Both the propensity g and the outcome regression
bar Q are fitted by post-lasso; the one-dimensional
targeting step is unpenalised.
Usage
morie_penalised_tmle(y, D, X, penalty = 0.05, iters = 100)
Arguments
- y
Numeric outcome in [0,1].
- D
Numeric treatment vector.
- X
Numeric covariate matrix.
- penalty
Lasso penalty.
- iters
Maximum targeting iterations.
Value
A list with the estimate, SE, CI, EIC, and support.
References
Belloni, A. & Chernozhukov, V. (2013); van der Laan,
M. J. & Gruber, S. (2016).
Examples
set.seed(11)
n <- 300
X <- matrix(rnorm(n * 5), n)
D <- rbinom(n, 1, plogis(X[, 1]))
y <- plogis(-0.3 + 0.5 * D + X[, 1] + rnorm(n, sd = 0.3))
morie_penalised_tmle(y, D, X, penalty = 0.05)$estimate
#> [1] 0.09580897
