Main characteristics of multiple imputation (MI)
Replace each missing value \(𝑦^{mis}\) with M plausible values
Main characteristics of multiple imputation (MI)
Each imputed value is generated conditionally from the observed data
Main characteristics of multiple imputation (MI)
Borrowing of information across variables requires to define a multivariate distribution
Strategies to allow borrowing of information
Simultaneous approaches that define a multivariate distribution f(Y ) directly (“joint modelling”)
Sequential approaches that build up a multivariate distribution using a ladder of conditional distributions, where the model for each variable conditions only on those earlier in the sequence (“fully conditional specification”)