eif_additive: Estimate treatment effects under additivity using the...

In michaelpollmann/parTreat: Efficiently Estimate Treatment Effects Based On A Parametric Model For Treatment Effects

Description

Efficiently estimate the treatment effect under an additive model based on the influence function. Finds the parameter θ such that

\int f_{0}'(x)/f_{0}(x) f_{0}(x) dx = \int f_{0}'(y-θ)/f_{0}(y-θ) f_{1}(y) d y (*)

and estimates the standard error of \hat{θ} (under correct specification). Beware of warnings (about failures to find a solution) coming from pracma::fzero. These are currently not handled properly.

Usage

eif_additive(
  X,
  Y,
  xf = NULL,
  d1_logf = NULL,
  d2_logf = NULL,
  theta_init = NULL,
  calc_se = TRUE,
  ...
)

Arguments

X	numeric vector, the outcomes of the control observations.
Y	numeric vector, the outcomes of the treated observations.
xf	numeric vector, points at which the density is given in d1_logf (and d2_logf). If xf is NULL but d1_logf is supplied, it is assumed that the density was estimated at each points in X.
d1_logf	numeric vector, (estimates of) the first derivative of the log density of the control observations X. If d1_logf is null, the density (and its derivatives) of X is estimated using estimate_density_d_logs.
d2_logf	numeric vector, (estimates of) the second derivative of the log density of the control observations X. Used to estimate standard errors. If d2_logf is null but calc_se is TRUE (default), the density (and its derivatives) of X is estimated using estimate_density_d_logs.
theta_init	numeric, initial value to find the solution to (*). Defaults to the difference in medians.
calc_se	logical, should standard errors be calculated? (default = TRUE) This is slower if d1_logf but not d2_logf is provided as then the density needs to be estimated for standard errors but not for the point estimate.
...	additional arguments passed on estimate_density_d_logs for density estimation.

Value

list of two elements if calc_se = TRUE or one element otherwise:

tau	the point estimate of the treatment effect
se	(if calc_se = TRUE) the estimated standard error

Examples

# draw a random sample with additive treatment effect
X <- rexp(n=1000, rate=2)
Y <- 0.5 + rexp(n=200, rate=2)
eif_additive(X,Y)

michaelpollmann/parTreat documentation built on Dec. 21, 2021, 5:58 p.m.