Aeja tables figures

Tables & Figures (aeja-tables-figures)

When to trigger

• The main result is settled and must be made readable at a glance
• Tables are dense, over-decimaled, or bury the headline coefficient
• An event-study / RD / first-stage plot needs to carry the identification visually
• You are preparing exhibits for submission and want them AEA-house-style compliant

The AEJ: Applied exhibit bar

At AEJ: Applied the main causal estimate should be findable in seconds and every exhibit should earn its place. AEA house style permits significance stars but expects standard errors in parentheses, clear notes that make each exhibit self-contained, and clean figures over chartjunk. Lead with the design's signature visual — the event-study plot, the RD scatter, or the balance table — because at this journal the picture of the identification is half the persuasion.

Exhibit	What it must show	Common failure
Main results table	headline coefficient, SE in parentheses, N, controls indicated, dependent-var mean	too many columns; no SEs; over-precision
Balance table (RCT)	baseline means by arm, differences, joint test	missing joint test; no attrition row
Event-study figure	leads + lags, CIs, reference period, flat pre-trends visible	no CIs; ambiguous reference period
RD figure	binned scatter + fitted lines, bandwidth, density	overfit polynomial; no density panel
First-stage / IV table	first-stage F, exclusion logic in notes	weak first stage hidden
Robustness exhibit	point-estimate stability across checks	a starred wall with no map

Exhibit craft

1. One table for the headline. Table 1 (or 2) should let a referee read the main causal estimate, its SE, and N without flipping pages.
2. Standard errors always, stars optional. AEA permits stars, but SEs in parentheses are the load-bearing object; report the dependent-variable mean so magnitudes are interpretable.
3. Self-contained notes. Each exhibit's note states sample, units, clustering level, controls, and what an asterisk (if used) means — a referee should not need the text to read the table.
4. Figures carry identification. Event-study leads, RD continuity, and balance are more convincing as figures than as prose; make them publication-clean (vector output, readable fonts, CIs shown).
5. Right precision. Two to three significant figures; do not report coefficients to five decimals.

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.

• Tables: etable (multi-model columns) or did_summary_to_latex straight from the result_id — one variable definition, one set of numbers, body and appendix in sync.
• 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 (from the result's diagnostics) and states the magnitude in interpretable units.

See a full fitted-result → exhibit chain in the JF execution walkthrough.

Checklist

• Main causal estimate readable in one table: coefficient, SE in parentheses, N, dep-var mean, controls flagged
• Standard errors reported everywhere; clustering level stated in notes; stars (if used) defined
• Identification figure present (event-study with CIs / RD with density / balance with joint test)
• Notes make every exhibit self-contained (sample, units, clustering, controls)
• Figures are clean vector output, legible, no chartjunk, consistent scales
• Precision sensible (2–3 sig figs); no redundant exhibits

Anti-patterns

• A main results table with 12 columns where the headline coefficient is hard to find
• Reporting stars or t-stats but omitting standard errors / clustering level
• Over-precision (coefficients to 5 decimals) implying false accuracy
• An event-study plot with no confidence intervals or unclear reference period
• An RD figure with a high-order global polynomial that manufactures a jump
• Exhibit notes that force the reader back to the text to interpret the table

Worked vignette (illustrative)

A draft's Table 4 has 12 columns sweeping every control combination, and the headline coefficient is buried in column 9 with only t-statistics shown. The AEJ: Applied fix: promote the preferred specification to a two-panel Table 2 — Panel A the main estimate (coefficient 3.1, s.e. 0.9 in parentheses, N, dependent-var mean 0.44), Panel B the same with the full controls — and move the sweep to the online appendix. Add Figure 1, the event-study with confidence intervals and a clearly marked reference period, so the identification is visible before the reader reaches the table. The result is now findable in seconds.

Referee pushback mapped to the exhibit fix

• "I cannot find your main estimate." → One headline table with the coefficient, SE, N, and dep-var mean; everything else demoted to the appendix.
• "Where are the standard errors / what is the clustering?" → SEs in parentheses everywhere; clustering level and (if used) star meaning stated in the self-contained note.
• "This RD jump looks like an artifact of the polynomial." → Replace the global high-order fit with a local-linear binned scatter plus a density panel.

Output format

【Headline exhibit】one table/figure carrying the main estimate? [Y/N]
【Inference shown】SEs in parentheses + clustering level in notes? [Y/N]; stars defined if used
【Identification figure】event-study / RD / balance present with CIs? [Y/N]
【Self-contained notes】sample/units/clustering/controls in every note? [Y/N]
【Figure quality】vector, legible, no chartjunk? [Y/N]
【Next step】aeja-writing-style