Post-process models to recover Qini curves via propensity trimming

for any model in a results list whose qini_objects are null or empty, this function applies an overlap restriction on the estimated\ n#' propensity scores and recomputes in‑sample Qini curves on the trimmed data, without touching the original ATE or forest objects.

Usage

margot_rescue_qini(
  model_results,
  propensity_bounds = c(0.05, 0.95),
  verbose = TRUE
)

Arguments

model_results

a list from margot_causal_forest() with save_models = TRUE and save_data = TRUE

propensity_bounds

numeric length‑2 vector of lower/upper bounds for forest$W.hat (default c(0.05, 0.95))

verbose

logical; if TRUE prints progress messages (default TRUE)

Value

modified model_results with rescued qini_data and qini_objects for any models that initially had empty gain