Randomization Techniques

Chapter 6 — Randomization Techniques Randomization is the engine of causal inference, but “how to randomize” is less obvious than it sounds. This chapter surveys the landscape of classical assignment mechanisms—Bernoulli trials, completely randomized experiments, randomized block (stratified) experiments, rerandomization, and matched-pairs designs—explaining the three conditions each must satisfy: nonzero probability, individualism, and unconfoundedness. It also covers design-conscious inference, ensuring that statistical tests properly account for the randomization scheme used.

1. Classical randomized assignment mechanisms satisfy three necessary conditions: nonzero probability, individualism, and unconfoundedness.
2. Five popular randomization approaches that satisfy these conditions are Bernoulli trials, completely randomized experiments, randomized block (stratified) experiments, rerandomization, and paired randomized experiments.
3. Conducting statistical inference accounting for the assignment mechanism (design-conscious inference) is crucial to ensure that statistical tests have appropriate error rates.