Which issue arises when using MICE (Multivariate Imputation by Chained Equations) on a dataset with high multicollinearity?

Question

Data Science — Hard

Which issue arises when using MICE (Multivariate Imputation by Chained Equations) on a dataset with high multicollinearity?

Accepted Answer

MICE relies on regressing each feature with missing values against other features. When features are highly multicollinear, the design matrix becomes ill-conditioned. This leads to unstable coefficient estimates in the underlying regression models, potentially resulting in extreme or nonsensical imputed values. While MICE is robust, it is sensitive to the stability of the regression models used within the chain.