R/rank_average_treatment.R
rank_average_treatment_effect.fit.RdProvides an optional interface to rank_average_treatment_effect which allows for user-supplied
evaluation scores.
rank_average_treatment_effect.fit( DR.scores, priorities, target = c("AUTOC", "QINI"), q = seq(0.1, 1, by = 0.1), R = 200, sample.weights = NULL, clusters = NULL )
| DR.scores | A vector with the evaluation set scores. |
|---|---|
| priorities | Treatment prioritization scores S(Xi) for the units in the evaluation set. Two prioritization rules can be compared by supplying a two-column array or named list of priorities (yielding paired standard errors that account for the correlation between RATE metrics estimated on the same evaluation data). WARNING: for valid statistical performance, these scores should be constructed independently from the evaluation dataset used to construct DR.scores. |
| target | The type of RATE estimate, options are "AUTOC" (exhibits greater power when only a small subset of the population experience nontrivial heterogeneous treatment effects) or "QINI" (exhibits greater power when the entire population experience diffuse or substantial heterogeneous treatment effects). Default is "AUTOC". |
| q | The grid q to compute the TOC curve on. Default is (10%, 20%, ..., 100%). |
| R | Number of bootstrap replicates for SEs. Default is 200. |
| sample.weights | Weights given to an observation in estimation. If NULL, each observation is given the same weight. Default is NULL. |
| clusters | Vector of integers or factors specifying which cluster each observation corresponds to. Default is NULL (ignored). |
A list of class `rank_average_treatment_effect` with elements
estimate: the RATE estimate.
std.err: bootstrapped standard error of RATE.
target: the type of estimate.
TOC: a data.frame with the Targeting Operator Characteristic curve estimated on grid q, along with bootstrapped SEs.
# \donttest{ # Estimate CATEs with a causal forest. n <- 2000 p <- 5 X <- matrix(rnorm(n * p), n, p) W <- rbinom(n, 1, 0.5) event.probability <- 1 / (1 + exp(2 * (pmax(2 * X[, 1], 0) * W - X[, 2]))) Y <- 1 - rbinom(n, 1, event.probability) train <- sample(1:n, n / 2) cf.cate <- causal_forest(X[train, ], Y[train], W[train]) # Predict treatment effects on a held-out test set. test <- -train cate.hat <- predict(cf.cate, X[test, ])$predictions # Estimate AIPW nuisance components on the held-out test set. Y.forest.eval <- regression_forest(X[test, ], Y[test], num.trees = 500) Y.hat.eval <- predict(Y.forest.eval)$predictions W.forest.eval <- regression_forest(X[test, ], W[test], num.trees = 500) W.hat.eval <- predict(W.forest.eval)$predictions cf.eval <- causal_forest(X[test, ], Y[test], W[test], Y.hat = Y.hat.eval, W.hat = W.hat.eval) # Form doubly robust scores. tau.hat.eval <- predict(cf.eval)$predictions debiasing.weights.eval <- (W[test] - W.hat.eval) / (W.hat.eval * (1 - W.hat.eval)) Y.residual.eval <- Y[test] - (Y.hat.eval + tau.hat.eval * (W[test] - W.hat.eval)) DR.scores <- tau.hat.eval + debiasing.weights.eval * Y.residual.eval # Could equivalently be obtained by # DR.scores <- get_scores(cf.eval) # Form a doubly robust RATE estimate on the held-out test set. rate <- rank_average_treatment_effect.fit(DR.scores, cate.hat) rate#> estimate std.err target #> 0.2250861 0.02385237 priorities | AUTOC# Same as # rate <- rank_average_treatment_effect(cf.eval, cate.hat) # In settings where the treatment randomization probabilities W.hat are known, an # alternative to AIPW scores is to use inverse-propensity weighting (IPW): # 1(W=1) * Y / W.hat - 1(W=0) * Y / (1 - W.hat). # Here, W.hat = 0.5, and an IPW-based estimate of RATE is: IPW.scores <- ifelse(W[test] == 1, Y[test] / 0.5, -Y[test] / 0.5) rate.ipw <- rank_average_treatment_effect.fit(IPW.scores, cate.hat) rate.ipw#> estimate std.err target #> 0.2225889 0.04545525 priorities | AUTOC# IPW-based estimators typically have higher variance. For details on # score constructions for other causal estimands, please see the RATE paper. # }