Covariate-balancing propensity score (Imai & Ratkovic 2014)

Native just-identified CBPS: solves the covariate-balance moment conditions \(E[(T - p(X)) X / (p(X)(1-p(X)))] = 0\) directly (the exactly-identified estimator), via Newton iterations on the logistic index. Balance is thus built into the score rather than checked after the fact.

Usage

morie_weight_cbps(
  data,
  treatment,
  covariates,
  estimand = "ATE",
  max_iter = 100L,
  tol = 1e-10
)

Arguments

data

Data frame.

treatment

Binary treatment column (0/1).

covariates

Covariate column names.

estimand

"ATE" (default), "ATT", or "ATC".

max_iter

Maximum damped-Newton iterations. Default 100.

tol

Convergence tolerance on the balance-moment norm. Default 1e-10.

Value

A morie_weight object.

References

Imai & Ratkovic (2014) JRSS-B 76(1).

Examples

d <- data.frame(t = rbinom(120, 1, 0.5), x = rnorm(120))
morie_weight_cbps(d, "t", "x")
#> morie_weight: CBPS (Imai-Ratkovic 2014, exact) (estimand ATE)
#>   n = 120  ESS = 118.7  weight range [1.671, 2.493]