AIPW / one-step estimator (= DML with cross-fit nuisances)

Doubly robust: consistent if EITHER g or Q is correct. The estimate is the plug-in plus the mean of the efficient influence function; the standard error comes from the influence curve (no bootstrap). With strategy = "crossfit" nuisances this is exactly double machine learning (DML).

Usage

dr_aipw(Y, A, nu)

Arguments

Y

Numeric outcome vector.

A

Binary treatment vector (0/1).

nu

A nuisance object from dr_nuisance().

Value

A list with estimate, se, ci, ic (influence curve), method.

References

Robins, Rotnitzky & Zhao (1994); Chernozhukov et al. (2018).