Comparing groups and modelling outcomes

This branch covers statistical comparisons, associations and fitted models, divided by the method or measure being interpreted and reported. Descriptive statistics, general uncertainty and hypothesis testing, study design and analysis planning, missing-data strategy, critical appraisal and presentation of numbers sit outside this branch.

• Comparing means and adjusting for covariates — independent and paired t tests, ANOVA, covariate adjustment, factorial and multivariate designs, and repeated measures
• Comparing categorical outcomes — contingency tables, paired and independent tests, and pooled or stratum-specific comparisons
• Rank-based comparisons — methods for independent, paired and repeated observations, their assumptions and the questions they test
• Correlation and agreement — coefficient choice, association reporting and agreement between measurement methods
• Risk and treatment effects — occurrence measures, risk ratios and differences, odds ratios, numbers needed to treat and attributable fractions
• Regression and model reporting — model families, dependent observations, predictor handling, covariate adjustment, parameter interpretation and reporting
• Effect modification and interaction — effect modification, moderators and mediators, statistical interaction and subgroup reporting
• Survival analysis — event and censoring definitions, Kaplan–Meier estimates, hazards, comparisons and time-to-event reporting
• Diagnostic and prediction performance — test accuracy, predictive values, likelihood ratios, discrimination, calibration and fairness across subgroups
• Resampling and Bayesian inference — Bayesian updating, Markov chain Monte Carlo, analysis reporting and bootstrapping
• How a machine learning model is built and where it fails — model types, supervised training, bias, leakage, overfitting, misleading accuracy and performance changes over time