Is the second answer assumed correct?

No. Establishing accuracy may require external validation or a justified reconciliation process.

Can overall percentages agree while many units disagree?

Yes; opposing changes can cancel in the aggregate.

Is every duplicated sample a completely repeated survey?

No. It may consist only of the common subset investigated by both teams.

AI Glossary / Statistics and probability / Intermediate

Duplicated sample

Definition

A duplicated sample, in the OECD glossary's survey terminology, is a sample whose units are investigated twice by different parties of investigators. It can also be the common subset deliberately included in two interpenetrating samples for field-operation control. Its defining feature is repeated enquiry about the same units. This differs from a duplicate sample, which is an additional concurrent sample collected under comparable conditions. Neither term means accidentally copying records in a data file.

What it means

When two investigators collect information about the same unit, differences between their records can reveal problems that a comparison of unrelated samples would obscure. The shared units create a basis for checking response consistency or investigating collection procedures. However, neither interview is automatically the truth. Disagreement may reflect respondent interpretation, interviewer behavior, a changed reference period, recording errors or genuine change. Agreement can also coexist with a shared misunderstanding. The purpose and design of the repeated enquiry determine which quality questions can reasonably be answered.

How it works

Select the units or overlap according to a documented plan and establish matching identifiers. Specify whether the second team is blinded to the first responses, what questions are repeated and whether reconciliation will occur. Keep the reference period and measurement definitions comparable when assessing consistency. Link records at the unit level and analyze paired differences, transitions or agreement measures rather than pretending the records represent unrelated units. Account for the selection design, unequal weights and reinterview nonresponse where relevant. A study seeking response bias requires additional assumptions or validation evidence beyond simple disagreement rates.

Three practical examples

Interview quality control: ask a selected subset of households the same questions through another investigating team.
Coding assessment: compare independently recorded classifications for the same business units and identify ambiguous instructions.
Survey development: examine whether a question produces inconsistent answers during repeated enquiry under comparable conditions.

Think of it this way

Think of two inspectors checking the same set of labeled packages. Their paired reports can show which packages receive different classifications. The comparison is useful because the packages match, but deciding whose classification is correct may require a separate authoritative check rather than a vote between the inspectors.

Worked example

Consider a hypothetical duplicated sample of 40 units with a yes-or-no item. Both teams record yes for 24 units and no for six. Team A records yes while Team B records no for six units; the reverse occurs for four. Agreement is (24 + 6)/40 = 75%, and disagreement is (6 + 4)/40 = 25%. Team A's yes rate is 30/40 = 75%, while Team B's is 28/40 = 70%, a five-percentage-point difference. The net difference is much smaller than the disagreement rate because opposite changes partially cancel. These descriptive calculations do not identify which answers are correct, establish interviewer causation or estimate population bias. Inference would require the sampling design and appropriate paired analysis, with attention to any nonresponse or changes between enquiries.

Illustrative case study

Fictional case: Maple Survey Operations asks a separate team to recontact a planned subset of business respondents about the same reporting period. The team records answers before any reconciliation. Analysts find that opposing classification changes mostly cancel in the overall rate, while unit-level disagreement remains substantial. They review question wording and field instructions, then seek validation records where permitted. This invented case illustrates why aggregate agreement can conceal individual inconsistencies. It does not portray every difference as misconduct or assume the second interview is more accurate.

Common misconceptions and mistakes

Do not count two interviews with one unit as two independent population units. Do not confuse a duplicated sample with two independently selected samples merely because both involve repetition. Recontact can introduce memory, learning or consistency effects; an unblinded interviewer can also influence the second record. A later interview may refer to a genuinely changed situation unless the reference period is controlled. Disagreement alone measures inconsistency, not the direction or magnitude of bias relative to truth. Removing unmatched or nonresponding units without examination can distort the quality assessment.

Business use

Businesses commissioning surveys can use repeated enquiry to improve training, question design and collection controls. Define the quality objective before selecting the overlap and preserve both original records alongside any reconciled value. Report the number selected, successfully matched and analyzed, together with reasons for exclusions. Use identified patterns to guide investigation rather than immediately assigning blame. The exercise has costs and respondent burden, so its design should focus on meaningful risks while maintaining a clear distinction between operational checking and estimation of population characteristics.

Check your understanding

If ten of 40 paired records disagree but the two yes totals differ by only two, which number describes inconsistency?

Answer: Ten disagreements, or 25%. The net difference of two summarizes direction after cancellations and answers a different question.

Related terms

Duplicate sample
Response Bias
Measurement error
reinterview
response variance
paired observations
agreement
interpenetrating samples

Learn next

construct a paired response table, distinguish gross disagreement from net change, and study what additional validation is needed to estimate error relative to an underlying true response.

Sources

Sources: OECD Glossary of Statistical Terms, Duplicated sample, printed page 154: https://www.oecd.org/content/dam/oecd/en/publications/reports/2008/09/oecd-glossary-of-statistical-terms_g1gh9ad7/9789264055087-en.pdf ; Statistics Canada, Reinterview programs and response errors: https://www150.statcan.gc.ca/n1/en/catalogue/12-001-X197500100009 ; US Census Bureau, Content Reinterview Survey evaluation purpose: https://www.census.gov/programs-surveys/decennial-census/decade/2010/program-management/cpex/2010-cpex-206.html ; US Census Bureau, American Community Survey 2012 Content Reinterview Survey: https://www.census.gov/content/dam/Census/library/working-papers/2014/acs/2014_Murphy_01.pdf .

Previous: Duplicate sample Next: Durability (database systems)