Targeted maximum likelihood estimator (TMLE)

Doubly robust. Fluctuates the outcome fit Q through a logistic submodel with the clever covariate H(A, X), then plugs in. Continuous outcomes are scaled to [0, 1] (Gruber & van der Laan) so the logistic update is valid; binary outcomes use y_range = c(0, 1) (the scaling is then the identity). With strategy = "crossfit" nuisances this is CV-TMLE.

Usage

dr_tmle(Y, A, nu, y_range = NULL)

Arguments

Y

Numeric outcome vector.

A

Binary treatment vector (0/1).

nu

A nuisance object from dr_nuisance().

y_range

Optional outcome range c(min, max) used for scaling. Defaults to range(Y); pass c(0, 1) for binary outcomes.

Value

A list with estimate, se, ci, ic, eps (fluctuation), method.

References

van der Laan & Rubin (2006); Gruber & van der Laan (2012).