American Economic Review Response to Reviewers: An R&R Evidence Guide

Every reply should identify the changed page and line. For formal or empirical material, add the proposition, equation, figure, table, or appendix. A response such as "we added robustness checks" is not auditable; name the threat each check targets and the result it produces.

AER asks that essential methods, variable definitions, data sources, and formal proofs remain in the self-contained paper rather than requiring the reader to consult a supplemental appendix. That boundary matters in revision. Do not answer a central identification concern only with a large appendix table and leave the main paper unchanged.

Rebuild the response locations after final pagination. Equation and appendix numbers frequently shift during a large economics revision.

Use a stable distinction: comments in bold or shaded blocks, responses in regular text, and explicit labels for Editor, Referee 1, Referee 2, and Referee 3. Avoid a color-only scheme.

Use the bold font and labels to distinguish each reviewer comment from the author response even when the PDF is printed in grayscale. That visual rule should remain consistent in every section.

Long economics reports often contain nested comments. Preserve their hierarchy. If you divide one comment into parts 2a, 2b, and 2c, reproduce enough of the original wording that the referee recognizes each part and can verify your answer without searching the old report.

AER revision review combines broad contribution judgment with high methodological rigor. The editor synthesizes reports, while referees may focus on design, theory, measurement, or institutional context. Your opening letter should therefore explain how those specialist requests now support one general-interest economics contribution, not several parallel papers.

The response is strongest when it lets the editor distinguish a resolved validity concern from an optional extension. Referees can reasonably disagree about preferred specifications; the editor needs a coherent account of which evidence is decisive and why the revised claim stops at the supported boundary.

An requested method is not automatically the right answer. Identify the threat, then choose the design or analysis that actually addresses it.

Statistical significance is not a response to design validity. Report magnitude, uncertainty, and what the new analysis changes about the economic conclusion.

In our pre-submission review work with American Economic Review manuscripts and revisions, we connect each response to the design code, analysis output, manuscript claim, and replication documentation. We audit every controlling issue across those artifacts and observe where the revised inference still depends on an old result or undocumented step. That cross-artifact review is necessary because a long economics rebuttal can look rigorous while the identifying assumption, economic magnitude, or computational path remains unchanged. These anonymized manuscript patterns are not AER editorial records or access to referee deliberations. Each is testable from an author's own response, revised paper, and data-and-code package.

A robustness appendix that never states the threat. American Economic Review revisions sometimes add twenty specifications without explaining which one tests selection, measurement, functional form, or equilibrium spillovers. The response becomes longer without making identification stronger. We route each table and figure to one threat, record the result and consequence, and remove redundant variants that do not change the inference.

A mechanism asserted through heterogeneity. In an American Economic Review response, an effect may be larger in one subgroup and then be called proof of a channel. Unless competing mechanisms predict different patterns, heterogeneity is compatible with several stories. A credible response derives a discriminating prediction or labels the evidence descriptive, then updates the introduction and conclusion to use the same boundary.

A changed estimate with an unchanged contribution paragraph. New data cleaning, sample restrictions, or design corrections move the magnitude materially, but the abstract still carries the original number and policy interpretation. Returning referees notice the high-salience mismatch before reading the appendix.

A replication package that reproduces only the final regression. Data construction, exclusions, intermediate merges, random seeds, or figure scripts remain manual. The AEA Data Editor needs a traceable path from permitted inputs to published outputs. Treat provenance and environment as scientific documentation.

Our final Sullivan-style test asks whether another author could infer the same identification claim from the reported design and reproduce the published tables without private explanation from the team. When the answer is no, the response still depends on tacit knowledge and is not ready for the same editor and referees.

Scan the revised AER argument, response, and reproducibility package as one system before resubmission.

Push back when a requested design estimates a different object, relies on an indefensible assumption, or would move the paper away from the contribution the editor invited. State the underlying concern first, then offer a better test.

For example, a reviewer may request an instrumental-variable estimate to address reverse causality. If the proposed instrument lacks a credible exclusion restriction, do not add it as decoration. Strengthen the existing design, add a targeted sensitivity analysis, and state the remaining causal boundary.

Before acceptance, empirical, simulation, and experimental work must comply with AEA data-and-code policy. Use the revision to create:

1. a top-level README with the computational sequence;
2. provenance and access notes for each input;
3. an environment or dependency specification;
4. deterministic seeds where randomness is used;
5. a crosswalk from scripts to every table and figure;
6. documented handling for confidential or restricted data;
7. a clean run that starts from the permitted inputs.

If data cannot be shared, follow the policy's documentation and preservation requirements. Do not imply public availability where legal access is restricted.

Most rejection-on-revision risk comes from a controlling issue that accumulated analyses but did not become more credible. The response is long, yet the identifying assumption, distinct contribution, or mechanism remains unchanged. Another common failure is hiding a changed result in an appendix while preserving the original headline.

An R&R is not acceptance. If a new analysis contradicts the original interpretation, revise the paper around the result you can support. Editors can evaluate a narrower true contribution; they cannot rely on a response that treats contradiction as a robustness check.

1. Extract the editor's controlling issues before drafting replies.
2. Map every referee request to contribution, identification, mechanism, or generality.
3. State the result of each new analysis, not only that it was run.
4. Update abstract, introduction, tables, and conclusion after material changes.
5. Keep essential evidence and proofs in the self-contained paper.
6. Cite page, line, equation, figure, table, and appendix locations.
7. Distinguish editor/referee comments from author responses visually.
8. Build the data-and-code package during revision.
9. Verify every response location after final typesetting.
10. Recheck AER journal fit if the revised contribution changes materially.

This guide was reviewed on July 12, 2026. AEA sources define policy; the issue-to-evidence framework is Manusights analysis.