statsmodels

Overview

Statsmodels provides classical statistical modeling with rigorous inference for Python. It covers linear models, generalized linear models, discrete choice, time series, and comprehensive diagnostics. Unlike scikit-learn (prediction-focused), statsmodels emphasizes coefficient interpretation, p-values, confidence intervals, and model diagnostics.

When to Use

• Fitting linear regression (OLS, WLS, GLS) with detailed coefficient tables and diagnostics
• Running logistic regression with odds ratios and marginal effects for clinical/epidemiological studies
• Analyzing count data with Poisson or negative binomial regression
• Time series forecasting with ARIMA, SARIMAX, or exponential smoothing
• Performing ANOVA, t-tests, or non-parametric tests with proper corrections
• Testing model assumptions (heteroskedasticity, autocorrelation, normality of residuals)
• Model comparison using AIC/BIC or likelihood ratio tests
• Using R-style formula interface (y ~ x1 + x2 + C(group)) for intuitive model specification
• For prediction-focused ML with cross-validation and hyperparameter tuning, use scikit-learn instead
• For Bayesian modeling with posterior inference, use pymc instead