All Chapters

10 chapters · read in any order, though 1 → N works best.

  1. 1

    Why Experiment?

    Most things that look like cause and effect aren't. This chapter explains why — and why randomized experiments are so powerful.

    foundationscausalityconfounding
    18 min
  2. 2

    Designing Experiments

    How to design an experiment that actually answers your question — power analysis, block designs, compliance, and pre-registration.

    experimental designpowerrandomization
    20 min
  3. 3

    Causal Graphs

    DAGs (directed acyclic graphs) let you reason visually about causation. Learn the backdoor criterion, what to control for, and the collider trap.

    DAGsconfoundinggraphical models
    22 min
  4. 4

    Counterfactuals

    The potential outcomes framework, average treatment effects, heterogeneous treatment effects, and how machine learning estimates them.

    counterfactualspotential outcomesheterogeneous effects
    20 min
  5. 5

    Bayesian Thinking

    Bayes' theorem, priors and posteriors, the Beta-Binomial model, credible intervals, and why Bayesian methods are natural for experiments.

    Bayesianprobabilityposteriorscredible intervals
    22 min
  6. 6

    Multi-Armed Bandits

    Learn how multi-armed bandits adapt in real time — Thompson sampling, UCB, and why they beat fixed A/B tests for online decisions.

    banditsThompson samplingexplorationonline learning
    25 min
  7. 7

    Reinforcement Learning

    MDPs, Q-learning, policy gradient methods, and the connection between RL and causal inference for decision-making.

    reinforcement learningMDPQ-learningdecision-making
    28 min
  8. 8

    Effect Estimation

    OLS, propensity scores, AIPW doubly-robust estimation, and how machine learning improves causal estimation.

    estimationregressionpropensity scoresAIPWdoubly-robust
    24 min
  9. 9

    When Experiments Fail

    How to stress-test causal claims: E-values for unmeasured confounding, Rosenbaum bounds, placebo tests, and observational methods.

    sensitivityE-valuesunmeasured confoundingobservational
    20 min
  10. 10

    Causal AI

    Why ML alone isn't enough for decisions. Causal discovery, invariant risk minimization, counterfactual prediction, and the future of causal AI.

    causal AImachine learningcausal discoverycounterfactual ML
    26 min