jhr-tables-figures - SKILL.md Agent Skill

name: jhr-tables-figures description: Use when preparing Journal of Human Resources (JHR) tables, figures, online appendix exhibits, reconciliation tables, event-study and first-stage diagnostics, and policy-readable empirical displays that fit inside the page limit counting tables and figures.

Tables & Figures (jhr-tables-figures)

When to trigger

Results are ready but tables are dense or over the page limit
You need reconciliation, robustness, or design diagnostics in exhibit form
The Online Appendix needs a clear structure

Exhibit plan

Table 1: sample, descriptive statistics, and key balance where relevant.
Main table: preferred specification with transparent controls and clustering.
Design diagnostic: pre-trends, first stage, manipulation, balance, or event study depending on design.
Reconciliation table: compare your estimate to prior estimates and explain the bridge.
Appendix: robustness, sensitivity, alternative samples, and extra outcomes.

Notes must state

Unit, sample, period, outcome units
Fixed effects and controls
Clustering level
Treatment definition
Survey weights or population weights if used
Page/appendix location

Main-text exhibit budget

Use the scarce main-text pages for exhibits that change a reader's belief:

Sample and balance: proves the population and comparison are understandable.
Main estimate with magnitude: preferred result plus units and confidence interval.
Design diagnostic: pre-trend, first stage, manipulation test, balance, or attrition.
Reconciliation: prior estimate vs. your bridge specification vs. preferred specification.
Policy heterogeneity: only if it maps to a real policy margin, not a fishing cut.

Everything else belongs in the Online Appendix with clear cross-references.

Appendix map

Organize appendix exhibits by reviewer use, not by the order scripts happen to run:

Design validity: balance, pre-trends, manipulation, attrition, first stage, or placebo evidence.
Specification sensitivity: alternative controls, bandwidths, estimators, samples, weights, and clustering levels.
Reconciliation: bridge specifications that explain differences from prior estimates.
Mechanism and heterogeneity: only after the main effect and design validity are clear.
Data construction: variable definitions, sample filters, merges, missingness, and coding decisions.

Each appendix table should be referenced from exactly one main-text claim or robustness sentence. Orphaned appendix exhibits create page and credibility costs.

Event-study figure standard

The event-study plot is often the single most scrutinized exhibit in a JHR design paper. Hold it to this bar:

Name the estimator in the note (heterogeneity-robust group-time aggregation, interaction-weighted, or imputation — not just "event study").
Reference period marked (usually t = -1) and at least four pre-periods shown when the data allow; binned endpoints labeled as bins.
95 percent confidence intervals from SEs clustered at the assignment level, with the cluster count in the note.
Y-axis in outcome units, not standardized indices, so the policy reader can judge magnitude directly.
If TWFE and robust estimates diverge, plot both series rather than choosing silently.

First-stage and RD display conventions

IV papers: a first-stage table adjacent to the 2SLS table — coefficient, effective F per endogenous regressor, and the reduced form; referees read these three together.
RD papers: the binned outcome plot and the density plot are a pair; show the bandwidth on the figure and put manipulation-test results in the note.
Lottery papers: a balance exhibit within randomization strata precedes any effect figure.

Worked exhibit ledger

Illustrative ledger for a childcare-subsidy DID paper (titles invented):

Fig 1  Rollout map + timing of county adoption          claim: variation exists
Tab 1  Sample means, adopters vs not, pre-period        claim: comparability
Tab 2  ATT on maternal employment, 3 estimators         claim: main effect
Fig 2  Event study, 5 pre / 6 post, CIs, clusters=42    claim: no pre-trends
Tab 3  Bridge to prior state-level estimate             claim: reconciliation
Tab 4  Heterogeneity by single-parent status            claim: policy margin
App A  Sensitivity: windows, controls, clustering       referenced from Tab 2

Seven main exhibits is a sensible ceiling under the page cap; every appendix entry must be cited from one main-text sentence.

Execution bridge (StatsPAI / Stata MCP)

Generate exhibits from the fitted result, not by retyping numbers. Full map: execution-with-mcp. JHR is labor/education economics — program evaluation with selection; DiD/IV/RDD and the selection objection are central.

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.

Output format

[Exhibit] main table / diagnostic / reconciliation / appendix
[Claim] ...
[Required note fields] ...
[Page-limit action] keep / move to appendix / compress
[Next step] jhr-writing-style