Data Analysis with R, Second Edition - Second Edition

Methods of imputation

The method of imputation that we described for the univariate case, norm, works best for imputed values that follow an unconstrained normal distribution; it could lead to some nonsensical imputations otherwise. For example, since the weights in wt are so close to zero (because it's in units of 1,000 pounds) it is possible for the norm method to impute a negative weight. Though this will no doubt balance out over the other m-1 multiply imputed datasets, we can combat this situation by using another method of imputation called predictive mean matching.

Predictive mean matching (mice calls this pmm) works a lot like norm. The difference is that the norm imputations are then used to find the d closest values to the imputed ...