RDD Robustness Checks

Performs robustness checks for regression discontinuity design by varying key specification choices including bandwidth, polynomial order, and kernel.

Usage

rdd_robustness_checks(
  rdd_results,
  data,
  bandwidth_multipliers = c(0.5, 0.75, 1.25, 1.5, 2),
  polynomial_orders = c(1, 2, 3),
  kernels = c("triangular", "rectangular", "epanechnikov"),
  include_bias_correction = TRUE,
  placebo_cutoffs = c(-0.5, 0.5)
)

Arguments

rdd_results

Object of class "rdd_results" from regression_discontinuity()

data

Original data used in RDD estimation

bandwidth_multipliers

Numeric vector of multipliers for original bandwidth (default: c(0.5, 0.75, 1.25, 1.5, 2.0))

polynomial_orders

Integer vector of polynomial orders to test (default: c(1, 2, 3))

kernels

Character vector of kernel functions to test (default: c("triangular", "rectangular", "epanechnikov"))

include_bias_correction

Logical indicating whether to test with/without bias correction (default: TRUE)

placebo_cutoffs

Numeric vector of placebo cutoffs relative to true cutoff (default: c(-0.5, 0.5) in bandwidth units)

Value

A list with the following components:

• bandwidth_robustness: Results varying bandwidth

• polynomial_robustness: Results varying polynomial order

• kernel_robustness: Results varying kernel

• bias_correction_robustness: Results with/without bias correction

• placebo_tests: Results for placebo cutoffs

• summary_table: Summary table of all robustness checks

Details

This function systematically varies key RDD specification choices to assess the robustness of the main results. It tests:

• Different bandwidth choices (multiples of optimal bandwidth)

• Different polynomial orders (local linear vs. quadratic vs. cubic)

• Different kernel functions (triangular, rectangular, Epanechnikov)

• Bias correction on/off

• Placebo cutoffs where no effect should be present

Examples

if (FALSE) { # \dontrun{
# Run main RDD
rdd_results <- regression_discontinuity(
  data = employment_data,
  outcome = "employment_rate",
  running_variable = "age",
  cutoff = 65
)

# Robustness checks
robustness <- rdd_robustness_checks(
  rdd_results = rdd_results,
  data = employment_data,
  bandwidth_multipliers = c(0.5, 1.0, 1.5, 2.0)
)
} # }