Randomization balances known and unknown confounders
The flow diagram shows the machinery of an RCT: eligible participants are enrolled, randomized (usually by computer or sealed envelopes to prevent bias), assigned to treatment or control, and followed to an outcome. Randomization ensures that baseline characteristics are balanced between arms on average, including factors not yet measured or even discovered.
A non-random assignment (for instance, sicker patients getting treated, healthier ones getting control) introduces bias that no statistical adjustment can fully remove. Randomization breaks that causal arrow.
Analysis must match the randomization
All randomized participants should be analyzed in the group they were assigned to (intention-to-treat), even if they did not receive the assigned treatment. Analyzing only those who completed treatment breaks the balance created by randomization. The flow diagram's endpoint is typically a parallel analysis comparing outcome frequency between the two arms.