Robustness — Triage Against the Cap (aeri-robustness)

When to trigger

• The robustness section is sprawling and competing with the result for word budget
• You have ten checks and at most one or two can appear in-text
• A referee will ask for a check the short paper has no room to add in full
• Unsure what is load-bearing robustness vs. reassurance that can go online

The short-format robustness problem

A standard paper answers every threat in-text; AER: Insights cannot. With ≤7,000 words (minus 200 per exhibit, ≤5 exhibits), robustness must be triaged: keep in-text only the one or two checks that, if they failed, would kill the headline result, and move everything else to the Supplemental Appendix (which has no length cap but is read only by skeptics and the data editor). The skill is deciding the minimum credible set of in-text checks and signposting the rest in a single sentence.

The triage rule

For each candidate check, ask: "If a referee saw only the main text, would the absence of this check make them disbelieve the headline result?"

| Answer | Placement | |--------|-----------| | Yes — it defends the core identifying assumption | In-text (1–2 max), often a single robustness exhibit or a sentence | | No — it reassures but the result survives without seeing it | Supplemental Appendix, named in one in-text sentence | | It is really an extension / new question | Appendix or a different paper, not "robustness" |

What usually earns an in-text slot

• The one check that directly tests the central identifying assumption (e.g., pre-trends for DiD, the donut/bandwidth for RDD, the exclusion falsification for IV).
• A single specification curve summary or a one-line "results are stable across N specifications (Appendix Table X)" rather than the full grid.
• For an experiment: the pre-registered primary outcome's robustness to attrition bounds.

What goes to the Supplemental Appendix

• Alternative bandwidths, functional forms, sample windows, fixed-effect structures
• Placebo tests beyond the single most informative one
• Heterogeneity splits (these are not robustness; they are extensions)
• Multiple-testing-adjusted versions, alternative clustering, leave-one-out

Signposting economically

Use one sentence to point to the appendix: "The result is robust to list] (Supplemental Appendix Section X; Tables X1–X6)." This buys credibility for the price of a sentence and keeps the word budget for the insight. Reproduce nothing in-text that the sentence already covers.

Execution bridge (StatsPAI / Stata MCP)

Run the battery, don't just enumerate it. Full map: execution-with-mcp. AER: Insights is a short format built around one decisive result, so the body/appendix split is even tighter — run the design cleanly the first time.

• Many outcomes / specifications: romano_wolf (step-down FWER, accounts for

cross-test correlation) or benjamini_hochberg — report the adjusted threshold.

• OVB sensitivity: oster_delta / sensemakr — the confounder strength that would

overturn the headline.

• Inference: wild_cluster_bootstrap (few clusters), twoway_cluster / conley.
• Re-fit off one handle: audit_result(result_id) lists the missing checks and the

exact suggest_function for each — no guessing the battery.

• Exhibits: etable / did_summary_to_latex from the handle — no retyped numbers.

Keep the decisive checks in the body and the exhaustive (now actually-run) battery in the appendix. See the executed chain in the JF execution walkthrough.

Checklist

• ] Every candidate check classified by the triage rule
• ] ≤2 robustness items in-text, each defending the core assumption
• ] All other checks in the Supplemental Appendix, named in one in-text sentence
• ] Heterogeneity treated as extension (appendix), not in-text robustness
• ] In-text robustness fits within the five-exhibit budget shared with the result
• ] No significance asterisks; SEs / confidence sets reported

Anti-patterns

• A full in-text robustness section that blows the word/exhibit cap
• Reproducing the appendix robustness grid in the main text
• Treating heterogeneity splits as "robustness" to justify keeping them in-text
• Omitting the central-assumption check to save space (it must stay)
• No signpost to the appendix, leaving referees unsure the checks exist

Output format

【Core-assumption check (in-text)】<the 1–2 that must stay>
【Signpost sentence】"Robust to [list] (Suppl. Appendix §X, Tables …)."
【To the appendix】alt bandwidths / forms / placebos / heterogeneity / clustering
【Exhibit budget】in-text robustness uses __ of 5 exhibits
【Next step】aeri-tables-figures