Generalizability Overlap Analysis for PATE Estimates

Compute generalizability overlap scores and optionally trim observations with low overlap to produce a trimmed PATE estimate.

Usage

general_BART_overlap(
  object,
  threshold = NULL,
  overlap_value = c("zero", "NA"),
  verbose = FALSE
)

Arguments

object

A general_pate object from general_BART()

threshold

Numeric threshold for trimming based on standardized selection scores. Observations with |e_s_tilde| > threshold are excluded. Default is NULL (no trimming, just compute overlap).

overlap_value

Value to replace tau with for trimmed observations. Either "zero" (set to 0) or "NA" (exclude completely). Default is "zero".

verbose

Logical; print progress messages. Default is FALSE.

Value

A list with components:

overlap

Data frame with overlap metrics: pi_c, pi_t, pi_s, e_s_tilde

e_s_tilde_source

Standardized selection scores for source data

n_trimmed

Number of observations trimmed (if threshold used)

pate_trimmed

Trimmed PATE estimate (if threshold used)

ci_trimmed

95 percent CI for trimmed PATE (if threshold used)

See also

general_BART(), general_BART_transportability()

Examples

if (FALSE) { # \dontrun{
fit <- general_BART(princebart_fit, newdata, ...)
overlap <- general_BART_overlap(fit, threshold = 2)
} # }