Statistical Analysis
Workflow inheritance
For complex tasks, first apply the global codex-workflow-protocol skill. This skill only adds domain-specific knowledge, gates, and validation requirements. It must not weaken the global completion, escalation, or verification rules.
Workflow
- Translate the user question into estimand, outcome, predictors/groups, unit of analysis, and dependency structure.
- Inspect data quality, missingness, measurement scale, outliers, and study design before choosing tests.
- Choose the simplest defensible analysis: descriptive summary, hypothesis test, regression, survival model, Bayesian model, or simulation.
- Check assumptions and identify robust or nonparametric alternatives.
- Report effect sizes, uncertainty intervals, p-values only when appropriate, multiplicity caveats, and practical significance.
- Validate by rerunning calculations, checking sensitivity choices, and making the code reproducible.
Completion boundary
Do not claim completion from a smoke simulation, tiny dryrun, notebook import, or one successful toy run alone. Simulation, inference, or theorem-heavy research coding is complete only when the estimand, model assumptions, data-generating or likelihood assumptions, and requested validation evidence are checked at the intended scale or a precise residual risk is reported.
Bayesian/domain-specific acceptance should include prior justification, convergence diagnostics where applicable, posterior predictive or sensitivity checks where applicable, and a clear distinction between exploratory and confirmatory claims.
References
- Read
references/statistical-decision-notes.md for test/model selection and reporting conventions.
- Read
references/legacy-full-skill.md for older examples and extended background.
- Use
statsmodels for Python model implementation, pymc for PyMC models, bayesian-ppl-diagnostics for convergence and PPL diagnostics, and simpy for process-based simulation.
Validation
- The analysis states assumptions, missing-data handling, and unit of analysis.
- Results distinguish exploratory, confirmatory, and sensitivity analyses.
- Budget warnings for full Bayesian/statistics domain installs are advisory only.
1---2name: statistical-analysis3description: Statistical Analysis4---5# Statistical Analysis67## Workflow inheritance89For complex tasks, first apply the global `codex-workflow-protocol` skill. This skill only adds domain-specific knowledge, gates, and validation requirements. It must not weaken the global completion, escalation, or verification rules.1011## Workflow12131. Translate the user question into estimand, outcome, predictors/groups, unit of analysis, and dependency structure.142. Inspect data quality, missingness, measurement scale, outliers, and study design before choosing tests.153. Choose the simplest defensible analysis: descriptive summary, hypothesis test, regression, survival model, Bayesian model, or simulation.164. Check assumptions and identify robust or nonparametric alternatives.175. Report effect sizes, uncertainty intervals, p-values only when appropriate, multiplicity caveats, and practical significance.186. Validate by rerunning calculations, checking sensitivity choices, and making the code reproducible.1920## Completion boundary2122Do not claim completion from a smoke simulation, tiny dryrun, notebook import, or one successful toy run alone. Simulation, inference, or theorem-heavy research coding is complete only when the estimand, model assumptions, data-generating or likelihood assumptions, and requested validation evidence are checked at the intended scale or a precise residual risk is reported.2324Bayesian/domain-specific acceptance should include prior justification, convergence diagnostics where applicable, posterior predictive or sensitivity checks where applicable, and a clear distinction between exploratory and confirmatory claims.2526## References2728- Read `references/statistical-decision-notes.md` for test/model selection and reporting conventions.29- Read `references/legacy-full-skill.md` for older examples and extended background.30- Use `statsmodels` for Python model implementation, `pymc` for PyMC models, `bayesian-ppl-diagnostics` for convergence and PPL diagnostics, and `simpy` for process-based simulation.3132## Validation3334- The analysis states assumptions, missing-data handling, and unit of analysis.35- Results distinguish exploratory, confirmatory, and sensitivity analyses.36- Budget warnings for full Bayesian/statistics domain installs are advisory only.
YuukiAS