Does direct sampling require an in-person interview?

No. The defining issue is obtaining information from the population units themselves, not a single contact mode.

Does it guarantee equal selection probabilities?

No. Selection probabilities depend on the sampling design, which is a separate matter.

Is sampling database records always inappropriate?

No. It can be appropriate when records match the research question and their coverage, accuracy and reference period are understood.

AI Glossary / Statistics and probability / Intermediate

Direct sampling

Definition

Direct sampling, in the OECD glossary's survey-statistics sense, means that sampled units are actual members of the population being studied and information is obtained from those units themselves. It contrasts with sampling records that stand for the population, such as previously completed forms. The distinction concerns the directness of observation, not the selection algorithm. Direct sampling therefore does not automatically mean random sampling, simple random sampling or an unbiased estimate.

What it means

Suppose a business wants to know about its equipment. Selecting machines and measuring them directly differs from selecting old inspection reports and extracting the recorded measurements. Both approaches may be useful, but they observe different sources and potentially different reference periods. An electronic list can still help select the machines without making the eventual observation indirect in this sense. This historical glossary usage also differs from computational uses of direct sampling to mean generating values from a probability distribution without an iterative sampling chain.

How it works

Define the target population and identify its units. Choose a sampling design appropriate to the desired estimates, then collect the specified information from selected population members. Record how units were selected, contacted or measured, including omissions and failures to obtain observations. Keep separate documentation for the selection mechanism and the observation method. If selection probabilities are known, use suitable design weights and uncertainty estimation; if they are not, do not assume direct observation supplies them. The direct-versus-indirect label identifies the data source relationship, while randomization, stratification, clustering and response handling determine other important properties of the study.

Three practical examples

Equipment review: select devices from an inventory and measure their current operating characteristics rather than sampling historical maintenance reports.
Customer research: select eligible customers and obtain their responses directly under a documented survey design.
Agricultural measurement: observe selected plants or plots themselves instead of extracting values solely from previously filed production records.

Think of it this way

Think of checking selected books on a shelf rather than checking selected catalog cards describing them. The catalog may help locate the books, but looking at the books is a different observation step. Neither method alone tells you whether the selection represents the entire library or whether the recorded information is accurate.

Worked example

Suppose a hypothetical company has a complete inventory of 1,200 devices. It selects 120 by simple random sampling without replacement and directly inspects every selected device. Twelve meet a prespecified defect definition. The observed defect proportion is 12/120 = 0.10, and the expansion estimate of the total is 1,200 x 0.10 = 120 devices. Each device's inclusion probability is 120/1,200 = 0.10 and its design weight is ten, so summing the twelve defect indicators with those weights also gives 120. The total is an estimate, not a count of all defective devices. Direct inspection explains the observation source; simple random selection and complete observation justify this particular calculation. Sampling 120 old inspection records would instead be indirect observation of the device population under the glossary distinction.

Illustrative case study

Fictional case: Maple Facilities wants a current estimate of faulty sensors. Its team considers reviewing archived work orders but recognizes that these records mainly describe previously reported problems. It instead designs a sample of installed sensors and obtains current measurements from the selected units. The team documents inaccessible sensors, measurement procedures and inventory coverage before reporting results. This invented case shows how direct observation can align the source with a current question. It does not establish that such a study is always cheaper, more accurate or less biased than an appropriately designed records-based approach.

Common misconceptions and mistakes

Do not equate direct sampling with probability sampling: a convenience sample can observe units directly. Conversely, using a register to select units does not necessarily mean the observations come from that register. Direct observation does not remove measurement error, nonresponse, incomplete coverage or observer effects. Records can be valuable and sometimes preferable, particularly when they provide relevant historical information that cannot be recollected accurately. State the population and observation target carefully because records themselves can also be the units of interest in a different study. The terminology depends on what the study is trying to describe.

Business use

For businesses, this distinction supports data provenance and fit-for-purpose assessment. A new measurement campaign and a review of existing records have different costs, timeliness, access requirements and error risks. Teams should compare those tradeoffs rather than assuming one source is universally superior. AI systems trained on either source inherit its definitions and coverage limitations. Documenting which information came directly from units and which came from records helps reviewers understand the resulting dataset and avoids describing historical administrative data as newly observed current conditions.

Check your understanding

Which feature makes the device example direct sampling, and which feature supports the weight of ten?

Answer: Inspecting the selected devices themselves makes the observations direct in the glossary sense. The simple random design selecting 120 of 1,200 devices gives inclusion probability 0.10 and inverse-probability weight ten. These are distinct aspects of the study.

Related terms

Indirect sampling
Population
Coverage error
Measurement error
Administrative data
sampling unit
sampling frame
probability sampling
inclusion probability
survey weight
data provenance

Learn next

distinguish target populations, sampling frames and observation units. Compare direct collection with record-based analysis for the same question, identifying their reference periods and likely errors. Then study probability sample designs and weighting so that the source of an observation is not confused with the basis for statistical inference.

Sources

OECD Glossary of Statistical Terms, Direct sampling and Indirect sampling: https://www.oecd.org/content/dam/oecd/en/publications/reports/2008/09/oecd-glossary-of-statistical-terms_g1gh9ad7/9789264055087-en.pdf Statistics Canada, Probability sampling: https://www150.statcan.gc.ca/n1/edu/power-pouvoir/ch13/prob/5214899-eng.htm Statistics Canada, Non-sampling error: https://www150.statcan.gc.ca/n1/edu/power-pouvoir/ch6/nse-endae/5214806-eng.htm

Previous: Diphthong Next: Direct utility estimation