HAD Phase 4: trends_lin (Eq 17/18) + R-package end-to-end parity

CHANGELOG.md

@@ -7,6 +7,10 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Added
+- **HAD `trends_lin=True` linear-trend detrending mode** on `HeterogeneousAdoptionDiD.fit(aggregate="event_study")`, `joint_pretrends_test`, and `joint_homogeneity_test`. Mirrors R `DIDHAD::did_had(..., trends_lin=TRUE)` (paper Eq. 17 / Eq. 18 / page 32 joint-Stute homogeneity-with-trends). Per-group linear-trend slope estimated as `Y[g, F-1] - Y[g, F-2]` and applied as `(t - base) × slope` adjustment to per-event-time outcome evolutions. Requires F ≥ 3 (panel must contain F-2). The "consumed" placebo at our event-time `e=-2` is auto-dropped (R reduces max placebo lag by 1 with the same effect). Mutually exclusive with survey weighting (`survey_design` / `survey` / `weights`): raises `NotImplementedError` per `feedback_per_method_survey_element_contract` (weighted slope estimator not derived from paper; tracked in TODO.md as a follow-up). Bit-exact backcompat for `trends_lin=False` (default). Patch-level (additive keyword-only kwarg).
+- **HAD R-package end-to-end parity test** vs `DIDHAD` v2.0.0 (`Credible-Answers/did_had`) on the **`design="continuous_at_zero"` (Design 1') surface**. New parity fixture `benchmarks/data/did_had_golden.json` generated by `benchmarks/R/generate_did_had_golden.R` covers 3 paper-derived synthetic DGPs (Uniform, Beta(2,2), Beta(0.5,1)) × 5 method combinations (overall, event-study, placebo, yatchew, trends_lin). The harness explicitly forces `HeterogeneousAdoptionDiD(design="continuous_at_zero")` because R `did_had` always evaluates the local-linear at `d=0` regardless of dose distribution; our default `design="auto"` may legitimately choose `continuous_near_d_lower` or `mass_point` on dose distributions with boundary density bounded away from zero (e.g., Beta(2,2)) and thereby diverge from R numerically — that divergence is methodologically defensible but out of scope for this parity test. Python parity test `tests/test_did_had_parity.py` asserts point estimate / SE / CI bounds at `atol=1e-8` and Yatchew T-stat at `atol=1e-10` after a documented `× G/(G-1)` finite-sample convention shift. Two intentional convention deviations from R, documented in `docs/methodology/REGISTRY.md`: (a) we report the bias-corrected point estimate (modern CCF 2018 convention; R's `Estimate` column reports the conventional estimate with the bias-corrected CI separately — our `att` matches R's CI midpoint); (b) Yatchew uses paper Appendix E's literal (1/G) variance-denominator convention while R uses base-R `var()`'s (1/(N-1)) sample-variance convention (parity is bit-exact after the `× G/(G-1)` shift). Yatchew on placebos with R's mean-independence null (`order=0`) is not yet exposed in our `yatchew_hr_test` (we currently only support the linearity null) and is skipped in the parity test; tracked as TODO follow-up.
+
 ### Changed
 - **Rust dependency upgrades**: bumped `rand` 0.8 → 0.10 and `rand_xoshiro` 0.6 → 0.8 in the Rust backend (the two crates are coupled through `rand_core` and must move together). MSRV bumped from Rust 1.84 → 1.85 to satisfy the new dependency requirements. Three call sites in `rust/src/bootstrap.rs` updated for the `rand 0.9` API rename: `gen::<bool>()` → `random::<bool>()`, `gen::<f64>()` → `random::<f64>()`, `gen_range(0..6)` → `random_range(0..6)`. **Webb wild bootstrap byte stream shifted** as a side effect: `rand 0.9` reworked the internal algorithm for `random_range` (improved rejection sampling), so `Xoshiro256PlusPlus::seed_from_u64(seed)` followed by `random_range(0..6)` consumes RNG bytes differently than the old `gen_range(0..6)` did. Distributional properties of Webb weights are unchanged (still uniform over the 6-point support); aggregate inference (SE, p-values, CI) converges to the same values for any reasonable `n_bootstrap`. Rademacher and Mammen byte streams are bit-identical to the prior release. Anyone with a saved Rust+Webb baseline pinning specific seeded results will see different numbers; the regression test suite uses within-build seed-reproducibility (not cross-version baselines) so all internal tests pass unchanged. New regression guard `TestRustBackend::test_bootstrap_weights_bit_identity_snapshot` pins fixed-seed weights for all three weight types, so any future RNG drift fails loudly with a localized error message.
 

TODO.md

@@ -101,7 +101,10 @@ Deferred items from PR reviews that were not addressed before merge.
 | `HeterogeneousAdoptionDiD` mass-point: `vcov_type in {"hc2", "hc2_bm"}` raises `NotImplementedError` pending a 2SLS-specific leverage derivation. The OLS leverage `x_i' (X'X)^{-1} x_i` is wrong for 2SLS; the correct finite-sample correction uses `x_i' (Z'X)^{-1} (...) (X'Z)^{-1} x_i`. Needs derivation plus an R / Stata (`ivreg2 small robust`) parity anchor. | `diff_diff/had.py::_fit_mass_point_2sls` | Phase 2a | Medium |
 | `HeterogeneousAdoptionDiD` survey-design API consolidation, **next minor bump**: drop the deprecated `survey=` and `weights=` kwargs on all 8 HAD surfaces (`HeterogeneousAdoptionDiD.fit`, `did_had_pretest_workflow`, `qug_test`, `stute_test`, `yatchew_hr_test`, `stute_joint_pretest`, `joint_pretrends_test`, `joint_homogeneity_test`); only `survey_design=` remains. Also fold the legacy back-end `weights=` paths (e.g. `_aggregate_unit_weights` ad-hoc routing) into the unified `_resolve_survey_for_fit`-driven path. The `_make_trivial_resolved` underscore alias on `survey.py` stays (one-line, harmless). DeprecationWarning ships in this PR; the removal PR is ~50 LoC of cleanup. | `diff_diff/had.py`, `diff_diff/had_pretests.py` | next minor bump | Medium |
 | `HeterogeneousAdoptionDiD` continuous paths: thread `cluster=` through `bias_corrected_local_linear` (Phase 1c's wrapper already supports cluster; Phase 2a ignores it with a `UserWarning` on the continuous path to keep scope tight). | `diff_diff/had.py`, `diff_diff/local_linear.py` | Phase 2a | Low |
-| `HeterogeneousAdoptionDiD` Eq 18 linear-trend detrending (Pierce-Schott style): the joint-Stute infrastructure shipped in the Phase 3 follow-up supports pre-trends (mean-indep) and post-homogeneity (linearity) nulls. The Pierce-Schott application (paper Section 5.2) uses a LINEAR-TREND detrending of pre-period outcomes before the joint CvM — `Y_{g,t} - Y_{g,t_anchor} - (t - t_anchor)*(Y_{g,t_anchor} - Y_{g,t_anchor-1})` — reaching p=0.51 on US-China tariff data. Extends `joint_pretrends_test` with a detrending mode or a separate Eq 18-specific helper. Deferred to Phase 4 replication harness (where the published p=0.51 serves as the parity anchor). | `diff_diff/had_pretests.py::joint_pretrends_test` | Phase 4 | Medium |
+| `HeterogeneousAdoptionDiD` Eq 17 / Eq 18 linear-trend detrending: SHIPPED in PR #389 (Phase 4 R-parity, 2026-04). Exposed as `trends_lin: bool = False` keyword-only kwarg on `HeterogeneousAdoptionDiD.fit(aggregate="event_study")`, `joint_pretrends_test`, `joint_homogeneity_test`. Mirrors R `DIDHAD::did_had(..., trends_lin=TRUE)`. Pierce-Schott published-number parity (paper p=0.51 / p=0.40) deferred indefinitely (LBD-restricted analysis panel); replaced by end-to-end R-package parity at `tests/test_did_had_parity.py`. | `diff_diff/had_pretests.py::joint_pretrends_test`, `diff_diff/had.py` | Phase 4 (shipped) | Done |
+| `HeterogeneousAdoptionDiD` `trends_lin × survey_design` follow-up: per-group linear-trend slope under survey weighting (weighted slope estimator? per-PSU slope?) is not derived from the paper. PR #389 raises `NotImplementedError` on the combination across all 3 trends_lin surfaces. If user demand emerges, derive the weighted variant and lift the gate. | `diff_diff/had.py::HeterogeneousAdoptionDiD.fit`, `diff_diff/had_pretests.py::joint_pretrends_test`, `diff_diff/had_pretests.py::joint_homogeneity_test` | follow-up | Low |
+| `HeterogeneousAdoptionDiD` `yatchew_hr_test(null="mean_independence")` mode: R `YatchewTest::yatchew_test(order=0)` fits `Y ~ 1` (intercept-only baseline) and tests mean-independence of Y from D; R's `DIDHAD::did_had(yatchew=TRUE)` uses this on placebo rows ("non-parametric pre-trends test"). Our `yatchew_hr_test` always fits `Y ~ D` (linearity null) — no `null=` parameter exposed. Adding the mean-independence mode would (a) give practitioners a more conventional pre-trends test surface, and (b) close the PR #389 R-parity feature gap on the placebo-Yatchew rows (currently skipped in `tests/test_did_had_parity.py::TestYatchewParity` because the two tests are not the same statistic). | `diff_diff/had_pretests.py::yatchew_hr_test` | follow-up | Medium |
+| `HeterogeneousAdoptionDiD` Stute family Stata-bridge parity: PR #389 R-parity covers the full HAD fit + Yatchew surfaces but skips Stute family (`stute_test`, `stute_joint_pretest`, `joint_pretrends_test`, `joint_homogeneity_test`) because no R `Stutetest` package exists publicly (chaisemartinPackages publishes only the Stata `stute_test` module; the paper cites a 2024c R Stutetest module that is not on GitHub or CRAN). Stata-bridge parity would add `benchmarks/stata/generate_stute_golden.do` + a Stata installation requirement. Low priority unless user demand emerges. | `benchmarks/stata/`, `tests/test_stute_test_parity.py` | follow-up | Low |
 | `HeterogeneousAdoptionDiD` Phase 3 Stute performance: Appendix D vectorized matrix form replaces the per-iteration OLS refit with a single precomputed `M = I - X(X'X)^{-1}X'` applied to `eps * eta`. Functionally identical, ~2x faster. Shipped literal-refit form in Phase 3 to match paper text and keep reviewer surface small. | `diff_diff/had_pretests.py::stute_test` | Phase 3 | Low |
 | `HeterogeneousAdoptionDiD` Phase 3 R-parity: Phase 3 ships coverage-rate validation on synthetic DGPs (not tight point parity against `chaisemartin::stute_test` / `yatchew_test`). Tight numerical parity requires aligning bootstrap seed semantics and `B` across numpy/R and is deferred. | `tests/test_had_pretests.py` | Phase 3 | Low |
 | `HeterogeneousAdoptionDiD` Phase 3 nprobust bandwidth for Stute: some Stute variants on continuous regressors use nprobust-style optimal bandwidth selection. Phase 3 uses OLS residuals from a 2-parameter linear fit (no bandwidth selection). nprobust integration is a future enhancement; not in paper scope. | `diff_diff/had_pretests.py::stute_test` | Phase 3 | Low |