tooluniverse-statistical-modeling

tooluniverse-statistical-modeling

Statistical modeling skill for biomedical data analysis, designed to solve BixBench regression, survival, and statistical testing questions.

Capabilities

Model Type	Implementation	Key Outputs
Linear Regression (OLS)	statsmodels.formula.api.ols	Coefficients, R-squared, F-test, AIC/BIC
Binary Logistic Regression	statsmodels.formula.api.logit	Odds ratios, CIs, p-values, pseudo-R-squared
Ordinal Logistic Regression	statsmodels.miscmodels.ordinal_model.OrderedModel	Odds ratios, thresholds, proportional odds test
Multinomial Logistic Regression	statsmodels.MNLogit / sklearn.LogisticRegression	Per-category ORs, relative risk ratios
Mixed-Effects Models	statsmodels.formula.api.mixedlm	Fixed effects, random variance, ICC
Cox Proportional Hazards	lifelines.CoxPHFitter	Hazard ratios, CIs, concordance index
Kaplan-Meier Estimation	lifelines.KaplanMeierFitter	Median survival, survival curves, log-rank test
Statistical Tests	scipy.stats	t-test, chi-square, Fisher, ANOVA, Mann-Whitney, etc.

BixBench Question Coverage

The skill handles these BixBench question patterns:

1. Odds ratio extraction: "What is the odds ratio of X associated with Y?"
2. Ordinal logistic regression: "What is the OR using ordered logit model?"
3. Percentage reduction in OR: "What is the percentage reduction after adjusting for confounders?"
4. Interaction effects: "What is the interaction odds ratio?"
5. Hazard ratios: "What is the hazard ratio for treatment in Cox regression?"
6. Survival estimates: "What is the median survival time?"
7. Model coefficients: "What is the coefficient and 95% CI?"
8. Model comparison: "Which model fits better (AIC/BIC)?"
9. Adjusted vs unadjusted: "How does the OR change after adjustment?"
10. Statistical significance: "Is the association statistically significant?"

Dependencies

statsmodels>=0.14.0
scikit-learn>=1.3.0
lifelines>=0.27.0
pandas>=2.0.0
numpy>=1.24.0
scipy>=1.10.0

Files

File	Description
SKILL.md	Full skill specification with all phases and code patterns
QUICK_START.md	8 worked examples covering all model types
EXAMPLES.md	Detailed BixBench-style worked examples
TOOLS_REFERENCE.md	Package/function reference and decision tree
test_skill.py	85-test comprehensive test suite
README.md	This file

Testing

python3 test_skill.py

Test coverage: 15 sections, 85 tests, covering:

• Package imports (5 tests)
• OLS linear regression + diagnostics (8 tests)
• Binary logistic regression + odds ratios (7 tests)
• Ordinal logistic regression + proportional odds (7 tests)
• Multinomial logistic regression (5 tests)
• Mixed-effects models + ICC (6 tests)
• Cox proportional hazards (7 tests)
• Kaplan-Meier estimation + log-rank (5 tests)
• Statistical tests (9 tests)
• Confidence intervals (3 tests)
• Model comparison (3 tests)
• BixBench question patterns (7 tests)
• Edge cases and robustness (6 tests)
• Data loading and processing (3 tests)
• Effect size and interpretation (3 tests)

Result: 85/85 tests pass (100.0%)

Model Selection Guide

Outcome Type           -> Model
------------------------------------
Continuous             -> OLS (smf.ols)
Continuous + clusters  -> LMM (smf.mixedlm)
Binary                 -> Logistic (smf.logit)
Ordinal (3+ levels)    -> OrderedModel (distr='logit')
Nominal (3+ levels)    -> MNLogit
Time-to-event          -> Cox PH / Kaplan-Meier
Count data             -> Poisson / NB