Plot local overlap for CATE fits — plot

Two views of treatment overlap inside the DNN neighbourhoods of a cate_dnn() fit. type = "mass" shows the weighted treated and control mass per target (for simplex weights the two masses partition one, so the treated mass is the local treated share) with dotted guides and shaded warning bands outside share_limits. type = "propensity" shows the local weighted propensity distribution: weighted densities per selected target when observation-level propensities are available, and the weighted mean with a q10–q90 whisker summary otherwise.

Usage

plot_overlap(
  object,
  type = c("mass", "propensity"),
  share_limits = c(0.1, 0.9),
  treatment = NULL,
  propensity = NULL,
  targets = NULL,
  ...
)

Arguments

object: A cate_result with the weight matrix W stored (the default).
type: One of "mass" or "propensity".
share_limits: Numeric length-2 vector (default c(0.1, 0.9)); treated shares outside this range are flagged as overlap warnings.
treatment: Optional 0/1 vector of length n; used when the fit does not store treatment (matrix interface).
propensity: Optional numeric vector of length n in (0, 1); used when the fit does not store propensities.
targets: Optional integer vector of target ids for the density view (defaults to the first/median-n_eff/last subset above eight targets).
...: Ignored; reserved for future options.

Value

A ggplot object.

Details

Inputs resolve in a fixed order: components stored on the fit (t_vec/e from the data-frame interface), then the treatment/ propensity arguments, then an actionable error. Nothing is recomputed from first stages.

Examples

if (requireNamespace("ggplot2", quietly = TRUE)) {
  set.seed(1)
  n <- 120
  df <- data.frame(X1 = rnorm(n), X2 = rnorm(n))
  df$e <- plogis(0.5 * df$X1)
  df$T <- as.integer(runif(n) < df$e)
  df$m0 <- df$X1
  df$m1 <- df$X1 + 1
  df$Y <- ifelse(df$T == 1, df$m1, df$m0) + rnorm(n, sd = 0.3)
  fit <- cate_dnn(
    data = df, x_cols = c("X1", "X2"), y_col = "Y", t_col = "T",
    e_col = "e", m0_col = "m0", m1_col = "m1",
    X0 = cbind(seq(-1, 1, length.out = 5), 0),
    k = 1L, s = 15L, se = FALSE
  )
  plot_overlap(fit, type = "mass")
}