Apology and repair

Why should an apology change anything when the harm it addresses has already happened?

The glass is broken, the words were said, the appointment was missed. Nothing an apology contains reaches back and alters any of that. So a sceptic has a fair question: is an apology anything more than a noise that makes both parties feel better about an unchanged situation?

The usual defence — that it shows remorse, and remorse is nice to know about — concedes too much, because it makes the apology a mere report. If that were all it did, a sincere private feeling would serve equally well, and a third party could deliver the news. Neither substitution works.

Begin by asking what the harm actually consisted of. Break someone's window by accident and you owe them a window. Break it in anger and you owe them a window and something else. The difference is not in the glass.

What the second act added was a claim — not stated aloud, but expressed by the doing. Treating someone as an obstacle asserts something about their standing relative to you: that your convenience, or your temper, outweighed their interest. That is why a wrong feels like a message and not just a loss. Paying for the window while insisting you would do it again settles the debt and leaves the message standing.

Now notice the key property of that claim: it does not expire with the act. It stays in force. Until something withdraws it, the relationship continues on terms in which one party has been placed below the other, and both know it. That is the harm still doing work in the present — and it is a live thing, which is why something said afterwards can touch it.

So what would withdraw it? Not a report of feelings, because the claim was made by the wrongdoer and only the wrongdoer can retract it. Not a third party's account, for the same reason. It has to be an act by the same agent that repudiates the earlier act — and that is exactly the shape of an apology. It is not a description of remorse but a performance: saying the words is the withdrawal, in the way that saying "I promise" is the undertaking rather than a report of one.

That immediately explains a familiar phenomenon. Why does "I'm sorry if anyone was offended" enrage rather than soothe? Because it withdraws nothing. It concedes no act, accepts no responsibility, and relocates the problem into the listener's sensitivity — a restatement of the original claim about standing, dressed as its withdrawal. The ingredients of an apology that works are the ones a non-apology omits: naming what was done, owning it, declining the excuse, indicating what will be different.

A second thing an apology does is why the words are not cheap. Admitting fault surrenders your deniability, hands the other person something they can repeat, and commits you publicly to a standard your future conduct can be measured against. That cost is what gives the words evidential weight.

And this is why an apology cannot be complete on its own. A withdrawal has to be received: the wronged party can accept it, refuse it, or leave it hanging, and until they respond the apology is an offer rather than a repair. Two people are required, which is what the sceptic's picture of a self-soothing noise leaves out.

Three refinements.

First, the two functions can come apart, and this is where most disagreement sits. On one reading the work is constitutive: the apology changes the moral relationship directly by retracting the claim, and would do so even if nothing about the future were in question. On another it is evidential: what matters is that apologising is costly and therefore signals a person less likely to reoffend. The readings usually recommend the same behaviour, but they disagree sharply about deathbed apologies, apologies between strangers who will never meet again, and official apologies for historical wrongs — cases with no future relationship for evidence to be about, where many people still think something real is accomplished.

Second, apology is not forgiveness and does not compel it. The offender withdraws the claim; whether the wronged party releases their resentment is a separate act, theirs to perform or withhold. Treating an apology as a transaction that purchases forgiveness — "I said sorry, what more do you want" — reasserts exactly the priority the apology was supposed to withdraw.

Third, a caution about evidence. It is often claimed that apologising reduces lawsuits, and there is real research on medical apology laws in the United States, but the findings are mixed rather than settled — some report fewer or cheaper claims, others no effect or effects in the opposite direction for some groups of doctors. The argument above does not depend on that literature and should not lean on it.

A boundary worth marking. How a few words can create an obligation that did not exist a moment earlier is a neighbouring question about promising, and its answer runs forward: a promise builds a new commitment out of nothing. This question runs backward, to a commitment already broken, and what changes is not an obligation but the terms on which two parties stand. Likewise, why one broken promise costs more than many kept ones earn is about how evidence accumulates asymmetrically; here an apology is not primarily a piece of evidence at all.

• What an institution or a state is doing when it apologises for something none of its present members did.
• Whether forgiveness given without any apology is a gift, a mistake, or something the wronged party owes themselves.

Every term the collection defines is gathered in the glossary.