---
tag: 734Q6Q51X7
kind: concept
name: Randomization
msc: [62K99]
sources:
  - url: https://en.wikipedia.org/wiki/Randomization
---

## Summary

Randomization is the assignment of treatments to experimental units by a chance mechanism with known probabilities.
It balances, on average, all unit characteristics between the treatment groups, including those that nobody measured.
So differences in outcome can be attributed to the treatments.
It also provides the probability model behind randomization tests and, in large samples, justifies the usual tests of [[analysis of variance|M9AQ5HKGP4]].
It is one of the three principles of [[design of experiments|2TE42ZWJ3E]] and the basis of randomized controlled trials and of [[causal inference|V3QJFN5800]].

## Explanations

### L1 human

If a doctor chooses which patients get a new drug, she may give it to those who look strongest, and the drug will seem better than it is.
Tossing a coin for each patient removes that choice.
Then the two groups differ only by chance, in age, health and everything else.
With four patients and two places in each group, there are $6$ ways to split them, and the randomization picks one of them, each with probability $1/6$.