jar-tables-figures

Tables & Figures (jar-tables-figures)

When to trigger

• Tables are cluttered, inconsistent, or not self-explanatory
• A referee cannot tell the sample, units, or SE clustering from the table notes
• You need the standard JAR exhibit set assembled in house style
• Identification needs a figure (pre-trends, RD plot) to be believed

The standard JAR exhibit set

An empirical-archival JAR paper is read through its tables. Build, in order:

1. Sample construction table — from the raw population to the final N, line by line, with each screen and the observations lost. Referees expect to trace the sample.
2. Descriptive statistics — N, mean, SD, and key percentiles for every variable; state winsorization (e.g., 1/99%).
3. Correlation matrix — Pearson (and often Spearman) among the main variables; flag significance.
4. Main results — the central regression(s): coefficients with t-/z-stats or standard errors beneath, the SE clustering stated in the note, fixed effects indicated, and N and R² (or pseudo-R²) reported.
5. Identification & robustness — first-stage (IV), DiD dynamics, RD estimates, falsification/placebo, alternative measures, and cross-sectional (channel) partitions.

House-style discipline

JAR uses a custom author-date house style; match the typographic conventions of recent JAR articles rather than importing a reference manager's defaults. For exhibits specifically:

• Self-contained: a title, the sample/period, the units, and the dependent variable are clear from the table and its note alone.
• Inference visible: report what is beneath the coefficients (t-stats / SEs) and state the clustering in the note; mark significance consistently.
• Variable definitions: every variable defined (often an appendix variable-definitions table) with its data source (Compustat/CRSP/I/B/E/S/Audit Analytics/EDGAR).
• Numbers consistent: Ns, coefficients, and signs match the text; decimal places consistent.

Figures that earn their place

Use figures where they do identification work a table cannot: DiD event-study plots (coefficients by period with confidence bands, showing flat pre-trends), RD plots (binned means around the cutoff), and time series of the treatment/setting. Avoid decorative charts; every figure should support the causal claim.

Checklist

• Sample-construction table traces raw population → final N
• Descriptives with winsorization stated; correlation matrix included
• Main table reports coefficients, inference statistics, FE, N, R²
• SE clustering stated in every regression-table note
• Variable-definitions table with data sources included
• Identification figure (pre-trends / RD) present where the claim is causal
• All numbers reconcile with the text; formatting matches recent JAR articles

Anti-patterns

• Mystery samples: a final N with no construction table.
• Naked coefficients: no SEs/t-stats and no clustering note.
• Reference-manager defaults instead of JAR house style.
• Decorative figures that do no identification work.
• Undefined variables or sources scattered through the text.

Output format

【Exhibit set】sample / descriptives / correlations / main / robustness present?
【Inference shown】t-stats or SEs + clustering stated in notes?
【Variable definitions】table with sources included?
【Identification figure】pre-trends / RD plot present where causal?
【Consistency】Ns and coefficients reconcile with text?
【House style】matches recent JAR articles?
【Next step】jar-writing-style

Resources

• ../../resources/official-source-map.md — official JAR/Chicago Booth/Wiley URLs (accessed 2026-06-01)
• ../../resources/external_tools.md — exhibit-export tooling (estout / outreg2 / modelsummary) and data sources