Randomised Experiment
Random assignment of , which makes independent of the error so that differences in outcomes can be read as causal.
When to use
Lecture 1 uses it as the way to estimate causality, as in the health and drug example ; without randomisation, confounders can bias the estimate.
See Lec 1.