Is a copied spreadsheet a duplicate sample?

Not in this survey-design sense; it contains no new sample selection.

Must the two estimates match?

No. Sampling variation ordinarily produces differences.

Does this repeat an entire experiment?

Not necessarily. Sometimes only the act of selecting an additional sample is replicated.

AI Glossary / Statistics and probability / Intermediate

Duplicate sample

Definition

In the survey terminology used by the OECD glossary, a duplicate sample is an additional sample collected alongside a first sample in the same survey under comparable conditions. It can provide a form of replication, particularly when the additional sample is selected independently. The repeated element may be the sampling operation itself rather than an entire experiment. This technical meaning differs from copying a dataset and from a duplicated sample, where different investigating teams enquire about the same sample units.

What it means

Two survey estimates can differ because the selected units differ. A properly designed additional sample helps examine this source of variation while keeping the target population and collection setting comparable. It does not automatically repeat every source of uncertainty in the original work. For example, both samples may use the same incomplete sampling frame or the same misleading question. Agreement between them would therefore not demonstrate freedom from shared bias. The term describes a sampling arrangement, so the actual design and estimation rules remain essential.

How it works

Define the population, reference period, selection procedure and measurement protocol for both samples. Where independent sampling is intended, implement that independence in the random selection rather than merely assigning different file names. Calculate estimates using the correct inclusion probabilities and weights. Compare their differences with their expected sampling variability and document any shared operational influences. Combining estimates requires an appropriate variance calculation, including covariance when independence does not hold. Two replicate estimates alone offer limited information about the variability of the entire procedure; additional replication or design-based variance methods may be needed for a dependable precision assessment.

Three practical examples

Survey planning: collect an independent concurrent sample to examine the stability of an estimated population mean.
Methods teaching: demonstrate why two legitimate probability samples can yield different answers without either being fraudulent.
Quality review: distinguish replication of sample selection from repetition of the questionnaire, interview team or experimental treatment.

Think of it this way

Think of drawing two separately randomized scoops from the same well-defined container while following the same measuring rules. The scoops may differ naturally. Repeating the draw provides information about that variation, but it does not reveal a shared defect in the container's contents or the measuring instrument.

Worked example

Suppose two hypothetical independent concurrent samples estimate the same population mean as 48 and 52 units. Assume each estimator is unbiased and has a known sampling standard deviation of three units. Their difference is four units, and its standard deviation is sqrt(3 squared + 3 squared) = sqrt(18), approximately 4.243. The observed difference is therefore about 0.943 standard deviations, not automatically evidence of a collection problem. Their equally weighted average is 50, with variance (9 + 9)/4 = 4.5 and standard deviation approximately 2.121. These calculations require the stipulated independence and comparable target. If their covariance were positive, the average's variance would be larger than 4.5. The example uses known variances for clarity and does not estimate a reliable variance from just the two observed values.

Illustrative case study

Fictional case: Juniper Services commissions two concurrent probability samples of customer accounts to estimate average service usage. Both teams use the same reference month and definitions, while selection is separately randomized. Analysts compare the estimates against their sampling uncertainties and investigate any differences in response rates. They also audit the shared account frame because concurrent agreement could conceal omitted customers. This invented case illustrates the value and limits of repeated sampling. It does not claim that buying a second sample is always the best use of a survey budget.

Common misconceptions and mistakes

Do not mistake duplicate sampling for duplicating records, which adds no new sampling information. Do not assume two samples are independent merely because they were collected by different contractors. Shared selection constraints, coordinated allocation or overlapping operational errors can affect their relationship. Independently selected samples can contain some common units by chance; deliberately excluding overlap changes the joint design and may change covariance. Different reference periods or definitions can create differences unrelated to sampling error. A small difference is not proof that nonresponse, undercoverage or measurement bias is absent.

Business use

Businesses can use the concept to clarify what a repeated survey actually checks. Before paying for replication, specify whether the objective is precision, assessment of collection operations or confirmation under a different method. Preserve separate sample identifiers and selection documentation so analysts can calculate combined uncertainty correctly. Report both the scope of the replication and its limitations. When both samples inherit the same weaknesses, resources may be better directed toward checking those weaknesses rather than interpreting concordant estimates as independent confirmation of total accuracy.

Check your understanding

Two independent mean estimates each have variance nine. What is the variance of their equal average?

Answer: (9 + 9)/4 = 4.5. Averaging variances directly would give the wrong result because the weights must be squared.

Related terms

Duplicated sample
Sampling error
Covariance
replicated sampling
survey weights
interpenetrating samples
nonsampling error

Learn next

derive the variance of a weighted combination of two estimates, then examine how shared frames, nonresponse and overlapping sample designs affect what replication can establish.

Sources

Sources: OECD Glossary of Statistical Terms, Duplicate sample, printed page 153: https://www.oecd.org/content/dam/oecd/en/publications/reports/2008/09/oecd-glossary-of-statistical-terms_g1gh9ad7/9789264055087-en.pdf ; Statistics Canada, Questions to ask about surveys, sampling and total error: https://www.statcan.gc.ca/en/customized-products-services/custom-surveys/questions-to-ask ; US Census Bureau, Nonsampling Error: https://www.census.gov/programs-surveys/sipp/methodology/nonsampling-error.html .

Previous: Dunning-Kruger Effect Bias Next: Duplicated sample