nahisaho Data Analysis
Statistical analysis, visualization, and result interpretation.
Use This Skill When
- Running hypothesis tests (t-test, ANOVA, chi-square, etc.).
- Building regression or classification models.
- Performing exploratory data analysis (EDA).
- Creating publication-quality figures.
- Interpreting statistical results in research context.
Workflow
Data assessment:
- Check data structure, types, and dimensions
- Identify missing values, outliers, and distributional properties
- Validate assumptions for planned analyses
Analysis execution:
- Apply appropriate statistical methods
- Report effect sizes and confidence intervals (not just p-values)
- Run sensitivity analyses when assumptions are questionable
Visualization:
- Generate publication-quality figures (English text only)
- Use colorblind-friendly palettes
- Save all figures to
figures/
Interpretation:
- State what the results mean in research context
- Acknowledge limitations and alternative explanations
- Distinguish statistical significance from practical significance
Deliverables
report.md: analysis narrative with embedded figure references.results/statistical-summary.md: test results, effect sizes, CIs.figures/: publication-quality plots (English labels).data/: processed datasets when transformation occurs.
Quality Gates
- Statistical assumptions are checked before applying tests.
- Effect sizes and confidence intervals are reported alongside p-values.
- Figures use English text and colorblind-friendly palettes.
- Multiple comparisons are corrected (Bonferroni, FDR, etc.) when applicable.
- Limitations of the analysis are explicitly stated.
If any gate fails: identify the specific failing check, fix the issue, and re-validate before proceeding.
Gotchas
- p値だけでなく効果量と信頼区間を必ず報告すること。「p < 0.05 で有意」だけでは不十分
- 多重比較を行う場合は補正が必須。検定の数が3以上なら Bonferroni または FDR 補正を適用
- 外れ値の除外は根拠を明示すること。「見た目で除外」は再現性を損なう
- 図のテキストは必ず英語。日本語のラベルが入った図はジャーナル投稿で再作成が必要になる
- データの前処理手順は
data/preprocessing-log.mdに記録すること。処理の再現性を担保する
Validation Loop
- 分析結果を生成
- チェック:
- 仮定の検証（正規性、等分散性等）が行われているか
- 効果量と信頼区間が報告されているか
- 多重比較補正が必要な場面で適用されているか
- 図が英語ラベルで colorblind-friendly か
- 不合格なら該当箇所を修正して再分析
- 合格後のみレポート確定