Qini Curve and AUUC for Uplift / Policy Evaluation

Computes and plots the Qini curve and Area Under the Uplift Curve (AUUC), which measure the quality of treatment effect heterogeneity estimates for targeting / policy evaluation. Compares a CATE-based targeting rule against random assignment and perfect targeting benchmarks.

Usage

qini_curve(
  cate_hat,
  Y,
  W,
  compare_random = TRUE,
  compare_oracle = FALSE,
  n_bins = 100,
  conf_level = 0.95,
  boot_reps = 500,
  seed = 42
)

Arguments

cate_hat: Numeric vector of estimated CATEs.
Y: Numeric vector. Observed outcomes.
W: Numeric vector. Binary treatment assignment (0 or 1).
compare_random: Logical. If TRUE (default), plot the random targeting baseline.
compare_oracle: Logical. If FALSE (default), do not plot an oracle curve (true CATEs rarely available).
n_bins: Integer. Number of bins for the curve. Default 100.
conf_level: Numeric. Confidence level for bootstrap CI on AUUC. Default 0.95.
boot_reps: Integer. Bootstrap replicates for AUUC CI. Default 500.
seed: Integer. Random seed. Default 42.

Value

A list with:

auuc: Numeric. Area under the Qini curve (normalized 0-1).
auuc_ci: Numeric vector of length 2. Bootstrap confidence interval for AUUC.
qini_df: Data frame with columns fraction, uplift.
plot: ggplot2 Qini curve plot.

Details

The Qini curve plots the fraction of treated units (ranked by estimated CATE, highest first) against the cumulative incremental outcome (uplift). The AUUC normalizes this to a value between 0 and 1, relative to the perfect oracle. Values above 0.5 indicate better-than-random targeting.

References

Radcliffe, N. J. (2007). Using control groups to target on predicted lift: Building and assessing uplift models. Direct Marketing Analytics Journal, 1(3), 14-21.

Athey, S., & Imbens, G. W. (2017). The econometrics of randomized experiments. In Handbook of Economic Field Experiments (Vol. 1, pp. 73-140). North-Holland.

Examples

if (FALSE) { # \dontrun{
library(grf)
n <- 1000; p <- 5
X <- matrix(rnorm(n * p), n, p)
W <- rbinom(n, 1, 0.5)
tau <- pmax(X[, 1], 0)
Y   <- tau * W + rnorm(n)
cf  <- causal_forest(X, Y, W)
tau_hat <- predict(cf)$predictions

result <- qini_curve(tau_hat, Y, W)
result$plot
result$auuc
} # }