Compute RATE (Rank Average Treatment Effect) On-Demand

This internal helper function computes RATE metrics (AUTOC or QINI) on-demand from a causal forest and treatment effect estimates. It ensures proper out-of-sample validation when possible.

Usage

compute_rate_on_demand(
  forest,
  tau_hat = NULL,
  target = c("AUTOC", "QINI"),
  q = seq(0.1, 1, by = 0.1),
  policy = c("treat_best", "withhold_best"),
  subset = NULL,
  use_oob_predictions = TRUE,
  verbose = FALSE,
  seed = 12345,
  ...
)

Arguments

forest

A causal_forest object from grf

tau_hat

Optional vector of treatment effect estimates. If NULL, will be computed from the forest using out-of-bag predictions by default.

target

Character; either "AUTOC" or "QINI"

q

Numeric vector of quantiles at which to evaluate the TOC. Default is seq(0.1, 1, by = 0.1) which matches the GRF default.

policy

Character; either "treat_best" (default) or "withhold_best"

subset

Optional indices for subsetting the evaluation data

use_oob_predictions

Logical; if TRUE and tau_hat is NULL, use out-of-bag predictions for better validity (default TRUE)

verbose

Logical; print informative messages (default FALSE)

seed

Random seed for reproducibility (default 12345)

...

Additional arguments passed to grf::rank_average_treatment_effect()

Value

A rank_average_treatment_effect object from grf

Details

For valid statistical performance, the prioritization scores (tau_hat) should be constructed independently from the evaluation forest training data. This function attempts to ensure this by: - Using out-of-bag predictions when tau_hat is not provided - Supporting subset indices for proper train/test splitting - Warning when validation might be compromised