statsmodels

What it does

Guides statistical modeling and econometric analysis using Python tools for estimation, inference, and diagnostics across methods like linear models, GLMs, discrete outcomes, and time series. It includes practical code examples for common workflows such as OLS, logistic regression, and ARIMA, plus a broad set of diagnostic and model-evaluation techniques.

How it works

• Recommends fitting regression models (OLS, WLS, GLS, GLSAR, Quantile Regression, Mixed Effects, Recursive/Rolling).
• Supports generalized linear models with multiple families (Binomial, Poisson, Negative Binomial, Gamma, Inverse Gaussian, Gaussian, Tweedie) and various link functions (Logit, Probit, Log, Identity, Inverse, etc.).
• Provides discrete choice models (Logit, Probit, MNLogit, Conditional Logit, Ordered Model; Poisson, Negative Binomial, Zero-Inflated, Hurdle) and outputs such as marginal effects, predicted probabilities, and model evaluation metrics.
• Describes time series capabilities (ARIMA, SARIMAX, VAR, VARMAX, Exponential Smoothing, ETS; with multivariate and advanced state-space options) and forecasting with prediction intervals.
• Includes extensive diagnostics and tests for residuals, influence, heteroskedasticity, autocorrelation, normality, and model specification, plus information criteria (AIC/BIC) and likelihood-based testing.
• Demonstrates a Formula API for model specification and provides model selection, cross-validation, and common workflows.
• Emphasizes best practices like adding a constant, checking assumptions, reporting effect sizes, and robust standard errors when appropriate.

When to use it

• Fitting regression models (OLS, WLS, GLS, quantile, mixed-effects)
• Generalized linear modeling with various distributions and links
• Analyzing discrete outcomes and count data
• Time series analysis and forecasting
• Running statistical tests and diagnostics, assessing model fit and assumptions
• Producing publication-ready summaries and tables

What it can touch

• Uses Python code paths shown in examples (import statsmodels.api as sm; sm.OLS; ARIMA; GLM; Logit; etc.)
• Demonstrates formula-based APIs via statsmodels.formula.api and model-specific calls (e.g., Logit, ARIMA, G LM).
• Executes diagnostic plots and checks (e.g., residual plots, ACF/PACF plots) as part of workflows.

Caveats

• License stated as Apache-2.0.
• Requires understanding of statistical modeling and proper interpretation of outputs (e.g., coefficients, odds ratios, rate ratios, confidence intervals).
• See specific model sections for limitations and recommended diagnostics; the skill emphasizes best practices rather than guaranteeing outcomes.