ConleySpatialHAC methodology-review-tracker promotion + Bertanha-Imbens 2014 citation correction

[igerber]

Summary

• ConleySpatialHAC methodology-review-tracker promotion: In Progress → Complete. Closes the Conley (1999) primary-source review on the methodology-review tracker. New tests/test_methodology_conley.py (~1600 LoC; 10 classes; 60 tests, 5 of them @pytest.mark.slow) with paper-equation-numbered Verified Components walk-through:

  • Eq. 4.2 cross-sectional sandwich (pairwise-distance specialization; Eq. 3.13 is the lattice-indexed form reserved for grid coordinates per conley-1999-review.md)
  • Eq. 4.2 HC0 + rank-1 limits + K(0)=1 diagonal contribution
  • Andrews (1991) HAC lag truncation matching conleyreg::time_dist.cpp
  • Haversine convention with Earth radius 6371.01 km
  • Phase 2 panel block-decomposed sandwich at atol=1e-12
  • Wave A Add git worktree management slash commands #120 sparse k-d-tree sparse-vs-dense bit-identity at atol=1e-10 (cross-sectional + panel + sparse-with-cluster paths)
  • R conleyreg v0.1.9 parity at atol=1e-6 on 6 fixtures (3 cross-sectional + 3 panel) + sparse-forced + time-asymmetric kernel literal-matching
  • Three deviations-area classes: TestConleyLibraryExtensions (Wave A Align TROP lambda conventions with paper (Athey et al. 2025) #119/Add git worktree management slash commands #120/Fix worktree-rm: detect squash-merged branches via GitHub PR status #123 + indefiniteness guard on both kernels), TestConleyDeviationsFromR (1-D radial vs 2-D separable, time-label normalization, independent temporal kernel deferred), TestConleyDeferrals (5 fail-closed NotImplementedError/TypeError contracts).

  Companion tests/test_conley_vcov.py drops 4248 → 3113 lines after methodology extraction; defensive surface preserved.

• Stale priority-queue cleanup at METHODOLOGY_REVIEW.md:L1386. PreTrendsPower removed (already Complete since 2026-05-19) and ConleySpatialHAC removed (this PR); substantive-review queue Add fixed effects and absorb parameters to DifferenceInDifferences #2-Add comprehensive code review for diff-diff library #5 renumbered to Add initial diff-diff library implementation #1-Claude/setup pip install k2 m4j #4; consolidation-pass queue Add multi-period DiD support #6-Add Synthetic Difference-in-Differences (SDID) estimator #8 renumbered to Add comprehensive code review for diff-diff library #5-Add multi-period DiD support #6.

• Bertanha-Imbens 2014 citation correction (16 sites across 5 files). NBER w20773 is on FRD external validity, NOT weighted spatial-HAC. Replaced across diff_diff/linalg.py (×8), diff_diff/conley.py (×1), diff_diff/guides/llms-full.txt (×2), docs/methodology/REGISTRY.md (×4), docs/api/spillover.rst (×1). New **Note (deferral status, 2026-05-26):** block frames the boundary as a tri-part contract:

  • Shipped — SpilloverDiD + Conley + survey via Wave E.1/E.2/E.3 (PR SpilloverDiD survey_design= on HC1/CR1 via Binder TSL (Wave E.1) #468/SpilloverDiD vcov_type='conley' + survey_design= via panel-aware stratified-Conley sandwich on per-period PSU totals (Wave E.2) #474/SpilloverDiD Wave E.3: SurveyDesign.subpopulation() full-design retention #482); TwoStageDiD + Conley + survey via Wave E.3 parity (PR TwoStageDiD Wave E.3 parity: always-treated full-design retention #485).
  • Deferred (generic linalg surface, any weight_type) — DiD/MPD/TWFE/LinearRegression generic path + Conley + survey_design=; LinearRegression / compute_robust_vcov Conley + weights= rejected for pweight, aweight, AND fweight.
  • Open methodological question (subset) — the pweight / survey_design portion of the deferral additionally lacks a canonical methodological extension of Conley (1999) for weighted spatial-HAC under probability sampling.

  Historical CHANGELOG entries (pre-[Unreleased]) intentionally retain Bertanha-Imbens 2014 attribution as accurate records of past release claims.

Methodology references

• Method name(s): ConleySpatialHAC

• Paper / source link(s): Conley, T. G. (1999), GMM Estimation with Cross-Sectional Dependence, J. Econometrics 92(1), 1-45. Paper review on file: docs/methodology/papers/conley-1999-review.md. Secondary: Andrews (1991) HAC theory; Düsterhöft (2021) conleyreg (CRAN v0.1.9) parity target; Colella, Lalive, Sakalli & Thoenig (2019) Stata acreg parallel.

• Any intentional deviations from the source (and why):

  • 1-D radial Bartlett on pairwise distance vs paper's 2-D separable Eq. 3.14 PSD-guaranteed form — practitioner specialization matching R conleyreg, Stata acreg, Hsiang (2010); indefiniteness guard at -1e-12 applies to both kernels.
  • Combined spatial + cluster product kernel (Wave A Align TROP lambda conventions with paper (Athey et al. 2025) #119) — library extension, no R correspondence; two limit-fixture anchors.
  • Callable conley_metric validation (Wave A Fix worktree-rm: detect squash-merged branches via GitHub PR status #123) — library extension.
  • Sparse k-d-tree path (Wave A Add git worktree management slash commands #120) — library extension, auto-activates at n > 5,000.
  • Time-label normalization via np.unique — deviation from R conleyreg literal; diff-diff is more robust on non-dense encodings.
  • Time-asymmetric kernel — matches R conleyreg literal (kernel arg controls spatial only).

  All documented in docs/methodology/REGISTRY.md § ConleySpatialHAC + METHODOLOGY_REVIEW.md § ConleySpatialHAC Deviations block.

Validation

• Tests added/updated:
  • tests/test_methodology_conley.py (CREATE, ~1600 LoC, 10 classes, 60 tests)
  • tests/test_conley_vcov.py (EDIT, methodology-anchored tests extracted; defensive surface preserved; ~1135 LoC removed)
• Backtest / simulation / notebook evidence: R conleyreg v0.1.9 parity at atol=1e-6 on 6 fixtures (3 cross-sectional + 3 panel) + sparse-forced cross-sectional + sparse-forced panel; internal block-decomposition cross-check at atol=1e-12.
• Local verification: pytest tests/test_methodology_conley.py + tests/test_conley_vcov.py = 176 pass + 5 slow deselected. grep -rn "Bertanha" diff_diff/ docs/ benchmarks/ = 0 hits. black + ruff clean.

Security / privacy

• Confirm no secrets/PII in this PR: Yes

---

Generated with Claude Code

[github-actions]

Overall Assessment

✅ Looks good

Executive Summary

• No P0/P1 issues found. The PR does not change Conley estimator math, weighting, variance computation, identification assumptions, or defaults; it is primarily a documentation/error-text/test reorganization PR.
• The methodology promotion is consistent with the Conley paper review and the registry: Eq. 4.2 pairwise form, Eq. 3.14 separable-Bartlett caveat, hardcoded Bartlett temporal kernel, and the generic weighted-Conley deferral all match the existing implementation.
• P3 — the citation/open-question cleanup is incomplete. The old generic-roadmap framing still remains in diff_diff/twfe.py and TODO.md, which now conflicts with the updated registry/code wording.
• I could not run pytest here because this environment is missing pytest and numpy, so the test review is static rather than execution-backed.

Methodology

• P3 — Impact: repository guidance on generic Conley + survey_design/weights is internally inconsistent after this PR. diff_diff/twfe.py (L102-L104) still describes survey_design + Conley as a generic “Phase 5 follow-up,” and TODO.md still cites Bertanha-Imbens 2014, while docs/methodology/REGISTRY.md (L3767-L3780), diff_diff/conley.py, and diff_diff/linalg.py now frame the generic surface as an open methodological question with estimator-specific shipped exceptions. Concrete fix: update the leftover TWFE docstring and TODO row to the same wording used in REGISTRY/conley.py/linalg.py, and keep SpilloverDiD/TwoStageDiD survey-aware Conley explicitly carved out.
• No other methodology findings. I did not find an undocumented change to estimator math, weighting, variance/SE, defaults, or assumptions in the diff.

Code Quality

• No findings.

Performance

• No findings.

Maintainability

• No findings beyond the stale documentation references noted in Methodology.

Tech Debt

• No findings. The deferred generic weighted-Conley work is already tracked, but the tracker text itself should be updated as noted above.

Security

• No findings.

Documentation/Tests

• No findings on the split itself; moving paper-anchored assertions into tests/test_methodology_conley.py and leaving defensive coverage in tests/test_conley_vcov.py is a sensible reorganization.
• Verification note: I could not execute the affected tests in this environment because pytest and numpy are unavailable.