qje-tables-figures
GitHub专为QJE论文设计的图表优化技能，强调“图表优先”风格。用于将密集表格转化为可视化图形（如事件研究、RDD），规范表格自包含注释与标准误报告，确保符合单PDF提交及作者-日期引用格式，提升结果可读性。
Trigger Scenarios
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill qje-tables-figures -g -y
SKILL.md
Frontmatter
{
"name": "qje-tables-figures",
"description": "Use when finalizing the main exhibits of a Quarterly Journal of Economics (QJE) manuscript — making the paper figure-forward, with clean tables and self-contained notes that read well in QJE's single-PDF, author-date format. Designs exhibits; it does not run new analysis."
}
Tables & Figures (qje-tables-figures)
When to trigger
- The main result is a dense table with too many columns
- The paper is "table-heavy" when the design would land better as a figure
- Table notes are incomplete (sample, units, clustering, significance unclear)
- An event-study / RDD / binscatter result is hidden in a table instead of plotted
QJE aesthetic: figure-forward, self-contained exhibits
QJE has moved firmly toward figure-forward presentation — the Opportunity Insights / Chetty-style QJE paper makes its central result legible in one well-designed graph (e.g., the binned mobility maps and exposure-effect plots of the QJE 2014/2018 neighborhoods papers). Identification designs are inherently visual: event-study plots, RDD discontinuity plots, and binned scatters communicate credibility better than a coefficient buried in a regression column. Tables remain essential for estimates and robustness, but the headline should often be a figure a reader grasps in five seconds. Practical QJE constraints: at initial submission everything is one PDF with figures embedded (no separate figure files), exhibits are numbered and called out in order, and in-text references are author-date (Chicago).
The headline-figure decision
| Design | Headline figure |
|---|---|
| DID / event std | Event-study plot: leads ≈ 0, clean post-treatment dynamics |
| RDD | Discontinuity plot: binned means + local polynomial fit |
| IV | First-stage and reduced-form scatter / binscatter |
| RCT | Treatment-vs-control outcome distributions or effect-by-arm |
| Descriptive | The new fact, plotted with the data doing the talking |
Table craft
- Width discipline. Main results table should be readable; if it sprawls past a handful of columns, split it or move variants to the appendix (no page limit means you can — but readability still wins).
- Self-contained notes. Every table/figure note states: sample and time span, unit of observation, what each column is, fixed effects included, standard-error clustering level, and how significance is denoted.
- Standard errors in parentheses, clustering level named in the note; report N and relevant fit statistics.
- Coefficients with meaning. Report units so the magnitude is interpretable (effect in SDs, in dollars, in percentage points), not just a bare number.
- Author-date (Chicago) in-text references; figures and tables numbered and called out in order.
Figure craft
- Show the data: binned scatters, confidence bands, and raw-ish patterns build credibility.
- Avoid chartjunk: no 3D, no needless color, legible axis labels with units; figures must remain legible embedded in the single submission PDF and at print resolution.
- Confidence intervals shown, not just point estimates; bandwidth/bin choices noted.
- A figure should be interpretable from its caption alone.
Checklist
- The central result has a headline figure a reader grasps quickly
- Main table is readable; sprawling variants moved to the appendix
- Every exhibit note is self-contained (sample, units, FE, clustering, significance)
- Magnitudes are interpretable (units stated), not bare coefficients
- Event-study / RDD / first-stage results are plotted, not only tabulated
- Confidence intervals / bands shown on figures
- Figures embedded and legible in the single submission PDF; numbered, author-date citations
Execution bridge (StatsPAI / Stata MCP)
Generate exhibits from the fitted result rather than retyping numbers (the usual source
of body-vs-appendix drift). Full map:
shared-resources/empirical-methods/execution-with-mcp.md.
- Tables:
etable(multi-column) ordid_summary_to_latexstraight from theresult_id— one definition, one set of numbers, body and appendix in sync. - Event-study / coefficient figures:
plot_from_result,enhanced_event_study_plot,event_study_table— axis units and the SE/clustering note baked in. - Every note names the estimator + clustering (from the result's diagnostics) and states the magnitude in interpretable units. See a full fitted-result → exhibit chain in the JF execution walkthrough.
Anti-patterns
- A 9-column main table when a single event-study figure would carry the result
- Table notes that omit the clustering level or the sample definition
- Reporting coefficients with no units, so magnitude is uninterpretable
- Decorative 3D/colored charts that add no information
- Burying the cleanest evidence (the discontinuity, the leads) in an appendix table
Output format
【Headline exhibit】figure type chosen + why
【Main table】column count + what moved to appendix
【Notes audit】sample / units / FE / clustering / significance present? [Y/N each]
【Magnitude legibility】units stated? [Y/N]
【Figures plotted】[event study / RDD / first stage / ...]
【Next step】qje-writing-style
Version History
- 1839142 Current 2026-07-05 14:18
