Logistic Regression Online: Model Binary Outcomes
What is Logistic Regression?
Logistic regression models a binary outcome such as pass/fail, default or respond/no-respond by linking a linear combination of predictors to a 0-1 probability through the logit function: ln(p/(1−p)) = b0 + b1X1 + ... + bkXk. The outcome follows a binomial distribution, and predictors need not be normally distributed. It is one of the most widely used models for classification and risk factor analysis, and its coefficients are easier to explain to business users than linear regression.
When to Use It
Use it to predict pass/fail outcomes, estimate default or churn risk, screen potential causes of defects, and study risk factors in medical or quality data. It is appropriate whenever the outcome is binary and you need both prediction and interpretable risk quantification.
How to Use It (Step by Step)
Prepare a data table with one binary outcome column (0/1) and one or more predictors, which can be continuous or categorical with automatic dummy coding. Select the variables and the tool fits the model, outputting coefficients, odds ratios, Wald p-values, goodness of fit, AUC and a confusion matrix, and can predict probabilities for new points. With few events, reduce the number of predictors first to avoid overfitting.
Key Formulas / Example
The odds ratio for predictor j is OR = exp(bj): each unit increase multiplies the odds of the event by OR. For example, b = 0.693 gives OR = 2.0, so each unit doubles the odds. Model fit is checked with the Hosmer-Lemeshow test, where a large p-value indicates good fit, and discrimination with AUC, where values closer to 1 mean better separation.