What is the primary risk of using IterativeImputer on a dataset with high multicollinearity among features?

Question

Data Science — Hard

What is the primary risk of using IterativeImputer on a dataset with high multicollinearity among features?

Accepted Answer

IterativeImputer models each feature as a function of others using regression. When features are highly multicollinear, the underlying regressor (often BayesianRidge or LinearRegression) faces ill-conditioned matrices, leading to numerical instability during coefficient estimation. While variance underestimation (B) is a concern for single imputation, the immediate technical failure in high multicollinearity is matrix inversion instability.