iRBCrd 2.1 -- October 29, 2026

Sourceable R functions for improved robust bias correction in regression
discontinuity designs.

New in 2.1:
- Confidence intervals use level (a percentage, default 95), replacing alpha.
  The numeric deriv argument replaces kink: 0 for levels, 1 for kinks.
  The former argument names are no longer accepted. Returned results also
  use level and deriv. Estimation and bootstrap formulas are unchanged.
- Bootstrap draws and per-draw diagnostics are not retained by default;
  request boot.keep = TRUE to retain them.
- Independent fixed/bootstrap choices for fuzzy and covariate combination
  weights in every applicable level and kink design. Fixed weights remain
  the default. Covariate-adjusted fuzzy designs support all four combinations.
- Simplified iRBCrd() interface: h and kernel accept a scalar or a left/right
  vector; fuzzy is the treatment input; deriv is 0 or 1. No separate
  side-specific input overrides. Returned kernel fields use the same spelling.
- Streamlined returned results, including nested component objects. h reports
  requested bandwidths; h.minus and h.plus retain effective cutoff bandwidths.
- deriv = 1 adds sharp, fuzzy, covariate-adjusted sharp, and covariate-adjusted
  fuzzy kink designs through iRBCrd(). Default kink polynomial order is p = 2.
- Slope-targeted iRBC bias correction and analytic covariance; bootstrap data
  and residuals continue to use local regression levels.
- Finite-sample kink Q and numerically evaluated population kink Q, including
  higher polynomial orders, with independent kernel-moment checks.
- Existing HC residual choices, side-specific bandwidths and kernels,
  regularization, and wild bootstrap weight choices for kink RD.
- Kink-specific fast standard errors use the appropriate slope-kernel ratios.
- Independent matrix formulas and brute-force bootstrap refits test all four
  kink designs. Existing level-RD implementations and results are preserved.

Retained features:

- Unified iRBCrd() entry point for sharp, fuzzy, covariate-adjusted sharp,
  and covariate-adjusted fuzzy RD.
- Analytic and bootstrap inference.
- Finite-sample and asymptotic Q adjustments.
- iRBC-HC0 through HC3 and CCT-HC0 through HC3 residuals.
- Rademacher, Mammen, and normal bootstrap weights.
- Scalar or side-specific bandwidths and kernels, with optional bandwidth
  regularization.
- Reproducible examples and standalone checks.

See README.txt for arguments and computational details. These files
are loaded with source(); this distribution is not an installable R package.
