Partially Linear IV (PLIV) / Local Average Treatment Effect

Native 2SLS in base R: first stage D ~ Z + X, second stage Y ~ D_hat + X. Cross-validated against DoubleML in the package's cross tests but not calling it at runtime.

Usage

estimate_pliv(
  data,
  treatment,
  outcome,
  instrument,
  covariates,
  n_folds = 5L,
  random_state = 42L
)

Arguments

data: Data frame with all required columns.
treatment: Endogenous treatment column name.
outcome: Outcome column name.
instrument: Instrument column name.
covariates: Exogenous covariate column names.
n_folds: Cross-fitting folds (DoubleML path). Default 5.
random_state: RNG seed. Default 42.

Value

Named list with late, se, ci_lower, ci_upper, pval, n_obs, method.

Examples

set.seed(1)
n <- 80
x1 <- rnorm(n)
x2 <- rnorm(n)
z <- rbinom(n, 1, 0.5)
d <- as.integer(plogis(0.3 + 0.8 * z + 0.4 * x1) > runif(n))
y <- 1 + 0.5 * d + 0.3 * x1 + rnorm(n)
df <- data.frame(y, d, z, x1, x2)
res <- suppressWarnings(estimate_pliv(df, "d", "y", "z", c("x1", "x2")))
res$late
#> [1] 4.354738