, subject to response and measurement limitations.

Related terms

Longitudinal data
Nonresponse
Sample rotation
Panel study
repeated cross-section
sample attrition
refreshment sample
sampling weights
paired observations

Learn next

Learn how longitudinal and cross-sectional estimands differ, then examine paired analysis and methods for missing repeated observations. Study rotation and refreshment designs to understand how surveys balance continuity, respondent burden and coverage of a changing population.

Sources

OECD, Glossary of Statistical Terms, fixed sample and sample rotation: https://www.oecd.org/content/dam/oecd/en/publications/reports/2008/09/oecd-glossary-of-statistical-terms_g1gh9ad7/9789264055087-en.pdf U.S. Census Bureau, SIPP organizing principles: https://www.census.gov/programs-surveys/sipp/methodology/organizing-principles.html U.S. Census Bureau, SIPP nonsampling error: https://www.census.gov/programs-surveys/sipp/methodology/nonsampling-error.html

Previous: Fixed-point solving Next: Fixed variable