Remove jackknife variance estimation from TROP

[igerber]

Summary

• Remove variance_method parameter from TROP.__init__() and TROPResults dataclass
• Delete _jackknife_variance() and _jackknife_variance_joint() methods (~180 lines)
• Remove dispatch conditionals — bootstrap is now called directly
• Update README, REGISTRY.md, tutorial notebook, and CHANGELOG

Breaking change: TROP(variance_method=...) now raises TypeError; TROPResults.variance_method no longer exists. Jackknife was never specified in the source paper — bootstrap is the only inference method described in Athey et al. (2025).

Methodology references (required if estimator / math changes)

• Method name(s): TROP (Triply Robust Panel Estimator)
• Paper / source link(s): Athey, S., Imbens, G. W., Qu, Z., & Viviano, D. (2025). Triply Robust Panel Estimators. https://arxiv.org/abs/2508.21536
• Any intentional deviations from the source (and why): None — this change removes a deviation. Jackknife was an implementation addition not described in the paper. Bootstrap (Algorithm 3) is now the sole variance method, matching the paper exactly.

Validation

• Tests added/updated: tests/test_trop.py — 4 jackknife tests deleted, 5 tests updated to remove variance_method kwarg. 80 tests pass.
• Backtest / simulation / notebook evidence: docs/tutorials/10_trop.ipynb cells 24-25 rewritten for bootstrap-only variance.
• Verified TROP(variance_method="bootstrap") raises TypeError (clean removal)
• Verified trop() convenience function still works
• Grepped codebase: no remaining TROP-specific jackknife/variance_method references

Security / privacy

• Confirm no secrets/PII in this PR: Yes

---

Generated with Claude Code

[github-actions]

Overall Assessment

• ✅ Looks good

Executive Summary

• Jackknife removal is consistent across code paths; both joint and twostep now always use bootstrap for SEs. diff_diff/trop.py:L1486-L1524 diff_diff/trop.py:L2067-L2107
• Methodology Registry now matches the paper: bootstrap-only inference for TROP. docs/methodology/REGISTRY.md:L548-L552 docs/methodology/REGISTRY.md:L972-L976
• Docs/tests reflect the API change; examples no longer mention jackknife. README.md:L1871-L1916 tests/test_trop.py:L189-L236
• Minor quality suggestion: consider validating n_bootstrap lower bounds now that bootstrap is mandatory.

Methodology

• P3 — Alignment confirmed. Impact: None; bootstrap-only inference in code matches Methodology Registry and paper. Concrete fix: None. diff_diff/trop.py:L1486-L1524 diff_diff/trop.py:L2067-L2107 docs/methodology/REGISTRY.md:L548-L552

Code Quality

• P3 — Consider validating n_bootstrap >= 2 (or warn) now that bootstrap is the sole variance path; otherwise SE can be 0/NaN with no explicit signal when reps are too small. Impact: Potential silent NaN inference if users set very low reps. Concrete fix: add a guard in TROP.__init__ or before bootstrapping to raise/warn on n_bootstrap < 2. diff_diff/trop.py:L452-L492 diff_diff/trop.py:L1486-L1524

Performance

• P3 — No performance issues introduced in the diff. Impact: N/A. Concrete fix: N/A.

Maintainability

• P3 — No maintainability concerns; removal simplifies inference branching. Impact: N/A. Concrete fix: N/A.

Tech Debt

• P3 — No new tech debt identified. Impact: N/A. Concrete fix: N/A.

Security

• P3 — No security issues found. Impact: N/A. Concrete fix: N/A.

Documentation/Tests

• P3 — Optional: add a regression test that TROP(variance_method=...) raises TypeError to lock in the breaking change and prevent silent reintroduction via kwargs. Impact: Prevents accidental API drift. Concrete fix: add a small test in tests/test_trop.py. tests/test_trop.py:L236-L258