Simulate Synthetic Data for Sharp or Fuzzy RDD Examples

A convenience generator that mirrors the data-generating processes used in the rdpartial documentation and unit tests. The function produces a data frame ready for bounds_sharp(), bounds_fuzzy(), and bootstrap_bounds(), while also returning the true treatment effect at the cutoff for quick sanity checks.

Usage

simulate_rdd_data(
  n = 4000,
  cutoff = 16,
  dist = c("poisson", "uniform"),
  lambda = 15,
  range = c(8L, 25L),
  beta = c(0, 0.1, 0.002),
  tau = 1,
  sigma = 1,
  design = c("fuzzy", "sharp"),
  p0 = 0.2,
  p1 = 0.8,
  manip_width = 0,
  manip_prob = 0,
  seed = NULL
)

Arguments

n: Integer sample size (≥ 1).
cutoff: Numeric threshold. Must be integer-aligned with the support of x.
dist: Either "poisson" (default) or "uniform".
lambda: Poisson mean when dist = "poisson".
range: Integer length-2 support when dist = "uniform".
beta: Numeric length-3 vector for the quadratic mean of Y(0).
tau: Constant treatment effect added to Y(1) (default 1).
sigma: Error SD for both potential outcomes (default 1).
design: "fuzzy" (default) or "sharp".
p0, p1: Treatment probabilities below / above cutoff when design = "fuzzy".
manip_width: Numeric ≥ 0 - interval width subject to manipulation just below the cutoff.
manip_prob: Probability of manipulation for units in that interval (default 0).
seed: Optional integer for reproducibility.

Value

A data.frame with columns:

x - running variable after manipulation.
y - realised outcome.
z - treatment status.
x_orig - running variable before manipulation.
manipulated - logical flag.

The object carries attributes:

ate_cutoff - true treatment effect at the cutoff (tau).
call - original function call.

Outline of the DGP

Draw an integer running variable x from either a Poisson or discrete uniform distribution.
(Optional) Manipulate units with x in (cutoff − manip_width, cutoff) above the threshold with probability manip_prob.
Generate potential outcomes Y(0) and Y(1) as quadratic functions of x plus Normal noise.
Assign treatment either sharply (z = 1(x >= cutoff)) or fuzzily via Bernoulli probabilities p0, p1 below / above the cutoff.
Observe realised outcome y = Y(z).

Examples

set.seed(42)
sim <- simulate_rdd_data(n = 2000, cutoff = 16, design = "fuzzy",
                         manip_width = 0.4, manip_prob = 0.25)
head(sim)
#>    x         y z x_orig manipulated
#> 1 20 4.1729792 1     20       FALSE
#> 2 12 1.2726848 0     12       FALSE
#> 3 15 3.1496858 0     15       FALSE
#> 4 10 0.6554904 0     10       FALSE
#> 5 14 1.9406387 0     14       FALSE
#> 6 20 2.6233708 1     20       FALSE
#> 
#> True ATE at cutoff: 1 
attr(sim, "ate_cutoff")
#> [1] 1