academic-paper-writer

name	academic-paper-writer
description	Drafts and revises empirical and theoretical economics papers in journal-style structure (introduction, literature, data, identification, results, robustness, discussion, conclusion), with explicit identification arguments, careful causal language, and reviewer-response discipline. Cross-links to `latex-econ-model` for theory typesetting, `latex-tables` for `esttab`/`modelsummary`/`pyfixest.etable` table production, and `econ-visualization` for code-driven figures, so every numeric claim, table, and figure in the paper is reproducible from one project repo. Use when the user asks to draft an introduction, write a methods or identification section, interpret results, polish a discussion, prepare a journal submission, write a referee response letter, or restructure an existing paper.
workflow_stage	writing
compatibility	["claude-code","cursor","codex","gemini-cli"]
author	JonasWeinert
version	2.0.0
tags	["LaTeX","academic-writing","papers","economics","identification","referee-response","dime"]

Academic Paper Writer

Draft and revise economics papers with the structure and prose that journal reviewers expect. The default style is empirical-applied economics: a clear identification argument carries the paper. Theory papers and literature surveys deviate from this template — flag the deviation explicitly.

Operating Principles

1. Identification carries the paper. In an empirical paper, the most-read and most-criticized section is the one that explains what variation drives the coefficient and why that variation is plausibly exogenous. Lead the methods section with the identification argument, not the regression equation.
2. Numbers are reproducible. Every statistic in the paper is the output of a script in the project repo (see latex-tables for esttab/modelsummary/pyfixest.etable; econ-visualization for figures). Never type a coefficient by hand.
3. Causal language is precise. "Causes" / "leads to" / "drives" requires a defended identification strategy. "Is associated with" / "predicts" / "covaries with" is the right phrasing for descriptive results. Reviewers track this carefully.
4. One claim per paragraph. Topic sentence first, evidence next, qualification last.
5. Cite where it matters. Heavy citations in the literature review and the identification section, lighter elsewhere. Engage with contradictory findings, don't ignore them.

Decision Policy

This skill follows the repo-wide Agent Policy.

ASK before proceeding (blocking):

1. Paper type (empirical / theoretical / referee response) and stage (outline / draft / R&R).
2. Target journal — affects length, structure, and identification framing.
3. The headline number and identification strategy in plain language. The agent does not invent these; the user provides them.
4. For R&R: which specific reviewer comment is being addressed.
5. Whether to make claims of causality (requires user to confirm the identification argument).

DEFAULT + flag (use this default; tell the user how to override):

• Five-paragraph introduction structure (motivation → contribution → identification → headline → literature).
• Identification paragraph appears in the intro (one paragraph) and in the empirical strategy section (full).
• Causal language convention pinned at the top of the draft: "causes" requires defended identification; "is associated with" for descriptive.
• Numbers come from \input{tabs/...} or \headlineEstimate{} macros, never typed in prose.
• Five-sentence abstract structure (question → setting → headline → second-order → contribution).

DOCUMENT and proceed (write into the decisions log at the top of paper.tex):

• Section structure if it deviates from IMRAD-style empirical default.
• Cross-references to tabs/ and figs/ artifacts for every numeric claim.
• Any rewriting of the user's prose, with a comment explaining the change.

PROCEED items: \Cref{} over \ref{}; tex/notation.tex shared with slides; tex/numbers.tex regenerated by the analysis script; one section per file via \input{}.

Pre-flight Checklist

Before drafting, confirm with the user — and write the answers in the paper repo's README.md:

• Paper type. Empirical (RCT / quasi-experimental / descriptive) or theoretical (model / mechanism design / proof)?
• Stage. Outline, first draft, revision-and-resubmit, or referee response?
• Target journal. AER / QJE / JoE / RES / JPE / field journal? Determines length, structure, and style guide.
• Identification strategy. RCT, IV, DiD (and which estimator), RDD, event study, or descriptive?
• Main result. State the headline number now: "We find a Y% increase in X following a Z change."
• Key contribution. What does this paper add that would make a reasonable reviewer recommend it for the target journal?
• Sections that need help. Intro, literature, data, identification, results, robustness, discussion, conclusion, abstract, response letter?

Empirical Paper Structure (default)

1. Introduction (1.5-2 pages, 5 paragraphs)
   - Motivation + question
   - What this paper does
   - Identification in one paragraph
   - Headline result with magnitude
   - Contribution to literature (3-5 strands, named papers)
   - (optional) road map

2. Background / Institutional Setting (often merged with intro
   or section 2 in a labor / public / development paper)

3. Data
   - Source(s), sample period, unit of observation
   - Sample construction with N reported at each step
   - Summary statistics (Table 1)
   - (For RCTs) Balance table (Table 2)

4. Empirical Strategy
   - Identification argument FIRST, equation SECOND
   - The estimating equation with notation defined
   - Threats to identification, addressed in turn
   - Inference (clustering level, why)

5. Results
   - Main table walk-through
   - Magnitude in economic terms (relative to mean / SD / counterfactual)
   - Heterogeneity (when meaningful)

6. Robustness
   - Alternative samples, controls, FE, clustering
   - Placebo tests
   - Specification curve / multiverse (when feasible)

7. Mechanisms (when paper has them)
   - Hypothesized channel
   - Reduced-form evidence per channel
   - Honest about what the paper cannot disentangle

8. Discussion / Policy
   - External validity
   - Magnitude in policy-relevant terms
   - Limitations

9. Conclusion (1 page, 3-4 paragraphs)
   - Restate question + answer (no new results)
   - Contribution
   - Open questions

The Identification Paragraph (the most important paragraph in the paper)

This is the paragraph reviewers stress-test. It belongs in the introduction (one sentence) and in the empirical strategy section (a full paragraph).

Template (full paragraph, fill the brackets):

Our identification strategy exploits [SOURCE OF VARIATION: the staggered rollout of program X across U.S. states between [Y1] and [Y2]]. The key identifying assumption is that, absent the program, [TREATED UNITS] would have followed the same trend in [OUTCOME] as [CONTROL UNITS]. We provide three pieces of evidence in support of this assumption. First, [PRE-TREND EVIDENCE]: in an event-study specification, the leads of the treatment indicator are jointly indistinguishable from zero ([Cref]). Second, [PLACEBO EVIDENCE]: applying the same design to [PLACEBO OUTCOME] yields a precise null. Third, [CHARACTERISTICS BALANCE]: pre-treatment characteristics are balanced across rollout cohorts ([Cref to Table 2]). The remaining identification concern is [HONEST CONCERN]; we address it in [SECTION/SECTION].

Causal-Language Discipline

You wrote	Implies	Use when
X causes Y	Defended causal effect	RCT or strong quasi-experiment (IV, DiD, RDD with checks)
X has a causal effect on Y	Same	Same
X drives / leads to / increases Y	Causal but slightly softer	Same; common in working-paper style
The effect of X on Y is ...	Causal	Same
X is associated with Y	Conditional correlation	Descriptive results, OLS without identification
X predicts Y	Conditional correlation, often time-ordered	Forecasting context
X covaries with Y	Pure correlation	Pure descriptives, no controls

Mixing these is the most common reviewer complaint. Pin the convention up front.

Common Pitfalls

• Burying the main result. The introduction must state the headline number in the first 1-2 paragraphs.
• Using "significant" without specifying statistical or economic.
• Over-claiming causality without a defended identification strategy.
• Literature review that's a list of papers — should organize by mechanism or design, not chronology.
• Conclusion that just summarizes — should also synthesize and signal contribution.
• Walking through tables column-by-column instead of leading with the headline coefficient.
• Hand-typing numbers in the prose: every statistic should come from \input{tabs/...} or be a \newcommand{\headlinenumber}{0.082} at the top.
• Treating standard errors as ornamental — discuss them where they matter (weak IV, few clusters, multiple testing).
• Skipping the "honest limitation" paragraph in the discussion. Reviewers will find them anyway.

Additional Resources

• reference.md — extended templates: introduction (5-paragraph plan), identification section, results walk-through, robustness section, response-to-referees letter, abstract structure, journal-specific deviations.
• examples/ — copyable LaTeX section templates:
  • examples/intro_template.tex — 5-paragraph introduction
  • examples/identification_template.tex — the identification section
  • examples/results_template.tex — results walk-through
  • examples/robustness_template.tex — robustness ladder
  • examples/conclusion_template.tex — conclusion + limitations
  • examples/abstract_template.tex — 5-sentence abstract
  • examples/response_to_referees_template.tex — point-by-point response

Cross-Skill Routing

• For LaTeX setup, theorem environments, and notation discipline → latex-econ-model skill.
• For esttab / modelsummary / pyfixest.etable table production → latex-tables skill.
• For event-study, balance, time-series, and binscatter figures → econ-visualization skill.
• For Beamer slides built on the same notation file → beamer-presentation skill.

References

• Cochrane, Writing Tips for PhD Students.
• Shapiro, How to Give an Applied Micro Talk — applies to writing as well.
• Thomson (2011), A Guide for the Young Economist.
• McCloskey, Economical Writing.
• Berk, Writing Tips for Economists.
• Roth, Sant'Anna, Bilinski & Poe (2023), What's Trending in DiD? — modern identification language.
• DIME Analytics, Reproducible Research — every number reproducible from code.
• DIME Analytics, Exporting Analysis — full-replicability tier for tables and figures.
• AEA, Manuscript Style Guide.