Is a donor always chosen at random?

No. Selection can use random hot-deck rules, nearest neighbors or other specified methods.

Does a valid donor guarantee a consistent recipient?

No. The copied values can conflict with information retained in the recipient, as the example shows.

Does imputation recover the original missing answer?

Not necessarily. It supplies a plausible value under assumptions, and that distinction should remain visible.

AI Glossary / Statistics and probability / Intermediate

Donor (imputation)

Definition

A donor in imputation is a record that supplies values used to replace missing or inconsistent values in another record, called the recipient. In the OECD glossary's hot-deck editing context, donors are selected from records that pass relevant checks and resemble the recipient on specified fields. Their observed values are transferred to corresponding recipient fields. A donor provides a plausible replacement source, not proof of the recipient's true missing value.

What it means

If a respondent omits an answer, another sufficiently similar respondent may provide a useful value for completing the dataset. Similarity must relate to the variable being imputed, not merely to convenient identifiers. Donor selection can be random within a class, based on nearest neighbors or guided by another documented rule. The source record may be valid on its own while its values conflict with the recipient's retained information. Consequently, donor eligibility and recipient consistency are separate checks, and imputed values should remain identifiable.

How it works

Identify fields needing imputation and construct an eligible donor pool from suitably observed and checked records. Choose matching variables and distance or class rules that reflect relationships with the missing items. Select a donor according to the method, transfer the specified values and rerun relevant recipient edits. When transferring several related fields, using a coherent donor pattern can preserve relationships better than independently mixing incompatible values, though it is not an automatic guarantee. Record donor identity or a suitable audit reference, selection details and imputation flags. Assess donor reuse, coverage and the uncertainty introduced by imputation rather than treating completed records as fully observed.

Three practical examples

Survey processing: fill a missing response using an eligible respondent with similar observed characteristics under a documented hot-deck procedure.
Business data preparation: replace incomplete related fields using a matched complete record while checking accounting identities.
Method evaluation: mask known values, apply donor selection and examine how well the imputations preserve relevant summaries and relationships.

Think of it this way

Think of borrowing an example from a similar completed form to fill a gap. The borrowed answer may be plausible, but it belongs to another case. You must still check whether it fits the recipient's other answers and remember that the completed form contains an estimate rather than a newly observed fact.

Worked example

Suppose a hypothetical recipient record reports a total of ten units, with both category counts missing. Eligible donor A reports total ten with category counts six and four. Eligible donor B reports total twelve with category counts eight and four. Each donor passes the edit that its own two category counts sum to its total: 6 + 4 = 10 and 8 + 4 = 12. Copying A's two category values into the recipient preserves the recipient total. Copying B's values while retaining total ten produces 8 + 4 = 12, which fails that edit. Thus an edit-passing donor does not necessarily produce an edit-passing recipient. Matching on the total or using another appropriate constrained procedure can address this particular conflict, but neither establishes the recipient's actual missing composition.

Illustrative case study

Fictional case: Elm Survey Services imputes missing business questionnaire fields. The team creates donor classes using relevant observed characteristics, tracks how often each donor is used and checks completed records against consistency rules. A small class has too few suitable donors, so analysts review the matching strategy rather than silently copying one record repeatedly. Reports identify the affected fields and explain the method. This invented case illustrates quality controls for donor-based imputation; it does not establish that the missingness mechanism is ignorable or that donor replacement removes bias.

Common misconceptions and mistakes

Do not confuse a donor record with an independent new observation about the recipient. Reusing donors can create dependence and affect uncertainty. Similarity on irrelevant variables does not ensure good imputations, and a narrow donor pool may fail to represent missing cases. Passing edit rules does not prove factual correctness. Single donor imputation generally does not by itself account for all missing-data uncertainty in later standard errors. Values from a previous period or external source may be useful, but their timing and conceptual definitions must align with the recipient data.

Business use

Businesses use donors to complete datasets while preserving plausible observed values and, sometimes, realistic multivariable patterns. Method selection should depend on the analytical purpose, available auxiliary information and reasons values are missing. Keep imputed fields separate from verified source observations in provenance records. For machine-learning evaluation, fit donor-selection logic within the training process to avoid leaking information from held-out data. Evaluate sensitivity of key results to imputation choices and use an uncertainty approach appropriate to the survey or modeling framework.

Check your understanding

Why is donor B unsuitable for direct transfer in the worked example despite passing its own edit?

Answer: B's counts sum to twelve, while the recipient's retained total is ten. The conflict arises from combining values across records, so donor-level validity alone is insufficient.

Related terms

Hot deck imputation
Missing data
Imputation
recipient
cold-deck imputation
nearest-neighbour imputation
matching variable
edit rule
donor reuse
multiple imputation
data provenance

Learn next

define donor eligibility and matching rules for a small incomplete table. Check both donor records and completed recipients, then examine how repeated donor use changes the dataset. Study assumptions about missingness and appropriate variance estimation before treating an imputed dataset as ready for inference.

Variance estimation

Sources

OECD Glossary of Statistical Terms, Donor (imputation): https://www.oecd.org/content/dam/oecd/en/publications/reports/2008/09/oecd-glossary-of-statistical-terms_g1gh9ad7/9789264055087-en.pdf Statistics Canada, Imputation: https://www150.statcan.gc.ca/n1/edu/power-pouvoir/ch3/imputation/5214784-eng.htm Statistics Canada, Quality Guidelines, Imputation: https://www150.statcan.gc.ca/n1/pub/12-539-x/2009001/imputation-eng.htm

Previous: Dominating policy Next: DOPIPE