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:
-
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.
-
Descriptive statistics — N, mean, SD, and key percentiles for every variable; state winsorization (e.g., 1/99%).
-
Correlation matrix — Pearson (and often Spearman) among the main variables; flag significance.
-
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.
-
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.
Execution bridge (StatsPAI / Stata MCP)
Generate exhibits from the fitted result, not by retyping numbers (the usual source of
body-vs-appendix drift). Full map: execution-with-mcp. JAR is archival/empirical accounting; foreground identification around disclosure and regulation shocks, with modern DiD where adoption is staggered.
-
Tables:
etable (multi-model columns) or did_summary_to_latex straight from the
result_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 effect size in
interpretable units.
See a full fitted-result → exhibit chain in the JF execution walkthrough.
Checklist
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
