技能 数据科学 空间计量经济学图表设计指南

空间计量经济学图表设计指南

v20260724
jue-tables-figures
这是一份为顶级经济学期刊(尤其是城市经济学领域)量身打造的空间结果展示指南。它指导用户如何设计具有高度说服力的地图、回归表格和事件研究图,核心目的是视觉化展示研究的“识别变异”(如处理边界、不连续点)。强调图表必须服务于经济学论证,而不是简单的数据装饰。
获取技能
105 次下载
概览

Spatial Exhibits: Maps, Tables & Figures (jue-tables-figures)

When to trigger

  • The paper has a spatial result but no map — the reader cannot see where the variation is
  • A boundary/RD or event-study design is in tables only; the figure that would convince is missing
  • Tables are dense with controls and the headline coefficient is hard to find
  • The geography of treatment vs control is described in words, not shown
  • Reviewers cannot tell whether the spatial pattern is real or an artifact of the units chosen

Maps earn their place — but only when they carry information

JUE is one of the few economics journals where a map is often the central exhibit, because the identifying variation is geographic. But a map must do econometric work, not decorate:

  • Show the identifying variation, not just the data. Map the treatment boundary, the corridor, the discontinuity, the shift-share exposure — the thing your design exploits — so the reader sees what generates the estimate.
  • Make scale and units explicit. State the spatial unit (tract, block, grid cell), include a scale bar and north arrow, and pick a classification (quantile vs equal-interval) that does not manufacture a pattern.
  • Pair the map with the design figure. A boundary map plus an RD plot; a treatment map plus an event-study; an exposure map plus the first stage.

Exhibit-by-design checklist

Design The figure that convinces The table that supports it
Boundary / spatial RD RD plot in distance-to-boundary with binned means + local-linear fit; covariate-smoothness panel RD estimates across bandwidths; density test
Place-based DiD / event study event-study plot with leads/lags and CIs; treatment-geography map heterogeneity-robust ATT vs TWFE; Bacon decomposition
Shift-share / Bartik map of exposure; first-stage scatter Rotemberg-weight / BHJ diagnostics; second stage
Capitalization / housing map of price gradient or treatment area hedonic estimates; spatial-SE comparison
QSM counterfactual map of the counterfactual spatial reallocation parameter table with identification source; sensitivity

Table craft for JUE

  1. Headline coefficient first, controls collapsed to "Yes/No" rows — the reader should find the estimate in two seconds.
  2. Report spatial standard errors (Conley / spatial cluster) alongside or instead of naive; note the distance cutoff in the table notes.
  3. Self-contained notes: sample, geography, FE level, clustering, units. A JUE table should be readable without the text.
  4. House style: follow the journal's significance-reporting convention (检索于 2026-06;以官网为准); always show standard errors, and prefer reporting the magnitude in interpretable units (% capitalization, elasticity).
  5. Figures legible in grayscale for print; do not encode the only signal in color.

Execution bridge (StatsPAI / Stata MCP)

Generate exhibits from the fitted result, not by retyping numbers. Full map: execution-with-mcp. JUE is urban/spatial economics — sorting and spatial dependence; identification + Conley/spatial-robust inference.

  • 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

  • A map shows the identifying variation (boundary / corridor / exposure), not just raw data
  • Map has spatial unit stated, scale bar, north arrow, and a non-misleading classification
  • The design's convincing figure (RD plot / event-study / first-stage) is present
  • Headline coefficient is immediately findable; controls collapsed to indicator rows
  • Spatial SEs reported with the distance cutoff noted
  • Table notes are self-contained (sample, geography, FE, clustering, units)
  • Figures remain legible in grayscale

Anti-patterns

  • A spatial paper with no map, or a map that shows the data but not the identifying variation
  • A map whose classification (bins/colors) manufactures a visual pattern the estimate does not support
  • Burying the headline coefficient under fifteen control rows
  • Naive standard errors in the table for geographically clustered data
  • An event-study or RD reported only as a table when the figure is the convincing object
  • Color-only encoding that disappears in print or for color-blind readers

Common map mistakes that draw referee fire

  • Classification gaming. Quantile bins on a skewed variable can manufacture a stark gradient; show the result is not an artifact of the binning, and prefer a classification the reader can interpret.
  • The wrong unit. Mapping at a coarse unit (county) when the identification is at a fine unit (tract/parcel) hides the variation that does the work; map at the scale of the design.
  • Projection distortion. An unstated or inappropriate CRS distorts distances and areas — fatal when the design is distance-based; state the projection.
  • Decorative maps. A choropleth of the raw outcome with no treatment boundary or exposure overlay shows the data but not the identifying variation; every map should advance the argument.

What belongs in the appendix vs the main text

JUE main text carries the map and the design figure that convince; the appendix carries the supporting battery. Put in the main text: the treatment-geography map, the RD/event-study/first-stage figure, the headline table. Push to the appendix: full robustness tables across scales and radii, covariate-balance batteries, alternative-classification maps, and the spatial-SE-cutoff grid. Do not let the appendix carry an exhibit the main claim depends on — the editor and first referee may not reach it.

Worked vignette (illustrative)

A boundary-discontinuity paper on school-zone capitalization first presents only a regression table. The JUE exhibit upgrade: (1) a map of the attendance boundary with the price gradient, so the reader sees the discontinuity is geographic and not confounded by a highway; (2) an RD plot in distance-to-boundary with binned means and a local-linear fit, the visual that makes the jump credible; (3) a covariate-smoothness panel showing pre-determined characteristics do not jump; (4) a table with Conley SEs (1km cutoff) and the estimate in percent. The map + RD plot carry the paper.

The one-figure test

Editors and referees often form an impression from a single exhibit. Ask: if a reader saw only one figure from your paper, which would it be, and does it convey both the geography and the result? For a boundary design it is the RD plot in distance-to-boundary; for a place-based policy it is the event-study with the treatment map inset; for an agglomeration paper it is the exposure map paired with the first-stage. Engineer that figure to be self-explanatory — labeled axes in interpretable units, the identifying variation visible, a caption that states the claim — because it is doing more persuasive work than any table. A paper whose single best figure is a generic choropleth has not yet found its convincing exhibit.

Output format

【Map】shows identifying variation? unit/scale-bar/classification ok? [Y/N]
【Design figure】RD plot / event-study / first-stage present? [Y/N]
【Headline table】coefficient findable; controls collapsed? [Y/N]
【Spatial SEs】Conley/spatial cluster in tables, cutoff noted? [Y/N]
【Notes】self-contained (sample/geography/FE/clustering/units)? [Y/N]
【Grayscale】figures legible without color? [Y/N]
【Next skill】jue-writing-style
信息
Category 数据科学
Name jue-tables-figures
版本 v20260724
大小 8.25KB
更新时间 2026-07-28
语言