Tables & Figures (io-tables-figures)

Exhibits are where an expert IR reviewer checks whether the result is real — and where IO's editorial staff will later confirm the numbers reproduce. At IO they also cost words: the word limit includes tables, figures, and notes (it excludes the bibliography). So every exhibit must pull its weight, and every value must match the deposited code.

When to trigger

• Designing the main results table/figure or a key descriptive/measurement exhibit
• Deciding what belongs in the article vs. the supplementary material (which should rarely exceed ~20 pp)
• A reviewer found an exhibit unclear, mislabeled, or non-self-contained
• Trimming to the word cap by moving exhibits to supplementary material

Principles

1. State the level explicitly. IR exhibits are misread when the unit is implicit. Put the

unit of analysis (state-year, directed-dyad-year, IGO, crisis) in the title or note, with the sample, N, time span, and what the quantity is. A dyadic coefficient and a monadic one look identical on a plot but mean different things.

1. Plot effects; tabulate sparingly. For the headline international effect, a coefficient/forest

plot, a marginal-effects or predicted-probability plot, or an event-study/RD plot conveys magnitude and uncertainty better than columns of coefficients. Always show the interval.

1. Survive print and color limits. Use colorblind-safe palettes that stay legible in grayscale; drop

3D, chartjunk, and decorative color. A referee skimming the PDF must read it in seconds.

1. Decide article vs. supplementary by argument weight. Keep only the exhibits that carry the IR

claim in the article; relegate balance tables, full specifications, robustness grids, and proof-supporting tables to supplementary material (which should rarely exceed twenty pages, especially at first submission). Because the word count counts exhibits, this is also how you land under 14,000 / 8,000 / 10,000 words.

1. Tie every number to the driver script. Each exhibit is regenerated by the analysis pipeline and

matches the deposited package — IO staff re-run it before final acceptance (see io-transparency-and-data-policy).

IR-specific exhibits

• Maps for geographic/spatial variation (conflict locations, alliance/trade ties); network diagrams for

alliance, trade, or IGO-membership structure where structure is the point.

• Dyad-level coefficient plots; conjoint AMCE plots and foreign-policy survey-experiment effect plots.
• For qualitative IR work: timelines of crises/negotiations, causal-process diagrams, evidence tables

linking claims to archival/interview sources.

Exhibit-placement decision table (article vs. SI, word-aware)

Because IO counts exhibits toward the cap and the SI carries a ~20-page norm, every exhibit needs a placement verdict tied to argument weight:

Rule of thumb: if removing an exhibit would not weaken the core IR claim, it belongs in the SI.

Worked exhibit vignette (illustrative)

A treaty-compliance paper drafts five tables, with the headline result in a dense six-column dyad-level table. The IO redesign converts it to one coefficient plot with 95% intervals, titled "Effect of delegated monitoring on subsequent compliance, directed-dyad-year, 1990–2015, N≈42,000" (numbers illustrative) so the unit, sample, span, and quantity are self-contained — a referee reads magnitude and uncertainty in seconds. The four remaining tables move to supplementary material, keeping the main text under cap. Every plotted value is regenerated by the master script to match the package IO staff will re-run.

Execution bridge (StatsPAI / Stata MCP)

Generate exhibits from the fitted result, not by retyping numbers (the usual source of body-vs-supplement drift). Full map: execution-with-mcp. International Organization is IR — country/dyad panels with difficult identification; foreground the source of variation and robustness to alternative explanations.

• 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.

Anti-patterns

• Leaving the unit/level implicit so a dyadic estimate reads like a monadic one
• An exhibit only intelligible after reading the prose around it
• Significance stars with no effect size or interval
• Pushing every robustness grid into the main text (it costs words; the SI also has a ~20-page norm)
• Color-only encoding that collapses in grayscale or for colorblind readers
• Printed values that diverge from the deposited code output (the IO re-run will flag it)
• A six-column dyadic table where one interval plot would convey the magnitude and save words

Output format

【Main exhibit】what it shows + why a figure/table
【Self-contained?】title + labels + note + N/units/level/time present? [Y/N]
【Accessible?】grayscale-legible + colorblind-safe? [Y/N]
【Article vs supplementary】split decided, word-budget + ~20-page SI noted
【Reproducible?】generated by master script, matches package for IO verification? [Y/N]
【Next】io-writing-style

Supplementary resources

• ../../resources/external_tools.md — plotting/mapping/network packages for IR data
• ../../resources/official-source-map.md — word-count inclusion rule and supplementary-material guidance