Randomization
Also known as: Random assignment
Letting chance decide who gets the treatment, so groups start out comparable.
What it means
Randomization is the allocation of units to experimental conditions by a chance mechanism, the defining feature that separates a true experiment from an observational study. By making assignment independent of every characteristic of the units — measured or not — it ensures the groups are, in expectation, balanced on all confounders, so any post-treatment difference can be attributed to the treatment. Randomization also provides the formal basis for inference, because the distribution of outcomes under the null can be derived from the assignment process itself. It does not guarantee balance in any single small sample, which is why blocking, stratification, and adequate sample sizes still matter.
Examples
Flipping a coin to decide which patients get the new drug guards against doctors steering healthier patients toward it.
An app assigns users to the new checkout by a random hash, so keen early adopters cannot pile into the new version and make it look better than it is.
A jobs programme with too few places allocates them by lottery; the applicants who miss out become a control group matched on motivation, not just on paperwork.
First described in Ronald Fisher (1925, 1935).