Jleo Robustness

Robustness Strategy (jleo-robustness)

When to trigger

• The main result holds but a referee will ask "is this an artifact of the sample / specification / measurement?"
• The robustness section is a mechanical appendix list ("Table A1–A8") with no threat each addresses
• The institutional measure (asset specificity, governance form, judicial independence) is a proxy whose validity is questionable
• An alternative mechanism could produce the same pattern and is not yet ruled out
• The estimate is sensitive to clustering, controls, or the treatment-timing window

Robustness as threat-defusal at JLEO

JLEO referees — institutional and organizational economists — do not reward a wall of robustness tables. They reward a robustness program organized by the specific threats to the institutional claim. For each headline result, name the threats, then show the check that defuses each. The institution-specific threats cluster as follows:

Sequencing the program

1. List threats before checks. For each main result write the two or three threats that, if true, would overturn the institutional interpretation. Robustness exists to kill those, not to fill pages.
2. Construct validity first. Because JLEO results hinge on measuring institutional constructs (specificity, governance form, judicial independence, committee power), validate the measure early — alternative operationalizations should not move the conclusion.
3. Mechanism vs. rival. The most persuasive JLEO robustness is a test that only your mechanism predicts (e.g., the effect is concentrated where transaction-cost theory says it should be, absent where it should not). This doubles as evidence, not just defense.
4. Inference robustness. Institutional treatments often have few clusters (states, courts, reforms). Use wild-cluster bootstrap or randomization inference and report it.
5. Bound the unobservable. Where selection cannot be fully ruled out, use Oster-style bounds or sensitivity analysis to show how strong unobserved selection would have to be to overturn the result.

Execution bridge (StatsPAI / Stata MCP)

Run the battery, don't just enumerate it. Full map: execution-with-mcp. JLEO is law-and-economics/organizations; institutional designs with endogeneity — foreground identification.

• 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

• Each headline result has a named list of two or three threats to its institutional interpretation
• Every robustness exhibit maps to one named threat (not a generic "additional specifications" dump)
• The institutional construct is validated with at least one alternative measure
• A test that separates the proposed mechanism from the leading rival is included
• Inference robustness shown for few-cluster / serial-correlation concerns
• Selection sensitivity (Oster δ or equivalent) reported where selection is the live threat

Anti-patterns

• A robustness appendix organized by table number rather than by threat
• Reporting twenty specifications without saying which threat each one rules out (reads as fishing)
• Ignoring measurement of the institutional construct while piling controls on the regression
• Leaving the leading alternative mechanism un-tested because the main result is "significant"
• Default one-way clustering with a handful of institutional clusters and no bootstrap

Worked vignette (illustrative)

A paper claims judicial independence reduces expropriation of firms. Threats: (1) independence is endogenous to economic conditions; (2) the independence index is noisy; (3) general rule-of-law improvement, not independence specifically, drives the result. The threat-organized program: an event study around a constitutional reform isolating timing (threat 1); two alternative independence indices that agree (threat 2); and a placebo on a rule-of-law dimension that the mechanism says should not matter for this outcome (threat 3). Each table is captioned by the threat it kills, not by its position in the appendix.

Output format

【Headline result】[...]
【Threats to the institutional interpretation】1) ___ 2) ___ 3) ___
【Check per threat】threat 1 → ___ ; threat 2 → ___ ; threat 3 → ___
【Construct validity】alternative measure(s): ___
【Mechanism vs. rival test】[...]
【Inference robustness】wild-cluster / randomization / alt-clustering: ___
【Selection bound】Oster δ or sensitivity: ___
【Next skill】jleo-tables-figures