Judicial recusal

Why must a judge withdraw from a case when nobody has shown that they are actually biased?

A judge holds shares in a company appearing before her. She insists, and may sincerely believe, that the holding will not move her by a hair. No party offers evidence that it has. She is disqualified anyway.

That looks like an accusation made without proof, and courts are elsewhere fastidious about not doing that. So either the rule is a superstition, or it is not really making a claim about her mind at all. Which is it?

Start with what a rule against biased judging would need in order to work. It would need to detect bias. Ask how, in a single case. You cannot open the judge's head. You could look at the outcome — but a judge who ruled for her shareholding is exactly what an honest application of the law might also produce, and one who ruled against it may have overcorrected. A single decision is consistent with both stories, so it carries almost no information about the cause.

You could ask her. But the influence of an interest on judgement is not usually experienced as pressure; it is experienced as finding one side's argument more persuasive. So self-report is unreliable in a direction that always favours staying on the case.

Notice what has happened. Actual partiality is unobservable in the single instance by any party, including its owner, so any rule keyed to it would be unenforceable. The law does what it always does when a target is unobservable: it keys the rule to something visible that correlates with it. Financial interest. Kinship to a party. Prior involvement as counsel. Facts anyone can check from outside, and the standing conditions under which partiality is more likely.

That makes recusal a proxy rule. But it does not explain why the proxy is stated in the language of appearance — the English test asks what a fair-minded and informed observer would think, and the American federal statute speaks of proceedings in which a judge's impartiality might reasonably be questioned. Why frame it around an observer rather than around risk?

Because a court produces something besides correct rulings. It produces rulings the loser will comply with uncompelled. That second output depends on what the loser can tell about the process, and the loser can no more read the judge's mind than we can. What they can see is the connection. A litigant who loses to a judge who owned the winner's shares has a complete, self-consistent explanation of their loss that does not involve the merits — and has it whether or not the shares mattered.

So the rule does two jobs with one instrument: reducing the real incidence of skewed decisions, and denying losing parties an unanswerable story about why they lost. The second is why the standard is objective and sincerity no defence.

What would falsify this? If the mechanism is proxy-plus-legitimacy, disqualification should attach to defined external relationships regardless of demonstrated effect, and should be unavailable on a showing that the judge merely holds strong views about the law at issue — since views are the product of judging, not a hidden channel into it. That is close to what we find: prior judicial rulings, even emphatic ones, are generally held not to be grounds. The refuting observation would be a system that required proof the judge's connection actually changed the outcome before disqualifying — or one that disqualified freely for known legal opinions. Either would mean I have the mechanism backwards.

Two refinements, and then the price.

First, "appearance" is not a licence for suspicion. The observer in the test is stipulated to be informed — knowing how courts work, what judges routinely do — exactly so the standard does not collapse into whatever an aggrieved litigant chooses to feel. The rule is objective in both directions: it ignores the judge's assurance, and it ignores the losing party's hunch.

Second, the rule is not uniform. Some systems treat a direct pecuniary or equivalent interest as automatic disqualification with no weighing at all; others route everything through the reasonable-observer test. The split is a real design choice: automatic rules are cheap to apply and impossible to fudge; weighing rules handle odd cases better and invite argument.

Now the costs, which are substantial. Recusal is a weapon: a party preferring a different judge can press a thin motion and gain delay even when it fails. Systems answer with timing rules and sanctions, but the exposure is structural: a rule that fires on appearance can be fed appearances.

Worse is the arithmetic at the top. On a trial bench, a recusing judge is replaced and nothing is lost. On a final appellate court of fixed membership there is no substitute, so recusal does not neutralise the seat — it removes it. In courts where an evenly divided vote leaves the decision below standing, disqualification hands a systematic advantage to whichever side won earlier, and it does so silently. The remedy for one distortion introduces another.

And there is the part usually left unsaid: at many apex courts the motion to disqualify a judge is decided by that judge alone, with no review and often no reasons. That is the one place in the system where the argument for external, checkable standards is strongest, and the one place it is least applied. The arrangement persists partly because changing it would require the agreement of the very people whose discretion it protects — a general feature of self-regulating professions rather than a peculiarity of courts. Defenders answer that any external disqualifying power could be used by a government to strip a court of inconvenient members, which is a serious objection and not obviously wrong. The dispute is live; I do not think the evidence settles it.

• Why appellate courts split evenly rather than defaulting to reversal, and what that does to recusal's arithmetic.
• How disclosure regimes — which target the hidden half — succeed and fail differently from disqualification rules.

Every term the collection defines is gathered in the glossary.