Agent Skills
› brycewang-stanford/Awesome-Journal-Skills
› jebo-tables-figures
jebo-tables-figures
GitHub为JEBO论文生成行为对比图表，涵盖处理效应、动态变化及分布可视化。确保灰度可读、含置信区间且自解释，不执行分析或撰写正文，旨在提升实验机制的清晰度与说服力。
Trigger Scenarios
回归表中处理差异不清晰
缺乏实验动态或分布可视化
不确定JEBO图表显著性标记规范
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill jebo-tables-figures -g -y
SKILL.md
Frontmatter
{
"name": "jebo-tables-figures",
"description": "Use when a Journal of Economic Behavior & Organization (JEBO) manuscript's exhibits must make a behavioral comparison legible — treatment bar charts, by-round dynamics, distribution plots, regression tables. Builds exhibits that show the mechanism; it does not run the analysis or write the prose."
}
Tables & Figures (jebo-tables-figures)
When to trigger
- Treatment differences are buried in a wall-of-coefficients regression table
- A reader cannot see the behavioral effect without reading the surrounding paragraph
- Experimental dynamics (learning over rounds, by-period contributions) are not visualized
- The figure shows means but hides the distribution the behavioral story depends on
- You are unsure whether significance markers belong in a JEBO exhibit
What a JEBO exhibit must do
The reader should see the behavioral comparison — treatment vs. control, by group or over time — before reading a word of text. JEBO's experimental core makes a few exhibit types load-bearing:
- Treatment-comparison plot. Bar/dot plot of the outcome by treatment with confidence intervals; the headline contrast is the visual focus, not a footnote.
- Dynamics plot. For repeated games / multi-round experiments, plot the outcome by period and treatment so learning, decay, or convergence is visible (the round-by-round graph is a JEBO signature).
- Distribution plot. Where the mechanism is about heterogeneity (e.g., types, bimodality of contributions), show the full distribution (CDF, histogram, or beeswarm), not just the mean — a difference in means can hide a difference in shape.
- Regression table. Used to support the figure, with the design-relevant comparison up top, clustered SEs, and pre-registered/primary specification flagged.
Exhibit decisions by what the result is about
| The behavioral claim is about… | Lead exhibit | Common mistake |
|---|---|---|
| A level difference across treatments | bar/dot plot with CIs | a 6-column table where the contrast is row 4 |
| Change over repeated interaction | by-period line plot per treatment | reporting only the pooled average |
| Heterogeneity / types | distribution / CDF / beeswarm | a single mean hiding bimodality |
| A predicted ordering of arms | ordered plot matching the model's prediction | unordered table that obscures the ranking |
| A structural/behavioral estimate | parameter plot with CIs + fit overlay | a table of parameters with no fit visualization |
House-style and Elsevier mechanics
- Confidence intervals over bare asterisks. Show SEs/CIs; let the reader judge. (Elsevier permits significance stars, but a behavioral comparison is far more persuasive shown as overlapping/non-overlapping intervals; reserve any asterisks for support tables, never as the sole evidence.)
- Self-contained captions: a referee should understand the exhibit without the body text — define treatment, unit, sample, error-bar meaning, and n.
- Figures legible in grayscale (Elsevier print) — use patterns/markers, not color alone.
- Report n per cell and the unit of observation (subject vs. subject-period vs. session); cluster the inference accordingly.
- Number tables/figures and place near first mention; high-resolution vector figures per Elsevier artwork specs (检索于 2026-06；以官网为准).
Execution bridge (StatsPAI / Stata MCP)
Generate exhibits from the fitted result, not by retyping numbers. Full map:
execution-with-mcp. JEBO spans behavioral/experimental and applied micro; randomization inference for experiments, DiD/IV for observational claims.
- Tables:
etable(multi-model) ordid_summary_to_latexstraight from theresult_id. - 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 and states the magnitude in interpretable units.
See a full fitted-result → exhibit chain in the JF execution walkthrough.
Checklist
- The headline behavioral comparison is visible in the lead figure before any prose
- Repeated-interaction results have a by-period plot per treatment
- Heterogeneity claims show the distribution, not only the mean
- Confidence intervals shown; significance stars are not the only evidence
- Captions are self-contained (treatment, unit, n, error-bar definition)
- Figures readable in grayscale; n-per-cell and clustering level reported
- Exhibit ordering matches the model's predicted ordering where one exists
Anti-patterns
- The main treatment effect appearing only as a coefficient in a dense table
- A bar of means concealing a bimodal contribution distribution
- Asterisks doing all the persuasive work with no CIs
- Color-only figures that collapse in grayscale print
- Captions that require the body text to interpret
- Reporting subject-period n as if it were independent subjects (wrong clustering)
Output format
【Lead exhibit】treatment-comparison / by-period dynamics / distribution / parameter plot
【What it shows at a glance】<the behavioral contrast>
【Inference shown】CIs / SEs (asterisks support-only)
【Unit + n per cell】<subject / subject-period / session>; clustering:
【Grayscale-safe + self-contained caption】[Y/N]
【Next step】jebo-writing-style
Version History
- 1839142 Current 2026-07-05 13:32
