drugclaw/stat-modeling-tools

Stat Modeling Tools

Use this skill when the user needs reproducible statistical analysis rather than only visual inspection.

Typical triggers:

• choose or run a hypothesis test on tabular data
• compare two groups or test association between variables
• fit OLS, logistic, or Poisson models with coefficient tables
• inspect residuals, p-values, confidence intervals, or effect sizes
• generate machine-readable statistical summaries for a manuscript or report

Environment Check

which python3 || true
python3 - <<'PY'
mods = ["numpy", "pandas", "scipy", "statsmodels"]
for name in mods:
    try:
        __import__(name)
        print(f"{name}: ok")
    except Exception as exc:
        print(f"{name}: missing ({exc})")
PY

If key modules are missing, say so explicitly and recommend the optional drug-sandbox image documented in docs/operations/science-runtime.md.

Bundled Assets

• templates/stat_test_report.py
• templates/statsmodels_regression.py

Preferred Workflow

1. Identify outcome type first: continuous, binary, count, or categorical contingency table.
2. Run a small deterministic statistical summary before fitting a larger model.
3. Report effect sizes and confidence intervals, not only p-values.
4. Save CSV and JSON outputs so the result is reusable.
5. Keep claim scope tied to the study design. Statistical association is not causal proof.

Hypothesis Tests

python3 templates/stat_test_report.py \
  --input stats/assay.csv \
  --test independent_ttest \
  --value-column response \
  --group-column arm \
  --group-a control \
  --group-b treated \
  --output stats/assay_ttest.csv \
  --summary stats/assay_ttest.json

Supported baseline tests in the bundled template:

• independent_ttest
• paired_ttest
• mannwhitney
• chi_square
• pearson
• spearman

Use this for quick but explicit statistical reporting.

Regression With Statsmodels

python3 templates/statsmodels_regression.py \
  --input stats/cohort.csv \
  --model ols \
  --outcome response \
  --feature age \
  --feature dose \
  --feature biomarker \
  --output stats/ols_coefficients.csv \
  --summary stats/ols_summary.json

Supported baseline models in the bundled template:

• ols
• logit
• poisson

Use this for:

• coefficient tables with confidence intervals
• basic inference and model-fit summaries
• prediction export for downstream review

Working Rules

• Prefer exact test names and explicit group labels.
• Check whether the data are paired before running paired tests.
• For regression, list the exact feature set and reference coding assumptions.
• Do not oversell significance when effect sizes are trivial.
• Distinguish exploratory testing from pre-specified confirmatory analysis.

Related Skills

For Kaplan-Meier, Cox models, and time-to-event workflows, activate survival-analysis-tools. For static or interactive figures, activate scientific-visualization-tools. For study design, reproducibility planning, or manuscript critique, activate scientific-workflow-tools or clinical-research-tools.