TWBG_Materials/文档润色流和知识库构建流/claude-scholar-upstream/skills/results-analysis/references/statistical-reporting.md

Statistical Reporting Standard

Minimum reporting package

For every major comparison, report:

• metric definition and direction,
• unit of analysis,
• sample size / run count,
• descriptive statistics,
• uncertainty estimate,
• inferential test,
• effect size,
• correction strategy when multiple contrasts exist,
• limitation if assumptions or sample size are weak.

Required fields

Descriptive

• mean ± std when repeated runs are comparable
• 95% CI when inference is discussed
• median / IQR when distribution is strongly non-normal

Inferential

• exact test name
• test statistic and degrees of freedom when applicable
• p-value format
• effect size
• correction method for multiple comparisons

Do not do these

• report only best run
• report only p-values
• hide non-significant comparisons
• treat unstable trends as conclusions
• switch tests without stating why

Default wording rule

Use three layers:

1. Observation — what changed numerically
2. Support — what the test/effect size says
3. Boundary — what remains uncertain