Estimate the Local Average Treatment Effect (LATE) via 2SLS / Wald

Uses a binary instrument $Z$ to identify the LATE (Imbens & Angrist, 1994): $$LATE = \frac{Cov(Y, Z)}{Cov(T, Z)}$$

Usage

morie_estimate_late(data, treatment, outcome, instrument, covariates = NULL)

Arguments

data: A data frame.
treatment: Name of the binary endogenous treatment column.
outcome: Name of the outcome column.
instrument: Name of the binary instrument column.
covariates: Optional character vector of exogenous covariates.

Value

Named list: late, se, ci_lower, ci_upper, first_stage_f, n.

Details

With covariates, the native 2SLS k-class engine is used (module 17; validated against ivreg in tests/cross/). Without covariates the closed-form Wald estimator and its delta-method SE are used.

References

Imbens GW, Angrist JD (1994). Identification and estimation of local average treatment effects. Econometrica, 62(2), 467-475.

Examples

set.seed(1)
n <- 300L
z <- rbinom(n, 1, 0.5)
t <- rbinom(n, 1, plogis(-0.2 + 1.5 * z))
y <- 0.8 * t + rnorm(n)
morie_estimate_late(data.frame(t = t, y = y, z = z), "t", "y", "z")
#> $late
#> [1] 0.3009515
#> 
#> $se
#> [1] 0.5737882
#> 
#> $ci_lower
#> [1] -0.8236734
#> 
#> $ci_upper
#> [1] 1.425576
#> 
#> $first_stage_f
#> [1] 32.12371
#> 
#> $n
#> [1] 300
#>