Agent Skills
› brycewang-stanford/Awesome-Journal-Skills
› jel-tables-figures
jel-tables-figures
GitHub用于为JEL综述构建总结性图表，包括“谁发现了什么”汇总表、概念框架图和元证据展示。旨在综合文献而非呈现新估计，确保可比性与可信度，避免不当合并或复制原图。
Trigger Scenarios
需要将争议或估计结果以表格形式清晰展示时
需要用图表呈现组织框架或机制时
准备制作综述用表而非原始回归表时
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill jel-tables-figures -g -y
SKILL.md
Frontmatter
{
"name": "jel-tables-figures",
"description": "Use when building exhibits for a Journal of Economic Literature (JEL) survey — summary \"who-found-what\" tables, conceptual\/framework figures, and meta-evidence exhibits that synthesize across studies. Designs survey exhibits; it does not produce original-estimate regression tables (a JEL survey reports no new estimates of its own)."
}
Tables & Figures for a Survey (jel-tables-figures)
When to trigger
- The synthesis is done and the reader needs to see the field at a glance
- A controversy or a body of estimates would be clearer as a table than as prose
- The organizing framework would land better as a diagram
- You are tempted to paste a regression table — but this is a survey, not a primary paper
The three exhibit types a JEL survey actually uses
JEL exhibits summarize across the literature; they do not present the author's own estimation. The workhorses:
| Exhibit | Purpose | Design notes |
|---|---|---|
| Who-found-what summary table | one row per study (or per design class): question/estimand, method, sample, finding (direction + magnitude), credibility note | rows ordered by the framework's cells, not chronology; columns let the reader compare comparable objects |
| Conceptual / framework figure | render the organizing spine — taxonomy tree, mechanism diagram, the simple unifying model | this is often the survey's signature exhibit; it should be restate-able from memory |
| Meta-evidence exhibit | a forest-style plot or funnel of effect sizes, a timeline of methods, a coverage map | use only when the estimates are commensurable; otherwise it manufactures false consensus |
Building a credible summary table
- Compare like with like. Group studies that estimate the same object; never put non-comparable estimands in one magnitude column (the cross-study version of the pooling error from
jel-literature-synthesis). - Carry uncertainty and credibility. A finding column without a credibility/identification note invites a referee to ask "but how good is that study?" — answer it in the table.
- Self-contained captions. A JEL exhibit is read on its own; the caption states what the table shows, the unit, and how to read a row, and points to the source studies.
- Source every cell. Each entry traces to a paper in the evidence matrix; a survey table with an unsourced number is indefensible.
Meta-analysis caution
If you assemble effect sizes into a quantitative synthesis (forest plot, meta-regression), you are doing a meta-analysis, with all its assumptions — comparable estimands, publication-bias diagnostics, weighting. Only do this where the literature genuinely supports it; otherwise a qualitative who-found-what table is more honest than a spurious pooled number. JEL readers include the methodologists who would catch an invalid pooling.
Reproduced vs. re-drawn figures
When a figure from a surveyed paper is central, prefer a re-drawn synthesis figure (your own panel that places several studies on common axes) over copying one paper's exhibit. A re-drawn figure serves the survey's argument and avoids the consensus-by-accident of reprinting whichever paper had the prettiest chart; if you do reproduce an original figure, attribute it and secure any permission the AEA style guide requires.
Checklist
- Each exhibit synthesizes across studies (no original-estimate regression output)
- Summary-table rows ordered by the framework's cells; only comparable estimands share a column
- Finding columns carry a credibility/identification note, not bare point estimates
- The conceptual figure renders the organizing spine and is restate-able from memory
- Any meta-evidence plot pools only commensurable estimates; bias diagnostics noted
- Captions are self-contained (what / unit / how to read a row / sources)
- Every cell sources to a paper in the evidence matrix
- Exhibits are placed where the argument needs them, sized for readability in print
Anti-patterns
- Pasting a regression table of the author's own new estimates — a survey reports no new results
- A who-found-what table that pools incomparable estimands into one magnitude column
- A forest plot implying a pooled consensus the literature does not support
- Finding columns with no credibility note (every reader then asks "is that study any good?")
- Decorative figures that do not encode the framework
- Exhibits whose captions require the body text to be intelligible
Output format
【Exhibit set】<list: summary tables / conceptual figure / meta-evidence>
【Summary table】rows by framework cell; comparable estimands only? Y/N
【Credibility column】present for every finding? Y/N
【Conceptual figure】renders the spine; restate-able from memory? Y/N
【Meta-evidence】pools only commensurable estimates (or omitted)? Y/N
【Sourcing】every cell traces to the evidence matrix? Y/N
【Next step】→ jel-writing-style (weave exhibits into the synthesis prose)
Version History
- 1839142 Current 2026-07-05 13:34
