hotdeck

Hotdeck is a data imputation technique used to fill in missing values in a dataset by borrowing observed values from similar records, called donors. It is a non-parametric approach that does not rely on fitting a statistical model to the data; instead, it uses the observed data within a defined donor pool to replace missing entries.

The imputation process typically involves three steps. First, the data are partitioned into imputation classes or decks based on variables related to the variable being imputed or the missingness mechanism. Second, for each record with a missing value, a donor is selected from the corresponding deck, either at random or using a distance-based criterion like nearest neighbors. Third, the donor’s value for the target variable is copied into the missing slot. In some applications, multiple donors may be used to create several imputed datasets, enabling analysis that accounts for imputation uncertainty.

Variants of hotdeck differ in how donors are chosen and how decks are formed. Classical hotdeck uses fixed imputation classes, while nearest-neighbor hotdeck selects donors based on similarity across multiple variables. Hotdeck can impute one variable at a time or iteratively across several variables. It is suitable for both numeric and categorical data, with imputed values chosen from the corresponding donor’s value.

Advantages include preservation of the original data distribution and simplicity of implementation. Limitations include potential bias if the donor pool is not representative or if the matching criteria do not adequately capture relationships among variables. Multiple-imputation extensions help mitigate uncertainty and improve inferential validity. Hotdeck is commonly used in survey data, census processing, and other contexts where maintaining observed data integrity is important.