Agent Skills
› brycewang-stanford/Awesome-Journal-Skills
› jep-evidence-without-equations
jep-evidence-without-equations
GitHub指导在JEP文章中用直觉、精选数据和诚实局限呈现证据，避免数学公式。通过量化效应大小、综合多项研究、明确识别策略与外部有效性，增强非专业读者的理解与信任，提升论证可信度。
触发场景
需在正文中展示实证或理论证据但不使用回归表
草稿过度依赖统计显著性表述或系数罗列
数据众多需筛选关键数字以突出核心发现
担心简化数学处理会削弱证据说服力
安装
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill jep-evidence-without-equations -g -y
SKILL.md
Frontmatter
{
"name": "jep-evidence-without-equations",
"description": "Use when presenting empirical or theoretical evidence credibly in a Journal of Economic Perspectives (JEP) article with minimal math — intuition, examples, well-chosen numbers, and honest caveats. Handles how evidence is conveyed in prose; defer figures\/tables to jep-exhibits-for-general-readers and fairness across studies to jep-balance-and-objectivity."
}
Evidence Without Equations (jep-evidence-without-equations)
When to trigger
- You need to convince a reader of a finding without a regression table in the body
- The draft leans on "significant at the 1% level" or coefficient dumps to make its case
- You have many numbers and aren't sure which few to feature
- You worry that going light on math will make the evidence look soft
The JEP evidence bar
JEP communicates evidence through intuition, well-chosen numbers, and honest caveats, not through equations and estimation tables in the body. The credibility comes from which evidence you feature, how clearly you convey magnitude, and how honestly you state limits — not from displaying machinery. A JEP reader should finish a paragraph knowing the size of an effect, how sure the field is, and what could be wrong.
How to present evidence credibly with little math
- Lead with magnitude in plain units. "Roughly a third of the cost is passed through to prices" beats a coefficient. Translate elasticities/log points into amounts a reader feels (dollars, percentage points, days).
- Pick a few load-bearing numbers. Two or three vivid, well-sourced figures land; a wall of estimates does not. Each featured number should earn its place in the argument.
- Convey precision plainly. Instead of asterisks, say how settled it is: "estimates cluster between 5% and 8%," or "studies disagree, ranging from near zero to large." Give the reader a sense of the range, not a star.
- Use examples and analogies. A concrete case or a familiar analogy makes a mechanism stick where a model would not.
- Characterize the body of evidence, not one study. "A dozen studies across countries find…" is more JEP than "Smith (2019) estimates…". Synthesis means describing the weight and consistency of evidence.
- State caveats in the same breath. "This holds for short horizons; over decades the evidence thins" builds trust and is the JEP norm.
Honesty devices (build credibility without machinery)
- Name the identification in words, once. A reader should know why a number is believable (a natural experiment, a long literature, a clean comparison) without seeing the design.
- Distinguish settled from contested. Tell the reader what is consensus, what is debated, and what is unknown — this is also the bridge to
jep-balance-and-objectivity. - Flag external validity. Say where a result travels and where it likely doesn't.
- Avoid false precision. "About 6%" is more honest than "6.23%" when the range is wide.
- Anchor with a benchmark. A number means more next to a familiar reference point: "about the size of a typical recession's effect on employment," or "roughly a month's rent." Benchmarks let a non-specialist judge whether an effect is big or small.
Checklist
- Effect sizes given in plain, feelable units (not raw coefficients)
- Only a few load-bearing numbers featured, each well-sourced
- Precision conveyed as ranges / degree of agreement, never asterisks
- Evidence characterized as a body (weight, consistency), not one study
- Identification/credibility explained in words, once, not displayed
- Caveats and external-validity limits stated alongside the claims
- Key magnitudes anchored against a familiar benchmark
- No false precision; magnitudes rounded to what the evidence supports
Worked vignette (illustrative)
A draft says: "The treatment effect is 0.184 log points (s.e. 0.061), significant at the 1% level (***)." A JEP rewrite drops the asterisks and the raw coefficient: "Workers in the program earned about 20% more a year later — an effect large enough to matter for a household budget, and one that a dozen studies in different settings find consistently, though the gains fade after a few years." The reader now has magnitude in feelable units, a sense of how settled it is (many consistent studies), and the key caveat (fading) — all without an equation or a star.
Anti-patterns
- A regression table or estimating equation in the body to "prove" a point
- "Significant at the 1% level" (and asterisks) standing in for magnitude and precision
- A blizzard of numbers with no hierarchy — the reader can't tell which matter
- Citing a single study as if it settled a question (synthesis means weighing many)
- Hiding caveats to make the evidence look stronger (a balance and an honesty failure)
- False precision ("6.23%") that overstates how much the field actually knows
Output format
【Headline magnitude (plain units)】[...]
【Load-bearing numbers (≤3)】1) … 2) … 3) … (sources)
【Precision conveyed as】[range / degree of agreement]
【Why believable (in words)】[natural experiment / long literature / clean comparison]
【Caveats + external validity】[...]
【Body of evidence framed?】many studies vs. one: [Y/N]
【Next step】jep-exhibits-for-general-readers
版本历史
- 1839142 当前 2026-07-05 13:34
