I am currently facing a set of data with missing values. I would like to impute these values with a Hot Deck. I have read upon the Hot Deck methods and decided to use Nearest Neigbour Hot Deck (NNHD) where k = 1. Additionally, I am using the Gower Distance to compute the similarity because my data set includes mixed data. I know that my data and its observations are split into donor and recipient.


I am sadly struggling to understand one thing. The NNHD imputes a missing value within a feature with a value from a donor within the same feature. Is the now imputed value going to be used for imputation within that feature? Or are only the "real" observed values used?

  • $\begingroup$ The imputed observation does not serve as a donor, of course (during the same run of the imputation procedure). However, repeated using of a donor (i.e. if a donor observation can impute into more than one missing observations) - this is a potential option to select or not, as you wish. For example, my macros for SPSS doing hot-deck imputation have such feature. $\endgroup$ – ttnphns Dec 13 '19 at 15:20
  • $\begingroup$ Thank you very much for your answer. It does help my understanding! Does this mean I can safely assume that I have two different sets. One for recipients one for donors. Once I start imputing the recipient will only take values from the donor set? $\endgroup$ – Alchtech Dec 13 '19 at 16:01
  • $\begingroup$ That is right. Imputation does not imply a snow-slip process whereby former recipients instantly enroll to the donors list. $\endgroup$ – ttnphns Dec 13 '19 at 18:26

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.