What problem does it solve? Choosing the right regression model, validating its assumptions, and interpreting coefficients correctly is error-prone — analysts routinely fit OLS to binary outcomes, forget intercepts, ignore overdispersion, or misread log-scale coefficients. This Skill provides a rigorous statsmodels workflow that matches models to outcome types and validates assumptions before trusting any p-value. ## Core Features & Use Cases - Model Selection by Outcome Type: Decision tables map continuous, binary, ordinal, count, zero-inflated, and time-ordered outcomes to the correct statsmodels class (OLS, Logit, MNLogit, OrderedModel, Poisson, NegativeBinomial, GLM, ARIMA/SARIMAX, VAR). - Diagnostics and Inference: Residual assumption tests (Breusch-Pagan, Ljung-Box, Jarque-Bera), influence detection (Cook's D, leverage, DFFITS), multicollinearity checks (VIF), robust/HAC/cluster standard errors, and power analysis. - Correct Interpretation: Guidance for reading coefficients on the right scale — odds ratios for logit, rate ratios for Poisson, marginal effects for nonlinear models — plus AIC/BIC and likelihood-ratio model comparison. - Use Case: A data analyst needs to know whether a treatment significantly affects a binary outcome. The Skill guides them to fit a Logit model, check convergence, report odds ratios with confidence intervals, and compute marginal effects rather than misreading raw coefficients. ## Quick Start Fit a logistic regression of churn on tenure and monthly charges with statsmodels, check the residual diagnostics, and report the odds ratios with confidence intervals.
