Logistic Regression

Summary

Builds logistic regression as the model for binary outcomes after explaining why linear regression fails through predictions outside zero-to-one, non-Normal residuals, and non-constant variance. Develops the logit link, maximum likelihood estimation, the S-shaped probability curve, and the practical translation of exponentiated coefficients into odds ratios with asymmetric confidence intervals. Extends to multiple logistic regression for confounder adjustment, categorical predictors, and interactions, then covers model fit through the Hosmer-Lemeshow test, discrimination via ROC curves and the C-statistic, calibration plots, separation problems, and influential observations.