tau_FRD = (outcome jump)/(treatment jump), each from
Causrdd with its own IK bandwidth (Imbens-Lemieux
2008 separate-bandwidth prescription per IK Section 5.1). Delta-
method se with independent jumps (covariance term omitted and
documented). Reduces exactly to the sharp estimate when the first
stage jumps 0 to 1.
Usage
Causrddf(
x,
y,
treat,
cutoff = 0,
h = NULL,
h_treat = NULL,
kernel = "triangular"
)
Arguments
- x
Running variable.
- y
Outcome.
- treat
Treatment received (0/1 or probability).
- cutoff
Threshold c.
- h
Outcome bandwidth; IK plug-in when NULL.
- h_treat
First-stage bandwidth; IK plug-in when NULL.
- kernel
"triangular" or "uniform".
Value
List with estimate, se, ci,
jump_outcome, jump_treatment, se_outcome,
se_treatment, h_outcome, h_treatment,
kernel, method.
References
Hahn, J., Todd, P. and van der Klaauw, W. (2001),
Econometrica 69(1), 201-209 (estimand); Imbens and Kalyanaraman
(2009), NBER w14726, Section 5.1 (implemented text, in the local
registry); Imbens and Lemieux (2008), J Econometrics 142(2),
615-635.
Examples
set.seed(1)
x <- runif(300, -1, 1)
treat <- rbinom(300, 1, ifelse(x >= 0, 0.85, 0.15))
y <- 1 + 2 * x + 0.6 * treat + rnorm(300, 0, 0.3)
r <- Causrddf(x, y, treat, cutoff = 0)
is.list(r)
#> [1] TRUE
