Must an old unit disappear before a new one can be added?

No; augmentation can broaden coverage without a missing unit.

Does adding more respondents always reduce bias?

No; the selection and coverage of the added units matter.

Is it the same as sample rotation?

No; rotation deliberately replaces units after a planned participation period.

AI Glossary / Statistics and probability / Intermediate

Sample augmentation

Definition

Sample augmentation, in survey statistics, means maintaining and adding units to an existing sample so that it continues to cover the evolving population appropriately. In establishment and price surveys, this can involve recruiting new businesses or outlets and adding relevant products or varieties. The purpose is to address sample depletion and changing coverage, not simply to increase the number of records.

What it means

A fixed panel gradually becomes less useful when businesses close, participants stop responding, or new types of activity appear outside its coverage. Adding suitable units can help keep the evidence connected to the population the survey intends to describe.
This entry follows the survey meaning in the OECD source list. In machine learning, augmentation often means transforming or synthesizing training examples. That separate practice does not recruit real survey units or automatically repair the coverage problems addressed here.

How it works

Review the target population and the current sampling frame. Determine whether the sample is losing units, missing newly created establishments, or underrepresenting growing segments. A stable headcount can still conceal outdated population coverage.
Distinguish reasons for loss. A business that has closed may no longer belong to the target population, whereas an eligible business that stops responding creates a nonresponse issue. These situations should not be treated as identical administrative events.
Decide what to add and how to select it. Use a documented design that accounts for relevant strata, size measures, and inclusion probabilities where applicable. Recruiting only the easiest available replacements can introduce further selection problems.
Coordinate the existing and additional samples. Check overlaps and duplicates, retain selection histories, and establish which records represent the same underlying unit. Combining lists without accounting for multiple selection routes can distort inclusion probabilities and weights.
Update estimation procedures to reflect the combined design and current population information. Old weights should not simply be copied to new units because their selection mechanisms may differ. Nonresponse adjustment and calibration have their own assumptions.
For price surveys, adding a new product or outlet also raises continuity questions. A higher-priced new variety is not automatically evidence of inflation relative to an old variety of different quality. Appropriate linking, comparability, and index methods are needed.
Evaluate the effect on coverage and series continuity. Compare the characteristics of retained and added units, examine contributions to estimates, and document whether a change reflects population movement or the sample update itself.
Record additions, losses, dates, reasons, selection procedures, and any revisions to weights or methods. Augmentation improves the opportunity for current coverage, but representativeness is an objective to assess, not a guarantee conferred by the act of adding units.

Three practical examples

Business panels: a survey recruits newly established firms that an older panel could not include.
Consumer price collection: compilers add relevant outlets or product varieties as purchasing patterns change.
AI adoption research: analysts refresh a business survey to include emerging service providers while preserving a documented estimation design.

Think of it this way

Think of maintaining a map while a town changes. Replacing a torn section is useful, but new streets also need to be added even when the old map remains intact. Likewise, a survey sample may require additions because the population changes, not only because existing participants disappear.

Worked example

A fictional establishment panel begins with 400 units. During a review period, 30 establishments close and 20 still-eligible establishments stop responding. The usable continuing panel has 400−30−20 = 350 units.
The survey recruits 70 additional establishments under a documented augmentation plan, bringing the usable count to 420. Added units make up 70/420 = 1/6, approximately 16.7% of the resulting panel.
Initially, 40 of the 400 units were in an emerging service segment. Suppose all 40 remain, and 35 of the 70 additions belong to that segment. Its unweighted share becomes 75/420, approximately 17.9%, compared with 10% initially.
These counts demonstrate replenishment and changed composition. They do not establish the segment's population share or prescribe weights. Closure and nonresponse remain distinct, and the combined selection design and current population totals are needed for appropriate estimation.

Illustrative case study

Fictional case: Silverstream Research maintains a panel tracking business software expenditure. Its original sample contains few firms using a newly popular delivery model, even though overall response numbers remain adequate.
The team adds eligible establishments from an updated frame and documents their selection probabilities and entry dates. It reviews weighting and comparability before interpreting any jump in reported spending as a market change. The sample update becomes part of the published methodological record.

Common misconceptions and mistakes

Augmentation is not necessarily one-for-one replacement of missing units. New segments or products may need inclusion while existing units remain.
Restoring the original sample size does not prove restored coverage or absence of nonresponse bias.
Synthetic AI examples are not interchangeable with observed establishments in a probability survey.
Adding a new product's price directly to a time-series average can confuse quality or composition changes with price change.
Augmentation and rotation can overlap operationally, but rotation specifically involves planned turnover of sampled units.

Business use

Use sample augmentation when a recurring information product risks becoming disconnected from its changing market or population. Plan recruitment, frame maintenance, weighting, and continuity together. For business intelligence based on repeated surveys, retaining this methodological history helps distinguish a genuine trend from an improvement or alteration in what the survey observes.

Check your understanding

Why are the 30 closures and 20 nonrespondents separated in the example?

Answer: The former may have left the population, while the latter remain eligible but unobserved.

Can the new 17.9% panel share be reported automatically as the population share?

Answer: No; selection and weights must be considered.

Related terms

Sample rotation
Nonresponse
Coverage error
Data augmentation
sampling frame
panel survey
sample depletion
survey weighting

Learn next

Review a recurring dataset's entry and exit records. Identify whether missing coverage concerns population births, attrition, nonresponse, or new categories, and design an augmentation plan for the specific problem.

Nonresponse

Sources

OECD Glossary of Statistical Terms, Sample augmentation: https://www.oecd.org/content/dam/oecd/en/publications/reports/2008/09/oecd-glossary-of-statistical-terms_g1gh9ad7/9789264055087-en.pdf Consumer Price Index Manual, Maintaining the Sample: https://www.elibrary.imf.org/view/book/9781484354841/ch07.xml The panel counts and illustrative case are original and fictional. IMF-hosted chapter PDF, Maintaining the Sample, sample augmentation and quality-comparability discussion: https://www.imf.org/-/media/files/data/cpi/chapter-7-maintaining-the-sample.pdf

Previous: Sample Next: Sample complexity