Module 4 — What MCMC Actually Does

Contents

Module 4 — What MCMC Actually Does#

Coming soon. This module will demystify Markov Chain Monte Carlo sampling using animated trace plots.

Preview#

TEXAS uses Stan’s No-U-Turn Sampler (NUTS) to draw samples from the posterior. Instead of computing the posterior analytically (impossible for complex models), MCMC explores the posterior by taking many small steps, accumulating a cloud of samples that represent the distribution.

Key ideas this module will cover:

  • Why we sample instead of solve

  • Reading a trace plot: what good mixing looks like vs. pathological chains

  • R-hat: a simple convergence diagnostic

  • Effective sample size (ESS): why 4,000 draws ≠ 4,000 independent estimates

  • Divergences: what they signal and why TEXAS reports them

Contents