Agent Skills
› brycewang-stanford/Awesome-Journal-Skills
› respol-tables-figures
respol-tables-figures
GitHub专为Research Policy稿件设计图表技能，聚焦回归表、专利地图及因果推断图。旨在打造跨学科读者易懂的自解释展示，突出创新机制，避免堆砌数据，确保结果清晰传达核心贡献。
Trigger Scenarios
关键结果被复杂的系数表淹没，难以解读
文献计量或专利网络图缺乏解释性
研究设计需要事件研究或剂量反应图但缺失
定性论文缺少连接证据与构念的数据结构表
审稿人指出图表未能回答研究问题
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill respol-tables-figures -g -y
SKILL.md
Frontmatter
{
"name": "respol-tables-figures",
"description": "Use when exhibits are the bottleneck for a Research Policy (RP) manuscript — designing tables and figures (regression tables, patent\/bibliometric maps, event-study plots, case data structures) that carry the innovation mechanism for an interdisciplinary readership. Builds exhibits; it does not run the analysis (respol-data-analysis) or write prose (respol-writing-style)."
}
Tables and Figures (respol-tables-figures)
When to trigger
- The key result is buried in a wall-of-coefficients table no reader can parse
- A bibliometric/patent map or network is decorative — pretty but uninterpretable
- An event-study or dose-response plot is missing where the design demands one
- A qualitative paper has no data-structure table linking evidence to constructs
- A referee says exhibits "don't answer the question" or descriptive stats are missing/opaque
The Research Policy exhibits bar
RP exhibits serve an interdisciplinary readership, so they must be self-explanatory to an economist, a management scholar, and a policymaker alike. Each exhibit should answer one question and visibly support the contribution; the headline result deserves a focused exhibit, not a dump of every specification. Innovation data also carry specific exhibit conventions: descriptive statistics that reveal the skew and zeros typical of patent/citation data, transparent variable definitions, and maps/networks that are interpreted, not merely displayed.
Designing the core exhibits
Regression / estimation tables
- Lead with a table that isolates the headline innovation effect; relegate the full specification grid to robustness.
- Report coefficients with standard errors and the relevant model statistics; state the estimator, sample, fixed effects, and clustering in the notes so the table stands alone.
- For count models, report incidence-rate ratios or marginal effects where they aid interpretation — a raw NB coefficient is opaque to many RP readers.
- Provide a descriptive-statistics and correlation table that shows the distribution (means, SDs, and the share of zeros for count variables).
Patent / bibliometric exhibits
- Co-occurrence/citation networks and technology maps must have an interpretive payoff: annotate clusters, state the layout algorithm and the tie definition, and tell the reader what to see.
- Time-series of patenting/diffusion should mark policy dates or structural breaks relevant to the claim.
Causal-design plots
- Event-study plots with leads/lags and confidence bands for DID; first-stage and reduced-form plots for IV; RD plots with binned means and the fitted discontinuity.
Qualitative exhibits
- A data-structure table (1st-order codes → 2nd-order themes → aggregate dimensions) and a representative-quotes table that ties evidence to each construct.
General craft
- Every exhibit has a number, a self-contained title, complete notes (source, sample, units), and is referenced and interpreted in the text.
- Units and variable definitions are explicit; do not assume the reader knows your patent indicator.
- Figures should be legible in greyscale and at print size; avoid chartjunk and uninterpreted color.
- Place exhibits to follow the argument's logic; the appendix holds robustness, not load-bearing results.
Execution bridge (StatsPAI / Stata MCP)
Generate exhibits from the fitted result, not by retyping numbers. Full map:
execution-with-mcp. Research Policy is innovation studies — patent/firm panels with selection; foreground identification and the selection objection.
- 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
- A focused exhibit isolates the headline innovation effect
- Descriptive stats reveal skew/zeros in patent/citation variables
- Count-model results are reported in interpretable units (IRR / marginal effects) where helpful
- Patent/bibliometric maps are annotated and interpreted, not decorative
- Design-required plots (event study / first stage / RD) are present
- Qualitative data-structure and quotes tables link evidence to constructs
- Each exhibit stands alone (title + notes: source, sample, estimator, clustering)
- Every exhibit is interpreted in the text, not just cited
Anti-patterns
- A single mega-table where the headline result is one column among twenty
- A network/map shown without telling the reader what to conclude from it
- Raw negative-binomial coefficients with no interpretive translation
- Descriptive tables that hide the zero-inflation of innovation counts
- Color-dependent figures that fail in greyscale
- A qualitative paper with quotes scattered in prose but no data-structure table
Output format
【Journal】Research Policy
【Skill】respol-tables-figures
【Headline exhibit】what it isolates and how it supports the contribution
【Descriptives】skew/zeros of innovation variables shown? [Y/N]
【Interpretation】count units / map annotation / design plots present? [Y/N]
【Stand-alone notes】source, sample, estimator, clustering in each note? [Y/N]
【Qualitative】data-structure + quotes table? [Y/N / NA]
【Verdict】pass / revise / reroute
【Next skill】respol-writing-style
Version History
- 1839142 Current 2026-07-05 14:19
