Tables & Figures (psychbull-tables-figures)
A meta-analysis communicates through a small set of standard, expected exhibits. Psychological
Bulletin reviewers look for a PRISMA flow diagram, a forest plot, funnel/bias plots, and
MARS-ready summary tables, all in APA 7th-edition format. This skill covers exhibit design;
the numbers come from the analysis skills.
When to trigger
- Building the figures and tables for the manuscript
- A reviewer asks for the PRISMA diagram, forest plot, or a study-characteristics table
- Making exhibits self-contained, accessible, and reproducible from the deposited scripts
The expected exhibits
-
PRISMA flow diagram — records identified → deduplicated → screened → full-text assessed →
included, with exclusion counts and reasons at each stage (matches the search log).
-
Forest plot — each study's effect size and CI, its weight, and the pooled estimate with
CI and (ideally) a prediction interval; order meaningfully (by year, effect, or subgroup).
-
Funnel plot (contour-enhanced) and any bias-diagnostic plots (e.g., PET-PEESE regression).
-
Moderator / meta-regression bubble plot — effect vs. continuous moderator, point size ∝ weight.
-
Study-characteristics table — one row per study: design, n, population, measures, effect size,
moderator codes — the backbone of a MARS-compliant report.
-
Summary-of-findings table — pooled effect, CI, k, I²/τ², prediction interval, bias results.
APA 7th & accessibility
- APA 7th table/figure format: numbered, titled, with notes defining abbreviations.
-
Self-contained: readable without the text; define every symbol and effect-size metric.
-
Colorblind-safe palettes; legible in grayscale; vector output (PDF/EPS) for print.
- Exhibits must regenerate from the deposited scripts and match the reported numbers.
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. Psychological Bulletin is a meta-analytic review venue.
-
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
- No PRISMA diagram, or counts that don't reconcile with the search log
- A forest plot without weights, pooled estimate, or a prediction interval
- A funnel plot presented as proof of (no) bias on its own
- A study table missing the inputs needed to recompute effect sizes
- Figures that can't be regenerated from the deposited code (numbers drift from text)
Exhibit expectations at the APA review flagship
Psychological Bulletin referees scan the exhibits before the prose: a synthesis missing its standard
visual vocabulary signals an immature manuscript. The decision table they apply:
| Exhibit |
Referee expects |
Common desk-reject pattern |
| PRISMA flow |
Counts that reconcile to the search log and text |
Diagram numbers that don't add up to the reported k |
| Forest plot |
Per-study CI, weights, pooled diamond, prediction interval |
A bare list of dots with no pooled estimate or PI |
| Funnel/bias |
Contour-enhanced, paired with a formal test |
Funnel alone, captioned "no bias" — over-reading a picture |
| Study table |
One row per study, all effect-size inputs recoverable |
Missing ns/SDs, so effect sizes cannot be reverified |
| Summary-of-findings |
k, pooled g, CI, I²/τ², PI, bias result in one place |
Scattered numbers a reader must reassemble |
Worked vignette — what the exhibits must show
Illustrative figures, not real data. For the k = 42, g = 0.34, I² = 68% self-affirmation synthesis
above, the exhibit set this skill requires looks like:
-
PRISMA diagram: 2,310 identified → 1,640 after dedup → 1,640 screened → 188 full-text →
42 included; exclusion reasons tallied so the bottom box equals the analyzed k.
-
Forest plot: 42 rows ordered by year, each with g and CI and an inverse-variance weight; a pooled
diamond at 0.34 [0.24, 0.44] and a wider prediction interval roughly [−0.10, 0.78] that visibly
exceeds the CI — making the 68% heterogeneity legible at a glance.
-
Contour-enhanced funnel beside an Egger caption (p = 0.03), not standing alone as "proof."
-
Bubble plot for the delivery-format moderator, point size proportional to weight.
-
Summary-of-findings table: one line carrying k, g, CI, I², τ², prediction interval, and the
trim-and-fill / selection-model sensitivity bounds.
Referee pushback → venue-specific fix
-
"Your forest plot has no prediction interval." → Add the PI band; with I² = 68% the CI alone
understates the spread of true effects.
-
"PRISMA counts don't match the text." → Reconcile the diagram, the search log, and the reported k so
every box is auditable.
-
"The funnel plot is presented as evidence of no bias." → Re-caption as one diagnostic among several
and cross-reference the Egger/selection-model results.
Output format
【PRISMA diagram】present + reconciles with search log? [Y/N]
【Forest plot】weights + pooled + prediction interval? [Y/N]
【Funnel / bias plots】present? [Y/N]
【Study-characteristics table】MARS-complete? [Y/N]
【APA 7th + accessible】numbered, colorblind-safe, vector? [Y/N]
【Reproducible】regenerates from scripts? [Y/N]
【Next】psychbull-writing-style
Supplementary resources
