Skip to contents

GRF-based doubly-robust ATT estimator (causal forest)

Source: R/est-grf-dr.R
est_grf_dr_att.Rd

This estimator wraps the grf::causal_forest() implementation to estimate the average treatment effect on the treated (ATT). It is treated as "doubly-robust" in the sense of GRF: the causal forest internally uses orthogonalization with nuisance estimates for propensity and outcome.

Usage

est_grf_dr_att(df, config = list(), tau = cs_tau_oracle, ...)

est_grf_dr(df, config = list(), tau = cs_tau_oracle, ...)

Arguments

df

Data frame as returned in dgp$df, containing at least y, w, and covariates. Covariates are taken as all non-core columns (i.e., excluding y, w, y0, y1, p, structural_te).

config

Optional list of configuration options. Common fields include:

  • ci_method: CI intent; one of "none", "default", "bootstrap", "native" (see cs_ci_methods).

  • seed: required when bootstrap CIs are requested.

  • n_boot: number of bootstrap draws if using bootstrap CIs.

  • GRF hyperparameters (e.g., num.trees, min.node.size).

For this estimator, ci_method = "default" maps to "native".

tau

Ignored. Included for API compatibility with other estimators.

...

Ignored. Reserved for future extensions.

Value

A legacy-compatible estimator result containing an att estimate and optional interval, qst = NULL, cf = NULL, and CI/provenance meta.

Details

This estimator is an optional registry entry and requires the grf package. The descriptor is discoverable without grf; execution requires the package.