Presumption of innocence

Why does a legal system accept a rule it knows will let some guilty people go free?

A criminal trial is a device for deciding a factual question with incomplete information. It will therefore be wrong sometimes. That much is unavoidable, and no rule of procedure can repeal it.

What is strange is the shape of the rule we chose. The presumption of innocence does not aim at being wrong as rarely as possible. It deliberately tilts the machine so that it errs in one direction more often, in order to err in the other direction less. Why would anyone design a fact-finder that is knowingly biased? Only, presumably, if the two ways of being wrong are not equally bad. So: are they?

Set out the four outcomes. A guilty person convicted and an innocent person acquitted are both correct. The two errors are a false conviction — an innocent person punished — and a false acquittal, a guilty person released.

Ask what each costs, and to whom. A false acquittal leaves a wrong unaddressed and a possibly dangerous person at liberty; the harm is real, diffuse, and often reversible in the sense that the person may be tried again on new evidence in many systems, or caught for a later offence. A false conviction is the state, deliberately and with its full authority, inflicting on a citizen exactly the harm it exists to prevent — imprisonment, ruin, sometimes death — while the actual offender remains free, so the second error is contained inside the first. It is also the error the person harmed can do least about, and the one that, once discovered, damages the standing of every other verdict the system issues.

If you accept that asymmetry, the design follows almost mechanically. Any fact-finder has a threshold: how convinced must it be before it condemns? Raise the threshold and false convictions fall while false acquittals rise; lower it and the reverse. There is no setting that eliminates both. So the only real question is where to put the dial — and the answer encodes a ratio of how much worse one error is than the other.

Blackstone gave the ratio its most quoted form in the 1760s: better that ten guilty persons escape than that one innocent suffer. The particular number is rhetorical, and the history of the phrase contains many others — Maimonides argued for a far larger figure, other jurists for smaller. What matters is not the arithmetic but the admission buried in it: we are choosing to be wrong more often overall, on purpose.

Now, the part most often blurred. Three distinct devices carry that choice, and they are routinely spoken of as one thing.

The presumption is the starting position. Before any evidence, the accused is treated as innocent — which is not a prediction about the facts but an instruction about where the trial begins. It follows that silence proves nothing and that the accused need not have an account of anything.

The burden of proof is the allocation: whose job is it to move the fact-finder off that starting position? The prosecution's. Woolmington v DPP in 1935 called this the golden thread of English criminal law. The accused is not required to establish innocence, because innocence was never in question — it was assumed.

The standard of proof is the height of the bar: how far must the prosecution move it? Beyond reasonable doubt, in criminal matters — as against the balance of probabilities in a civil suit, which is why the same conduct can lose a civil case and win a criminal one on the same evidence.

Three different things. The presumption says where we start, the burden says who must push, the standard says how far.

Two things this is not.

It is not a claim that the accused is probably innocent. The presumption is a rule about procedure, not a statistical estimate, and it applies with full force to someone caught in the act. Nor is an acquittal a finding of innocence: it records that the case was not proved to the required standard, which is why Scots law long kept a separate "not proven" verdict, and why an acquitted defendant can still lose a civil claim arising from the same events.

Nor is the tilt absolute in practice. Real systems carry reverse-onus rules — the insanity defence, some regulatory and possession offences — where the accused must establish something, and courts spend a good deal of effort deciding which of these are compatible with the presumption. The principle is close to universal; its exceptions are contested case by case.

One refinement worth holding onto: the asymmetry is a choice about error costs, so it moves when those costs move. The reason a civil dispute over money uses the balance of probabilities is not that truth matters less there, but that the two errors — wrongly ordering payment, wrongly refusing it — are roughly symmetrical between two private parties. It is the state's power to punish that unbalances the criminal ledger, and that is what the higher standard responds to.

• Why appeals are structured asymmetrically, and what that says about which error the system fears.
• How the same error-cost logic sets thresholds outside courts — in screening, safety engineering, and content moderation.

Every term the collection defines is gathered in the glossary.