Robustness Battery (jfe-robustness)
When to trigger
- The result holds in the baseline but you have not tried obvious alternatives
- A referee could ask "does this survive a different measure / sample / specification?"
- You are unsure which robustness tests belong in the main text vs. the Internet Appendix
- You suspect a result might be fragile and want to find out before a referee does
Why robustness is JFE's signature
JFE's referee culture rewards thoroughness: a result is credible only after it survives a wide battery of sensible alternatives, and reviewers expect every alternative explanation to be addressed, not merely the convenient ones. A fragile result that breaks under a routine variation is a classic JFE rejection. The headline tests live in the main text; the rest go to the Internet Appendix, which JFE asks you to append to the end of the main manuscript file (see jfe-internet-appendix).
Because JFE mandates a code + non-proprietary-data deposit (Mendeley Data) at acceptance for post-2021 submissions, your robustness battery must be runnable, not just described — referees increasingly probe reproducibility after the field-wide concerns about non-replicable factors (the Fama-French factor lineage is the benchmark) and fragile staggered-DID results.
The robustness dimensions
Work through each; for each, decide main-text vs. appendix.
1. Alternative measures
- Re-measure the dependent and key independent variables (different proxies, data sources, definitions).
- Show the result is not an artifact of one operationalization.
2. Alternative specifications
- Add/remove fixed effects; vary the control set; show coefficient stability (a coefficient-stability / Oster bounds argument for omitted-variable bias where relevant).
- Functional form: levels vs. logs, linear vs. nonparametric.
3. Alternative samples
- Drop influential years (e.g., the financial crisis), industries, or large firms.
- Subperiod stability; exclude the most/least-treated units.
4. Alternative inference
- Re-cluster at a different level; wild-cluster bootstrap when clusters are few.
- Permutation/randomization inference where appropriate.
5. Falsification & placebos
- Outcomes that should not respond if your story is right.
- Placebo timing or placebo treatment groups.
6. Ruling out alternative explanations
- For each rival story a referee could tell, design a test that distinguishes it from yours.
- This is the single most important block for JFE — list the alternatives explicitly and kill them one by one.
Execution bridge (StatsPAI / Stata MCP)
Run the battery, don't just enumerate it. Full map:
execution-with-mcp. JFE is finance top-3 (with JF, RFS) — corporate-causal chain for corporate papers, factor-zoo haircut for asset pricing; attribute canon to the correct top-3 journal.
-
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
Anti-patterns
- Reporting only the robustness checks that pass ("cherry-picked specifications")
- A "robustness" section that re-runs the same model with one extra control and calls it done
- Leaving an obvious rival explanation untested and hoping no referee notices
- A result that flips sign or loses significance under a routine variation, reported without comment
- Dumping every check into the main text so the paper becomes unreadable
Output format
【Headline result】...
【Robustness done】measures[...] specs[...] samples[...] inference[...] placebos[...]
【Alternatives ruled out】[story -> test] pairs
【Fragile under】[...] (if any — fix before submission)
【Main-text vs. appendix】split decided: yes/no
【Next】jfe-tables-figures
