Jle Robustness

Robustness Suite (jle-robustness)

When to trigger

• The main estimate of a legal/regulatory effect is in hand and you must show it is not an artifact of one specification
• A referee asks "is this robust to alternative controls / sample / which jurisdictions / how you date the rule?"
• The result depends on a bandwidth, a clustering choice, a treatment date, or a sample of included jurisdictions that could be questioned
• You suspect specification-search concerns and want to pre-empt them

The JLE robustness bar

JLE referees — economists who know the institution — probe whether the estimated effect of the rule is stable, honestly inferred, and not the product of researcher degrees of freedom, with special attention to legal-design choices: how you dated the rule, which jurisdictions you treated as controls, whether enforcement was uniform. Robustness here is not a wall of regressions; it is a targeted set of checks each tied to a specific threat to the legal identification. Map every plausible objection to the one check that answers it, and show the point estimate barely moves.

Robustness craft

1. Lock the primary specification first. Everything else perturbs around it; do not present five co-equal specs and let the reader guess the preferred one.
2. One threat → one check. A robustness table should read "here is the worry, here is the evidence it is not a problem."
3. Stress the legal-design choices specifically. Re-dating the rule, swapping the control jurisdictions, and varying enforcement assumptions are the JLE-characteristic checks a referee who knows the institution will demand.
4. Show stability of the point estimate, not just surviving significance.
5. Match inference to the data structure. With few states/jurisdictions, asymptotic clustered SEs over-reject; report a wild-cluster bootstrap or randomization inference — the single most common JLE robustness failure.
6. Be honest about where it weakens. A check that moves the estimate is information; report it and bound the implication.
7. Use placebos on uncovered legal areas. A distinctive and persuasive JLE robustness move is a placebo on an outcome the rule should not affect (an uncapped tort alongside a capped one, an unregulated adjacent market) — a null there isolates the legal channel far more credibly than another control regression.

Execution bridge (StatsPAI / Stata MCP)

Run the battery, don't just enumerate it. Full map: execution-with-mcp. JLE is empirical law-and-economics — DiD around legal/regulatory change is central.

• Many outcomes / specifications: romano_wolf (step-down FWER) or benjamini_hochberg.
• OVB sensitivity: oster_delta / sensemakr.
• Inference: wild_cluster_bootstrap (few clusters), twoway_cluster / conley.
• Re-fit off one handle: audit_result(result_id) lists missing checks + the exact suggest_function for each.
• Exhibits: etable / did_summary_to_latex from the handle — no retyped numbers.

Decisive checks in the body, exhaustive battery in the appendix. JF execution walkthrough.

Checklist

• Primary specification declared before perturbations
• Each robustness check mapped to a specific threat, not added for volume
• Treatment-date sensitivity shown (signing vs. effective vs. enforcement)
• Control-jurisdiction sensitivity shown (drop contaminated, leave-one-out, placebo legal area)
• Coefficient-stability evidence (Oster δ or stepwise) for selection on unobservables
• Inference at the legal-variation level + wild-cluster/randomization where jurisdictions are few
• Design-specific sensitivity (honest-DID / RD bandwidth / weak-IV set)
• Stability of the point estimate shown; any check that moves it reported honestly

Separating identification robustness from policy interpretation

Keep the robustness section about whether the estimate of the legal effect is stable, and do not let it drift into re-arguing the rule's normative merits. A referee wants to know the number survives re-dating, control swaps, and correct inference — not your view on whether the rule is good policy. Park the welfare and policy discussion in its own section (see jle-theory-model) so the robustness evidence reads as clean, mechanical stress-testing of the identified effect.