aejpol-writing-style

Writing Style — From Estimate to Policy Takeaway (aejpol-writing-style)

When to trigger

• The intro buries the policy question under data or method
• The abstract reports a coefficient but no policy lesson
• The paper has a clean estimate but no sentence a policymaker could act on — or it overclaims
• You are polishing for submission and need AEA house tone

> Late-stage polish: do not rewrite the intro until identification (aejpol-identification), the welfare bridge (aejpol-theory-model), and robustness (aejpol-robustness) have settled.

The AEJ: Policy introduction arc

Policy question → why credible identification is hard → the design that delivers it → headline causal estimate (with SE/CI) → welfare / cost-benefit / distributional reading → concrete, calibrated policy lesson → brief roadmap.

The distinctive moves vs. a general applied-micro intro:

1. Open with the policy question, not the dataset or estimator. A non-specialist AEA reader should know within two sentences what policy is at stake and why the answer matters.
2. Put the headline estimate, with its uncertainty, on page one — in policy-interpretable units (a percentage-point effect, a cost-per-outcome), never an asterisk.
3. Translate into a policy takeaway — the cost-benefit / MVPF / incidence reading — and state it as the contribution.
4. Calibrate the claim. Say exactly what the estimand is, for whom, and the conditions under which the lesson holds. The credibility of an AEJ: Policy paper rests as much on not overclaiming as on the estimate.

Translating estimates into policy language without overclaiming

• Convert coefficients into decision-relevant quantities ("a $1,000 expansion raises take-up by X and costs $Y per additional recipient") rather than leaving them as elasticities.
• Tie magnitude to a benchmark a policymaker recognizes (the program's budget, the status-quo level, a comparable policy).
• Hedge precisely, not vaguely: state the population the estimate applies to, the time horizon, and the assumptions the welfare reading needs. Replace "this shows policy X is good" with "for this population and horizon, the marginal dollar of X returns $Z, assuming [stated condition]."
• Separate what the data show (the causal estimate) from what the framework adds (the welfare reading) so a referee can grant one without the other.

AEA house style

• Author–year citations; SEs/CIs not asterisks (mirror aejpol-tables-figures); active voice; abstract that states question, design, headline estimate, and policy lesson. Online appendix carries the long material; the main text stays self-contained and readable. Review is single-blind, so the front matter names the authors — no need to anonymize the prose.

Checklist

• [ ] First two sentences state the policy question and why it matters
• [ ] Headline estimate with SE/CI appears early, in policy-interpretable units
• [ ] A one-sentence policy takeaway (cost-benefit / MVPF / incidence) is stated as the contribution
• [ ] The claim is calibrated: population, horizon, and assumptions named
• [ ] Causal estimate and welfare reading are separable in the prose
• [ ] No significance asterisks; author–year citations; abstract carries the policy lesson
• [ ] Self-citations cited normally (single-blind — no anonymization needed)

Anti-patterns

• An intro that leads with the dataset, the institutional weeds, or the estimator
• An abstract that ends at the coefficient with no policy lesson
• Overclaiming ("our results prove policy X should be adopted nationwide") beyond the estimand
• Vague hedging ("results should be interpreted with caution") instead of a precise scope statement
• Burying the policy takeaway in the conclusion where a policymaker will not find it
• Padding the prose with self-citations to signal a track record (cite only what the argument needs)

Worked vignette (illustrative)

Before: "Using administrative data and a difference-in-differences design, we estimate the effect of the reform on enrollment; the coefficient is 0.06 (s.e. 0.01)." After (AEJ: Policy): "Does auto-enrollment raise retirement-plan participation enough to justify its administrative cost? Exploiting the staggered rollout across employers, we find auto-enrollment raises participation by 6 percentage points (90% CI [4, 8]). At the program's per-worker cost this implies roughly $X per additional participant — cost-effective relative to a matching subsidy for this low-saver population, though the gain is concentrated among workers who would not have opted in (illustrative)." Question first, estimate with CI, policy lesson, calibrated scope.

Output format

【Opening policy question】one sentence
【Headline estimate】value + SE/CI in policy units, stated early
【Policy takeaway】cost-benefit / MVPF / incidence sentence
【Calibration】population + horizon + assumptions named
【Overclaim check】claim ≤ what design+framework support? [Y/N]
【Next step】aejpol-replication-package or aejpol-referee-strategy