feat: RegressionDiscontinuity - sharp RD estimator with rdrobust 4.0.0 parity (RDD PR-2) (#684)

* feat: RegressionDiscontinuity - sharp RD estimator with rdrobust parity

Estimation-path port + public estimator + docs sweep (PR-2 of 2).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QGca52n6H8oDDXALjjrsp4

* fix: run mass-point detection before the bandwidth branch + reject bools on integer knobs

rdrobust.R:365-380 runs the mass-point check in rdrobust() itself, ahead of
the manual-vs-data-driven bandwidth split, so the warning must fire on
manual-h fits too (verified on installed 4.0.0). fit() now performs the
detection and the port's rdbwselect-level warning is suppressed on the
estimator path to mirror R's single warning per fit. Integer knobs
(p/q/nnmatch/bwcheck) now reject bool (int subclass) inputs.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QGca52n6H8oDDXALjjrsp4

* chore: drop stale mypy ignores after rebase onto #683 (blocking mypy job)

The type: ignore[assignment] comments on the beta_p_* dataclass fields are
unused under the pinned CI mypy config, which now fails on unused ignores
(warn_unused_ignores = true).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QGca52n6H8oDDXALjjrsp4

---------

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

Author: igerber

CHANGELOG.md

@@ -8,6 +8,31 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## [Unreleased]
 
 ### Added
+- **`RegressionDiscontinuity` - sharp regression discontinuity estimation with robust
+  bias-corrected inference (alias `RDD`).** Local-polynomial sharp RD per Calonico,
+  Cattaneo & Titiunik (2014), parity-targeting R `rdrobust` 4.0.0 end-to-end: all 10
+  data-driven bandwidth selectors (`mserd` default, `msetwo`/`msesum`/comb and the
+  CER-optimal variants), triangular/epanechnikov/uniform kernels, `masspoints`
+  adjust/check/off, manual `h`/`b`/`rho` with R-exact resolution semantics (including
+  `rho`-without-`h` applying to selected bandwidths and the unconditional N<20
+  full-range fallback), and the three-row Conventional / Bias-Corrected / Robust
+  output. **Canonical binding:** `att`/`se`/`t_stat`/`p_value`/`conf_int` are ONE
+  coherent row - the robust bias-corrected row (`att = tau_bc`, CI centered on it,
+  `t_stat == att/se`), preserving the library-wide field identities; rdrobust's
+  printed headline coefficient is exposed as `att_conventional` with a full inference
+  row, and `summary()` prints the familiar three-row table. Estimation-path port
+  (`rdrobust_fit_sharp`: Q_q bias-correction score matrix, conventional/robust NN
+  sandwiches) validated against a new estimates golden
+  (`benchmarks/data/rdrobust_estimates_golden.json`, 16 configurations incl. the
+  Senate anchors) at rtol=1e-9 in `tests/test_rdd_parity.py`; R-free methodology
+  anchors (CCT 2014 Remark 7 bias-corrected == local-quadratic equivalence at rel
+  1e-10 across all kernels, invariances, joint-NaN degenerate contracts) in
+  `tests/test_rdd_methodology.py`; API/validation suite in `tests/test_rdd.py`.
+  Deviations from R (each labeled in the REGISTRY section): warn-instead-of-silent
+  NaN drops, warn-and-ignore `b`-without-`h`, fail-closed targeted errors on
+  degenerate designs, and the canonical-binding note above. Sharp designs only:
+  fuzzy RD, covariates (CCFT 2019 - review on file), cluster-robust variance,
+  weights, kink estimands, and rdplot/density diagnostics are documented follow-ups.
 - **Internal: mypy enforced at zero errors.** Triaged the 184 pre-existing
   `mypy diff_diff` errors to an enforceable zero and added a blocking Mypy job to
   the Lint CI workflow (pinned `mypy==2.1.0` + pinned numpy/pandas/scipy for stub
@@ -24,7 +49,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   already-stale ignores were removed. Tightening tracked in TODO.md. No public
   API or numerical behavior change.
 - **Internal: rdrobust sharp-RD bandwidth-selection port (`diff_diff/_rdrobust_port.py`).**
-  Step-1 machinery for the upcoming `RegressionDiscontinuity` estimator: a faithful
+  Step-1 machinery for the `RegressionDiscontinuity` estimator above: a faithful
   pure-Python NumPy/SciPy port of R `rdrobust` 4.0.0's `rdbwselect` sharp-RD path (all 10 data-driven
   selectors - mserd default, msetwo/msesum/comb and the CER-optimal variants -
   triangular/epanechnikov/uniform kernels, `masspoints` adjust/check/off, IK-style
@@ -36,10 +61,10 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   Cattaneo-Frandsen-Titiunik 2015) validated at rtol=1e-9 across 17 configurations x 10
   selectors in `tests/test_rdrobust_port.py` (kernels, p/q orders, deriv, masspoints
   modes, stdvars, bwrestrict, scaleregul, nnmatch), including the 2017 Stata Journal
-  published Senate anchors under `masspoints="off"`. **No public API change** - the module is private;
-  the estimator, robust bias-corrected inference, and docs surfaces land in the follow-up PR.
-  Port-level deviations documented in `docs/methodology/REGISTRY.md` (new
-  RegressionDiscontinuity section stub).
+  published Senate anchors under `masspoints="off"`. The module is private; the public
+  estimator, robust bias-corrected inference, and docs surfaces are the
+  `RegressionDiscontinuity` entry above. Port-level deviations documented in
+  `docs/methodology/REGISTRY.md` (RegressionDiscontinuity section).
 - **Internal: ungated `Lint` CI workflow.** Check-only `ruff check` + `black --check`
   (at the versions pinned in the `dev` extra) run on every PR push and on pushes to
   main — no `ready-for-ci` label needed; the aggregate `Lint Gate` job is the single
@@ -65,6 +90,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   `benchmarks/data/qte_golden.json`.
 
 ### Changed
+- `diff_diff/guides/llms-autonomous.txt` no longer lists regression discontinuity as
+  out of scope: sharp RD routes to `RegressionDiscontinuity`; only fuzzy RD is
+  referred to external tooling.
 - **Internal: repo-wide lint normalization + pinned tooling.** black/ruff/mypy are now
   pinned exactly in the `dev` extra (`black==26.3.1`, `ruff==0.15.13`, `mypy==2.1.0`;
   the tools require Python >= 3.10 — the library floor stays 3.9); full `black` +

README.md

@@ -112,6 +112,7 @@ Full guide: `diff_diff.get_llm_guide("practitioner")`.
 - [TripleDifference](https://diff-diff.readthedocs.io/en/stable/api/triple_diff.html) - triple difference (DDD) estimator for designs requiring two criteria for treatment eligibility
 - [ContinuousDiD](https://diff-diff.readthedocs.io/en/stable/api/continuous_did.html) - Callaway, Goodman-Bacon & Sant'Anna (2024) continuous treatment DiD with dose-response curves
 - [HeterogeneousAdoptionDiD](https://diff-diff.readthedocs.io/en/stable/api/had.html) - de Chaisemartin, Ciccia, D'Haultfœuille & Knau (2026) for designs where **no unit remains untreated**; local-linear estimator at the dose support boundary returning Weighted Average Slope (WAS) on Design 1' (`d̲ = 0` / QUG) or `WAS_{d̲}` on Design 1 (`d̲ > 0`, continuous-near-d̲ or mass-point), with a multi-period event-study extension (last-treatment cohort, pointwise CIs). **Panel-only** in this release - repeated cross-sections rejected by the validator. Alias `HAD`.
+- [RegressionDiscontinuity](https://diff-diff.readthedocs.io/en/stable/api/regression_discontinuity.html) - Calonico, Cattaneo & Titiunik (2014) sharp regression discontinuity with robust bias-corrected inference and rdrobust-parity bandwidth selection (all 10 selectors, mass-point handling). **Sharp designs only** in this release; canonical `att` is the bias-corrected estimate with a coherent robust CI (rdrobust's printed headline is `att_conventional`). Alias `RDD`.
 - [StackedDiD](https://diff-diff.readthedocs.io/en/stable/api/stacked_did.html) - Wing, Freedman & Hollingsworth (2024) stacked DiD with Q-weights and sub-experiments; optional covariate balancing (Ustyuzhanin 2026)
 - [EfficientDiD](https://diff-diff.readthedocs.io/en/stable/api/efficient_did.html) - Chen, Sant'Anna & Xie (2025) efficient DiD with optimal weighting for tighter SEs
 - [TROP](https://diff-diff.readthedocs.io/en/stable/api/trop.html) - Triply Robust Panel estimator (Athey et al. 2025) with nuclear norm factor adjustment

benchmarks/R/generate_rdrobust_estimates_golden.R

@@ -0,0 +1,136 @@
+# Golden-value generator for the diff-diff sharp-RD ESTIMATION port
+# (diff_diff/_rdrobust_port.py::rdrobust_fit_sharp and the public
+# RegressionDiscontinuity estimator).
+#
+# Deliberately a SEPARATE file/JSON from generate_rdrobust_golden.R so the
+# bandwidth fixtures reviewed in the machinery PR are never regenerated.
+# Same parity target: CRAN rdrobust 4.0.0 (tarball sha256
+# 78f0d6b4bdec4091cc8f42f6f1598704747f95926446d3aaee381ea1d613a36f);
+# the GitHub 4.1.0-dev tree must NOT be installed when regenerating.
+#
+# Outputs benchmarks/data/rdrobust_estimates_golden.json: per config the
+# embedded inputs (17 significant digits), selected/echoed bandwidths, the
+# full three-row (Conventional / Bias-Corrected / Robust) coef/se/z/pv/ci
+# block, effective counts, and per-side order-p coefficient vectors.
+#
+# Run from the repo root: Rscript benchmarks/R/generate_rdrobust_estimates_golden.R
+
+library(rdrobust)
+library(jsonlite)
+
+stopifnot(packageVersion("rdrobust") == "4.0.0")
+
+TARBALL_SHA256 <- "78f0d6b4bdec4091cc8f42f6f1598704747f95926446d3aaee381ea1d613a36f"
+
+run_estimate <- function(y, x, c = 0, masspoints = "adjust", kernel = "tri",
+                         p = 1, q = 2, h = NULL, b = NULL, rho = NULL,
+                         level = 95, bwselect = "mserd") {
+  args <- list(y = y, x = x, c = c, masspoints = masspoints, kernel = kernel,
+               p = p, q = q, level = level, bwselect = bwselect)
+  if (!is.null(h)) args$h <- h
+  if (!is.null(b)) args$b <- b
+  if (!is.null(rho)) args$rho <- rho
+  r <- suppressWarnings(do.call(rdrobust, args))
+  list(
+    c = c, masspoints = masspoints, kernel = kernel, p = p, q = q,
+    bwselect = bwselect,
+    h_in = if (is.null(h)) NA else h,
+    b_in = if (is.null(b)) NA else b,
+    rho_in = if (is.null(rho)) NA else rho,
+    level = level,
+    h_l = r$bws["h", "left"], h_r = r$bws["h", "right"],
+    b_l = r$bws["b", "left"], b_r = r$bws["b", "right"],
+    tau_cl = unname(r$coef[1, 1]), tau_bc = unname(r$coef[2, 1]),
+    se_cl = unname(r$se[1, 1]), se_rb = unname(r$se[3, 1]),
+    z = unname(r$z[, 1]), pv = unname(r$pv[, 1]),
+    ci_lower = unname(r$ci[, 1]), ci_upper = unname(r$ci[, 2]),
+    N = unname(r$N), N_h = unname(r$N_h), N_b = unname(r$N_b),
+    bias = unname(r$bias),
+    beta_p_l = unname(as.vector(r$beta_Y_p_l)),
+    beta_p_r = unname(as.vector(r$beta_Y_p_r))
+  )
+}
+
+golden <- list()
+
+golden$metadata <- list(
+  rdrobust_version = as.character(packageVersion("rdrobust")),
+  rdrobust_tarball_sha256 = TARBALL_SHA256,
+  seeds = list(dgp_lee_smooth = 42L, dgp_ties_moderate = 123L,
+               dgp_asymmetric_scaled = 777L),
+  generator = "benchmarks/R/generate_rdrobust_estimates_golden.R",
+  algorithm = paste(
+    "rdrobust() sharp-RD estimation blocks (three-row coef/se/z/pv/ci,",
+    "counts, per-side beta_p) for the vce='nn' no-covariate path,",
+    "complementing the bandwidth fixtures in rdrobust_golden.json."
+  ),
+  r_version = R.version.string
+)
+
+# Same seeded DGPs as generate_rdrobust_golden.R (kept in sync by seed).
+set.seed(42)
+n <- 1000
+x1 <- 2 * rbeta(n, 2, 4) - 1
+y1 <- 0.48 + 1.27 * x1 + 7.18 * x1^2 + 20.21 * x1^3 + 21.54 * x1^4 +
+  7.33 * x1^5 + 0.04 * (x1 >= 0) + rnorm(n, sd = 0.1295)
+
+golden$dgp_lee_smooth <- list(
+  x = x1, y = y1,
+  configs = list(
+    default    = run_estimate(y1, x1),
+    manual_h   = run_estimate(y1, x1, h = 0.15),
+    h_rho2     = run_estimate(y1, x1, h = 0.15, rho = 2),
+    rho_only   = run_estimate(y1, x1, rho = 2),
+    p2q3       = run_estimate(y1, x1, p = 2, q = 3),
+    p0q1       = run_estimate(y1, x1, p = 0, q = 1),
+    epa        = run_estimate(y1, x1, kernel = "epa"),
+    uni        = run_estimate(y1, x1, kernel = "uni"),
+    level90    = run_estimate(y1, x1, level = 90),
+    msetwo     = run_estimate(y1, x1, bwselect = "msetwo"),
+    cercomb2   = run_estimate(y1, x1, bwselect = "cercomb2")
+  )
+)
+
+set.seed(123)
+n2 <- 800
+x2 <- round(2 * rbeta(n2, 2, 4) - 1, 2)
+y2 <- 0.5 + 0.8 * x2 + (x2 >= 0) * 1.0 + rnorm(n2, sd = 0.3)
+
+golden$dgp_ties_moderate <- list(
+  x = x2, y = y2,
+  configs = list(
+    adjust = run_estimate(y2, x2, masspoints = "adjust"),
+    off    = run_estimate(y2, x2, masspoints = "off")
+  )
+)
+
+set.seed(777)
+n3 <- 300
+x3 <- 40 * rbeta(n3, 5, 2)
+y3 <- 2 + 0.05 * x3 - 0.001 * x3^2 + 0.8 * (x3 >= 28) + rnorm(n3, sd = 0.5)
+
+golden$dgp_asymmetric_scaled <- list(
+  x = x3, y = y3,
+  configs = list(
+    default = run_estimate(y3, x3, c = 28)
+  )
+)
+
+senate_path <- "benchmarks/data/rdrobust_senate.csv"
+stopifnot(file.exists(senate_path))
+senate <- read.csv(senate_path)
+ok <- complete.cases(senate$vote, senate$margin)
+sv <- senate$vote[ok]
+sm <- senate$margin[ok]
+
+golden$senate <- list(
+  csv = "benchmarks/data/rdrobust_senate.csv",
+  configs = list(
+    adjust = run_estimate(sv, sm, masspoints = "adjust"),
+    off    = run_estimate(sv, sm, masspoints = "off")
+  )
+)
+
+out_path <- "benchmarks/data/rdrobust_estimates_golden.json"
+write_json(golden, out_path, auto_unbox = TRUE, pretty = TRUE, digits = I(17))
+cat("Wrote", out_path, "\n")

benchmarks/data/rdrobust_estimates_golden.json

@@ -206,6 +206,10 @@
     TreatmentDoseShape,
     profile_panel,
 )
+from diff_diff.rdd import (
+    RegressionDiscontinuity,
+    RegressionDiscontinuityResults,
+)
 from diff_diff.results import (
     DiDResults,
     MultiPeriodDiDResults,
@@ -308,6 +312,7 @@
 DCDH = ChaisemartinDHaultfoeuille
 HAD = HeterogeneousAdoptionDiD
 CiC = ChangesInChanges
+RDD = RegressionDiscontinuity
 
 __version__ = "3.7.0"
 __all__ = [
@@ -508,6 +513,10 @@
     "HeterogeneousAdoptionDiDResults",
     "HeterogeneousAdoptionDiDEventStudyResults",
     "HAD",
+    # RegressionDiscontinuity (sharp RD, rdrobust parity)
+    "RegressionDiscontinuity",
+    "RegressionDiscontinuityResults",
+    "RDD",
     # HeterogeneousAdoptionDiD pre-test diagnostics (Phase 3)
     "qug_test",
     "stute_test",