Estimates \(\tau = \alpha_+ - \alpha_-\), the difference of the boundary intercepts of two kernel-weighted linear fits, one on each side of the cutoff (Hahn, Todd and van der Klaauw 2001; Imbens and Lemieux 2008, Sec. 4). Local LINEAR rather than local constant fitting is what removes the first-order boundary bias (Fan and Gijbels 1996, Ch. 3).
Arguments
- x
Running variable.
- y
Outcome.
- cutoff
Threshold, default 0.
- h
Bandwidth;
NULL(default) usesmorie_causrddh, the Imbens-Kalyanaraman plug-in.- kernel
"triangular"(default, the edge kernel that the IK constant is derived for) or"uniform"; both routes the sources give are available.
Value
A list with estimate (\(\tau\)), se,
ci, intercept_left, intercept_right,
slope_left, slope_right, h, kernel,
n_left, n_right, n_used, se_note,
method.
References
Hahn, J., Todd, P. and van der Klaauw, W. (2001). Identification and estimation of treatment effects with a regression-discontinuity design. Econometrica, 69(1), 201-209.
Examples
set.seed(1)
x <- runif(200, -1, 1)
y <- 0.5 * x + (x >= 0) * 0.3 + rnorm(200) * 0.1
morie_causrdd(x, y)
#> $estimate
#> [1] 0.2520205
#>
#> $se
#> [1] 0.03217331
#>
#> $ci
#> [1] 0.1889620 0.3150791
#>
#> $intercept_left
#> [1] 0.0019774
#>
#> $intercept_right
#> [1] 0.2539979
#>
#> $slope_left
#> [1] 0.4055003
#>
#> $slope_right
#> [1] 0.5978381
#>
#> $h
#> [1] 0.5324359
#>
#> $kernel
#> [1] "triangular"
#>
#> $n_left
#> [1] 61
#>
#> $n_right
#> [1] 56
#>
#> $n_used
#> [1] 117
#>
#> $se_note
#> [1] "HC0 sandwich per side, sides independent; rdrobust vce='hc0' convention"
#>
#> $method
#> [1] "sharp RDD, one-sided local linear fits at the cutoff"
#>