Jhe tables figures

Tables and Figures (jhe-tables-figures)

When to trigger

• The main result is settled and must be readable at a glance to a health economist
• Tables are dense, over-decimaled, or bury the headline policy effect
• An event-study / RD / first-stage plot needs to carry the identification visually
• The descriptive picture of the health setting (utilization, spending distribution, coverage) is missing or buried

The JHE exhibit bar

At JHE the main health-policy estimate should be findable in seconds, and the exhibits should let a health economist judge both the magnitude and the institutional plausibility of the result. Elsevier/JHE house style permits significance stars, but standard errors in parentheses are the load-bearing object and clustering must be stated. Two exhibit duties are specific to this journal: (1) a descriptive/institutional exhibit that shows the health setting — the coverage gap, the spending distribution (it is skewed — show it), the policy timeline — so readers trust the variation; and (2) the identification figure (event-study leads, RD continuity, first stage) that makes the design visible before the table.

Exhibit	What it must show	Common failure
Main results table	headline effect, SE in parentheses, N, clustering, dependent-var mean	too many columns; SEs missing; over-precision
Descriptive/institutional table	sample, coverage/utilization baseline, policy timing	generic summary stats with no health-system detail
Spending-distribution figure	skew, zero mass, where the effect sits in the distribution	mean-only reporting that hides the skew
Event-study figure	leads + lags, CIs, reference period, flat pre-trends	no CIs; ambiguous reference period
RD figure	binned scatter + local-linear fit, bandwidth, density panel	overfit global polynomial; no density
Heterogeneity exhibit	effects by clinically/policy-relevant subgroup with MHT	a starred subgroup fishing wall

Exhibit craft

1. One table for the headline. The preferred specification — effect, SE in parentheses, N, dependent-var mean, clustering level — should be readable without flipping pages; demote the controls sweep to the online appendix.
2. Show the distribution, not just the mean. Health spending and utilization are right-skewed and zero-heavy; a distribution figure or quantile effects tell the policy story a mean hides.
3. Make the setting legible. A descriptive exhibit naming the program, eligibility, and timing earns referee trust that the variation is what you say it is.
4. Figures carry identification. Event-study leads, RD continuity, and the first stage are more convincing as clean vector figures than as prose.
5. Right precision and self-contained notes. Two to three significant figures; each note states sample, units, clustering, controls, outcome definition, and (if used) what a star means.
6. Name the outcome in plain terms. "Any inpatient admission (0/1)" beats an opaque variable label; a health-policy reader should know exactly what was measured from the exhibit alone.

Execution bridge (StatsPAI / Stata MCP)

Generate exhibits from the fitted result, not by retyping numbers. Full map: execution-with-mcp. JHE is health economics — insurance/program reforms and selection; foreground DiD/IV/RDD and selection corrections.

• Tables: etable (multi-model) 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 magnitude in interpretable units.

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

Checklist

• Headline effect readable in one table: coefficient, SE in parentheses, N, dep-var mean, clustering stated
• A descriptive/institutional exhibit makes the health setting and policy timing concrete
• Skewed/zero-inflated outcomes shown as a distribution, not only a mean
• Identification figure present (event-study with CIs / RD with density / first stage)
• Heterogeneity reported by policy-relevant subgroup with MHT, not subgroup fishing
• Notes self-contained (sample, units, clustering, outcome definition, controls)
• Figures clean vector output; precision 2–3 sig figs; no redundant exhibits
• Outcome variables named in plain terms; magnitudes shown against a base rate

Anti-patterns

• A summary-statistics table with no health-system detail, so the variation looks generic
• A 12-column results table where the policy effect is buried in column 9
• Reporting stars or t-stats but omitting standard errors / clustering level
• Mean-only reporting that hides the skew and zero mass in spending/utilization
• An event-study with no confidence intervals or an unclear reference period
• An RD figure with a high-order global polynomial manufacturing a jump
• A heterogeneity wall of starred subgroups with no MHT and no clinical/policy logic
• An exhibit whose magnitude has no base rate, so the reader cannot judge whether it is large

Referee pushback mapped to the exhibit fix

• "I cannot find your main estimate." → One headline table with the effect, SE, N, clustering, and dep-var mean; the controls sweep demoted to the appendix.
• "Where are the standard errors and what is the clustering?" → SEs in parentheses everywhere; clustering level (usually state) and any star meaning stated in the self-contained note.
• "Your mean effect hides what happened in the tail." → Add a distribution or quantile-effect figure; health spending lives in the right tail, and the mean can understate or mask the policy story.
• "This RD jump looks like a polynomial artifact." → Replace the global high-order fit with a local-linear binned scatter plus a density panel.
• "I cannot tell whether your variation is credible." → Add the descriptive/institutional exhibit: eligibility, payment rule, and policy timing, so the design is plausible before the table.

Worked vignette (illustrative)

A coverage paper's Table 4 sweeps every control combination across 12 columns; the headline take-up effect hides in column 9 with only t-stats. The JHE fix: promote the preferred specification to a two-panel Table 2 (Panel A: effect 4.1pp, s.e. 1.0 in parentheses, N, dep-var mean, clustered on state; Panel B: with full controls), move the sweep to the online appendix, and add Figure 1 — the event-study with CIs and a marked reference period. Add Figure 2 showing the spending distribution shift, since the mean effect understates the policy story in the right tail. The result and its design are now legible in seconds.

Magnitudes a health-policy reader can act on

A JHE exhibit is persuasive when the reader can translate the number into a policy statement. Always give the base rate alongside the effect (a 4.1pp coverage gain reads differently against a 62% baseline than a 20% one), report dollar magnitudes for spending in current units, and reserve QALY/mortality language for effects you actually estimated. For heterogeneity, organize by the subgroup a regulator cares about (income band, age-eligibility, chronic-condition status), not by whatever split happens to be significant. The exhibit should let a policymaker read off "who was affected and by how much" without the prose.

Output format

【Journal】Journal of Health Economics
【Skill】jhe-tables-figures
【Headline exhibit】one table carrying the policy effect? [Y/N]
【Inference shown】SEs in parentheses + clustering level in notes? [Y/N]; stars defined if used
【Setting exhibit】descriptive/institutional + distribution shown? [Y/N]
【Identification figure】event-study / RD / first stage with CIs? [Y/N]
【Heterogeneity】policy-relevant subgroups with MHT? [Y/N]
【Next skill】jhe-writing-style

Handoff boundary

This skill makes a settled result legible; it does not establish the result (jhe-identification / jhe-robustness) or write the surrounding argument (jhe-writing-style). Do not polish exhibits while the estimate is still moving — that is wasted effort and it tempts presentation choices that flatter an unstable number. When the headline and identification figures are clean and self-contained, hand off to jhe-writing-style.