wber-tables-figures
GitHub为世行经济评论稿件设计并审计图表，确保符合40页限制且面向经济学家与政策制定者。核心要求包括：每图一问题、报告标准误而非星号、将系数转化为政策含义单位、用图形展示动态效应，并将次要结果移至附录以节省空间。
Trigger Scenarios
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill wber-tables-figures -g -y
SKILL.md
Frontmatter
{
"name": "wber-tables-figures",
"description": "Use when building the exhibits for a The World Bank Economic Review (WBER) manuscript so they read cleanly for a mixed economist + policymaker audience, fit the 40-page cap, and carry the policy magnitude. Designs and audits tables\/figures; it does not run the analysis or write the prose."
}
Tables and Figures (wber-tables-figures)
When to trigger
- Tables are dense regression dumps a policymaker cannot read
- Significance is shown with asterisks instead of standard errors / confidence intervals
- Effect sizes are reported only in coefficient units, not policy-meaningful magnitudes
- Exhibits are pushing the paper over the 40-page total cap
- An event-study or treatment effect is buried in a table when a figure would carry it
The WBER exhibit standard
WBER exhibits serve two readers at once: the applied economist who checks the design and the practitioner who wants the magnitude and its policy meaning. The best WBER tables and figures make the headline effect legible in policy units — percentage-point change in enrollment, dollars of consumption, cost per outcome — not just a coefficient. And remember the 40-page total cap includes tables, figures, references, and appendices: every exhibit competes for scarce space, so each must earn its place. Lead with the design (balance, first stage, pre-trends) and then the effect; do not bury the main result in column 6 of a kitchen-sink table.
Building exhibits that work
- One question per exhibit. A table that answers "is the design valid?" should not also try to answer "how big is the effect?" Split them.
- Report SEs / CIs, never asterisks. WBER, like serious applied journals, expects standard errors (and ideally confidence intervals) so readers judge magnitude and precision, not a star count. Note the clustering level in the table.
- Translate to policy units in the table or note. Beside the coefficient, give the effect as a share of the control mean, a percentage-point change, or a cost-effectiveness figure. State the units and the baseline.
- Figures for designs and dynamics. Event-study leads/lags, RD scatter with fitted lines and bins, and dose-response curves belong in figures — they show pre-trends, manipulation, and functional form at a glance.
- Self-contained notes. Each note states the sample, the data source (LSMS/DHS/admin), the estimator, the clustering, and the units. A practitioner skimming exhibits should understand the paper without the text.
- Space discipline for the cap. Move secondary specifications to the supplementary appendix; keep the main paper's exhibits to the load-bearing ones.
Exhibit-by-purpose map
| Purpose | Best exhibit | Must show |
|---|---|---|
| Design validity (RCT) | Balance table | covariate means by arm, normalized differences, attrition |
| Design validity (DiD) | Event-study figure | flat leads, dynamic effects, CI bands |
| Design validity (RD) | RD scatter + density plot | binned means, fitted lines, no density jump |
| Design validity (IV) | First-stage table | effective F, exclusion-falsification |
| Main effect | Compact results table | point estimate, SE/CI, clustering, control mean, policy-unit translation |
| Mechanism / heterogeneity | Coefficient/forest plot | subgroup effects with CIs, MHT-adjusted |
| Cost / policy magnitude | Small summary table or note | cost per outcome, benefit-cost, fiscal scale |
Referee pushback mapped to the exhibit fix
- "I can't tell how big this effect is." → Add the control mean and a policy-unit translation (share of mean / pp / cost per outcome) beside the coefficient.
- "Where's the evidence the design is valid?" → Promote balance / pre-trends / density / first-stage from text to a dedicated exhibit.
- "The asterisks tell me nothing about magnitude." → Replace with standard errors and confidence intervals; state the clustering level.
- "This table has too much in it." → Split into a design exhibit and an effect exhibit; move nuisance columns to the appendix.
- "The paper is over length." → Audit every exhibit against the 40-page cap; keep only load-bearing ones in the main text.
Execution bridge (StatsPAI / Stata MCP)
Generate exhibits from the fitted result, not by retyping numbers. Full map:
execution-with-mcp. WBER is development economics — RCTs and observational designs in low/middle-income settings; randomization inference + DiD/IV, magnitude in policy units.
- 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
- Each exhibit answers exactly one question
- SEs / CIs reported; no significance asterisks; clustering level stated
- Headline effect translated into policy units (share of mean / pp / cost per outcome)
- Design diagnostics (balance / pre-trends / density / first stage) shown as exhibits
- Event-study, RD, and dose-response shown as figures, not buried in tables
- Notes are self-contained: sample, source, estimator, clustering, units
- Main-text exhibits trimmed to load-bearing ones; secondary ones in the appendix (40-page cap)
Anti-patterns
- Significance asterisks instead of standard errors / confidence intervals
- A kitchen-sink table where the main effect hides among nuisance coefficients
- Coefficients reported with no baseline mean or policy-unit translation
- An event-study or RD presented as a table when the figure tells the story
- Notes that omit the data source, clustering, or units
- Pushing the paper over 40 pages with redundant exhibits that belong in the appendix
Worked vignette (illustrative)
A draft reports a transfer program's effect in one wide table: 14 columns, asterisks everywhere, coefficient = 0.08 with no context. The WBER rebuild: Table 1 is balance (means by arm, normalized differences, 6% attrition, balanced). Table 2 is the main effect — one preferred specification, coefficient 0.08 (s.e. 0.02, clustered at village), control-mean consumption stated, with a note translating it to a 9% increase and a cost of about $34 per 10% consumption gain (illustrative). Figure 1 is the event study showing flat pre-trends. The kitchen-sink columns move to the appendix. The result is now legible to both an econometrician and a finance ministry.
Reading exhibits for two audiences
A WBER exhibit is tested by a "two-reader skim": hand the tables and figures (without the text) to an econometrician and to a policy analyst.
- The econometrician should be able to verify the design from the exhibits alone — balance, pre-trends, first stage, clustering — and judge whether the estimate is credible.
- The policy analyst should be able to read off the magnitude in units they care about — how many more children enrolled, how many dollars, at what cost — without decoding a coefficient.
If either reader is lost, the exhibit set has failed. The note line is the bridge: it must state the data source (LSMS/DHS/admin), the estimator, the clustering level, the sample, and the units, so each reader can self-serve.
Output format
【Exhibit inventory】design / main / mechanism / cost — one question each
【Inference display】SE/CI shown, no asterisks, clustering noted? [Y/N]
【Policy-unit translation】effect as share of mean / pp / cost per outcome? [state]
【Figures】event-study / RD / dose-response as figures? [Y/N]
【Self-contained notes】source + estimator + clustering + units? [Y/N]
【Page-cap discipline】main exhibits trimmed; secondary in appendix? [Y/N]
【Next step】wber-writing-style
Version History
- 1839142 Current 2026-07-05 14:31
