double_robust_scores: Matrix Gamma of scores for each treatment a

In policytree: Policy Learning via Doubly Robust Empirical Welfare Maximization over Trees

Matrix \Gamma of scores for each treatment a

Description

Computes a matrix of double robust scores \Gamma_{ia} = \mu_a(x) + \frac{1}{e_a(x)} (Y_i - \mu_a(x)) 1(A_i=a)

Usage

## S3 method for class 'causal_forest'
double_robust_scores(object, ...)

## S3 method for class 'causal_survival_forest'
double_robust_scores(object, ...)

## S3 method for class 'instrumental_forest'
double_robust_scores(object, compliance.score = NULL, ...)

## S3 method for class 'multi_arm_causal_forest'
double_robust_scores(object, outcome = 1, ...)

double_robust_scores(object, ...)

Arguments

object	An appropriate causal forest type object
...	Additional arguments
compliance.score	An estimate of the causal effect of Z on W. i.e., Delta(X) = E(W | X, Z = 1) - E(W | X, Z = 0), for each sample i = 1, ..., n. If NULL (default) then this is estimated with a causal forest.
outcome	Only used with multi arm causal forets. In the event the forest is trained with multiple outcomes Y, a column number/name specifying the outcome of interest. Default is 1.

Details

This is the matrix used for CAIPWL (Cross-fitted Augmented Inverse Propensity Weighted Learning)

Value

A matrix of scores for each treatment

Methods (by class)

• double_robust_scores(causal_forest): Scores (\Gamma_0, \Gamma_1)

• double_robust_scores(causal_survival_forest): Scores (\Gamma_0, \Gamma_1)

• double_robust_scores(instrumental_forest): Scores (-\Gamma, \Gamma)

• double_robust_scores(multi_arm_causal_forest): Matrix \Gamma of scores for each treatment a

Note

For instrumental_forest this method returns (-\Gamma_i, \Gamma_i) where \Gamma_i is the double robust estimator of the treatment effect as in eqn. (44) in Athey and Wager (2021).

References

Athey, Susan, and Stefan Wager. "Policy Learning With Observational Data." Econometrica 89.1 (2021): 133-161.

Examples

# Compute double robust scores for a multi-arm causal forest
n <- 500
p <- 10
X <- matrix(rnorm(n * p), n, p)
W <- as.factor(sample(c("A", "B", "C"), n, replace = TRUE))
Y <- X[, 1] + X[, 2] * (W == "B") + X[, 3] * (W == "C") + runif(n)
forest <- grf::multi_arm_causal_forest(X, Y, W)
scores <- double_robust_scores(forest)
head(scores)

# Compute double robust scores for a causal forest
n <- 500
p <- 10
X <- matrix(rnorm(n * p), n, p)
W <- rbinom(n, 1, 0.5)
Y <- pmax(X[, 1], 0) * W + X[, 2] + pmin(X[, 3], 0) + rnorm(n)
c.forest <- grf::causal_forest(X, Y, W)
scores <- double_robust_scores(c.forest)

policytree documentation built on June 22, 2024, 9:47 a.m.