thomaswiemann / ddml

ddml: Double/Debiased Machine Learning in R
https://thomaswiemann.com/ddml/
GNU General Public License v3.0
15 stars 1 forks source link

Allow for different model specifications in first stage and residualization #17

Closed thomaswiemann closed 3 years ago

thomaswiemann commented 3 years ago

Currently, the same set of user-provided models are considered for the computation of E[D|X,Z], E[D|X], and E[y|X]. It would be an improvement to allow for a potentially different set of models in the computation of E[D|X,Z]. For example, lasso-selection among instruments, but no selection among controls.