Skip to contents

Sensitivity Analysis for Unmeasured Confounding

general_BART_transportability.Rd

Compute sensitivity bounds for the PATE estimate under potential unmeasured confounding using weight-shift optimization.

Usage

general_BART_transportability(object, gamma, n_sample = 100, verbose = FALSE)

Arguments

object

A general_pate object from general_BART()

gamma

Numeric sensitivity parameter > 1. Represents the maximum ratio by which observation weights can be shifted. Larger values allow more severe confounding.

n_sample

Number of posterior draws to use for sensitivity analysis. Default is 100.

verbose

Logical; print progress messages. Default is FALSE.

Value

A list with components:

gamma

The sensitivity parameter used

lower

Lower bound estimate with CI

upper

Upper bound estimate with CI

Details

The sensitivity analysis follows the weight-shift framework where observation weights can be multiplied by a factor between 1/gamma and gamma. The lower and upper bounds represent the extremes of the PATE estimate under this weight perturbation.

Requires the CVXR package for optimization.

See also

general_BART(), general_BART_overlap()

Examples

if (FALSE) { # \dontrun{
fit <- general_BART(princebart_fit, newdata, ...)
sens <- general_BART_transportability(fit, gamma = 1.5)
} # }