Expand description
Cluster-robust DML (Chiang, Kato, Ma & Sasaki, 2022), as DoubleML: one-way or two-way clustered observations.
With clustered data, sample splitting assigns whole clusters to folds (for
two-way clustering the folds are the K² combinations of the two cluster
dimensions, so no cluster appears in both the training and the test set),
θ solves a fold-weighted score equation, and the variance sums the score
within clusters.
Structs§
- Cluster
Fold - One fold of a cluster split: the observation-level fold and, for each cluster variable, the clusters in its test set.