Must every survey use every master-sample unit?

No. Survey-specific subsamples are a common arrangement.

Does reuse mean interviewing the same people each time?

Not necessarily. A design may reuse geographic areas while selecting different households or individuals.

Can a master sample serve an unexpected new survey?

Often, but only if its coverage and design can support the new population and precision requirements.

AI Glossary / Statistics and probability / Intermediate

Master sample

Definition

A master sample is a sample selected for use on multiple future survey occasions, rather than drawing a completely new sample independently for every inquiry. It may be a large sample from which smaller survey-specific samples are selected. In household survey programs, it often consists of geographic primary sampling units reused across different surveys or rounds. Its defining feature is planned reuse, not simply being a large, important, or centrally stored dataset.

What it means

Selecting and preparing survey locations can be expensive. A master sample allows several studies to share parts of that work, while drawing the final participants needed for each study. Different inquiries can therefore use the same sampled areas without necessarily interviewing exactly the same households.

A master sampling frame is the organized source of eligible units from which samples can be drawn. A master sample is the selected subset intended for reuse. A panel follows the same units over time; a master sample can support a panel, repeated cross-sectional surveys, or several different inquiries, depending on how subsampling is arranged.

How it works

Define the target population and the needs of the anticipated survey program. Build or select an appropriate frame, choose the master sample design, and record selection probabilities. For area-based designs, prepare the maps and lower-stage lists required within selected locations.

For each inquiry, select its sample or subsample using a documented rule. Calculate overall inclusion probabilities through the relevant stages, then construct weights and uncertainty estimates that reflect the actual design. Coordinate overlap when useful, but distinguish purposeful overlap from accidental duplication.

Maintain coverage as populations change. New buildings, businesses, households, or geographic boundaries can make old lists incomplete. Plan refreshing, rotation, and replacement carefully. A shared design must balance several studies' precision needs, and a master sample suitable for common groups may be inadequate for a rare subgroup that later becomes important.

Three practical examples

A statistical office reuses selected geographic areas for separate employment, spending, and technology-use surveys.
A business research program selects a probability sample of establishments and draws different subsamples for several planned inquiries.
An AI adoption study uses a maintained master sample to launch periodic surveys while tracking selection probabilities and respondent burden.

Think of it this way

Think of setting up a network of research bases before several expeditions. Each expedition can use some of the same bases while visiting different destinations nearby. Sharing the preparation saves effort, but the original placement of the bases constrains which later destinations can be reached well. Maintenance and planning still matter.

Worked example

Consider an original simplified design with 1,000 establishments on a complete frame. Select 200 by simple random sampling without replacement as a master sample. For survey A, select 50 establishments by simple random sampling from those 200.

An establishment's master-sample inclusion probability is 200/1,000 = 0.20. Conditional on being in the master sample, its survey-A inclusion probability is 50/200 = 0.25. Its overall probability is 0.20 × 0.25 = 0.05, so its base weight is 1/0.05 = 20.

If all 50 respond and 18 report using a specified AI tool, the weighted total estimate is 18 × 20 = 360 establishments, and the estimated proportion is 360/1,000 = 36%. These figures describe the estimator, not a guarantee about the population.

Suppose survey B independently selects 80 from the same realized master sample. Conditional on that master sample, the expected overlap with A is 50 × 80/200 = 20 establishments. The two surveys share an initial selection stage, so their samples are not independent selections from the full population. Nonresponse, unequal selection, or clustering would require additional treatment beyond this simple illustration.

Illustrative case study

Fictional case: EnterpriseWillow plans annual surveys on digital tools, skills, and operating practices. It creates a master sample of establishments to share recruitment preparation and coordinate contact schedules.

A later inquiry focuses on a rare type of specialist firm poorly represented in the original selection. Rather than claiming the existing sample can support precise estimates for every subgroup, the team develops a supplementary design and documents how it connects to the main survey. It also rotates contacts to manage burden. The case illustrates the efficiency of reuse and the need to revisit coverage when research objectives change.

Common misconceptions and mistakes

A master sample is not the entire population frame. It is also not automatically a longitudinal panel, because different final units can be selected on different occasions. Reusing a convenient volunteer list does not create known probability-sampling properties merely by changing its name.

Survey-specific weights must account for selection through the master sample. Using only the final subsampling fraction can seriously misstate population totals. Repeated use can create dependence between survey estimates and increase respondent burden. Finally, an old sample does not remain representative indefinitely without attention to population change, attrition, and frame maintenance.

Business use

Businesses and research organizations can use master samples to support a program of recurring studies with consistent definitions and coordinated fieldwork. The approach is most attractive when several surveys share compatible population coverage and design requirements.

Keep the master selection records separate from each inquiry's subsampling and response records. Budget for maintenance, monitor coverage, and preserve identifiers needed to assess overlap. When reporting changes across surveys, use methods that recognize the shared design where relevant. Operational convenience should be evaluated alongside precision, flexibility, and the experience of repeatedly contacted respondents.

Check your understanding

A unit has a 0.20 chance of entering a master sample and a 0.25 conditional chance of selection for a survey. Is its overall base weight four?

Answer: No. Four reflects only the subsampling stage. The overall inclusion probability is 0.05, giving a base weight of 20 before any further adjustments.

Related terms

Multistage Sampling
Master sampling frame
sampling frame
subsampling
inclusion probability
survey weight
panel
primary sampling unit

Learn next

Study how inclusion probabilities combine across sampling stages, then examine sample rotation, overlap, nonresponse adjustments, and variance estimation for coordinated survey programs.

Sources

OECD Glossary of Statistical Terms, master sample: https://www.oecd.org/content/dam/oecd/en/publications/reports/2008/09/oecd-glossary-of-statistical-terms_g1gh9ad7/9789264055087-en.pdf

Pettersson, United Nations household survey handbook, chapter V: https://unstats.un.org/unsd/HHsurveys/pdf/Chapter_5.pdf

The establishment example is original and its probabilities, weight, total, and expected overlap were checked. EnterpriseWillow is fictional.

Previous: Master data Next: Match rating approach