Plot per-observation contributions to a CATE estimate

For a target j, each observation contributes w_ij (psi_i - tau_hat_j) to the localized estimating equation; large contributions mean the estimate leans on few observations. Drawn against rank (or stored distance when available), with the largest three absolute contributions labelled by observation index and the weight-concentration diagnostics in the caption.

Usage

plot_influence(
  object,
  targets = NULL,
  pseudo_outcome = NULL,
  n_label = 3L,
  ...
)

Arguments

object: A cate_result with the weight matrix W stored (the default).
targets: Optional integer vector of target ids (defaults to the first/median-n_eff/last subset above eight targets).
pseudo_outcome: Optional numeric vector of length n; used when the fit does not store the pseudo-outcome.
n_label: Integer; number of largest absolute contributions to label per target (default 3).
...: Ignored; reserved for future options.

Value

A ggplot object.

Details

The pseudo-outcome is taken from the fit; supply pseudo_outcome for fits that do not store it.

Examples

if (requireNamespace("ggplot2", quietly = TRUE)) {
  set.seed(1)
  n <- 100
  X <- matrix(rnorm(n * 2), n, 2)
  psi <- rnorm(n, mean = X[, 1])
  fit <- cate_dnn(
    X = X, x0 = c(0, 0), pseudo_outcome = psi,
    k = 1L, s = 15L, se = FALSE, return = c("W", "dx")
  )
  plot_influence(fit)
}
#> Ignoring unknown labels:
#> • colour : "target"