What Counts as Coercion
The credible threat of harm to gain compliance
A mugger holds a loaded gun and says, “Give me your wallet or I’ll shoot.” Everyone agrees this is coercion. But the word does not stay put. Taxes are called coercion, and so are wage offers, advertising campaigns, social pressure, a mother’s guilt trip, an employer’s dress code. People use it to describe everything from violent threats to emotional manipulation, and the stretching is not innocent: coercion is a central concept in ethics, politics, and law, and a great deal of moral machinery is bolted to it. Rights, I argue elsewhere, are preferences we are willing to enforce through coercion and consider ethical to enforce that way — so a mushy definition of coercion makes mush of rights, and of everything downstream of rights. The previous chapter ended with an instruction: call coercion coercion. That instruction is empty until we know exactly what coercion is.
Here is the definition:
Coercion is the credible threat of actual harm to gain compliance.
The formulation owes a debt to Robert Nozick, who offered something similar in Anarchy, State, and Utopia; this version is refined for clarity and operational use. It has four elements — credible, threat, actual harm, to gain compliance — and every one of them is load-bearing. Remove any single element and coercion vanishes, leaving behind something else that deserves a different name. The way to see this is the way to test any definition: pair each element with a case that has it and a near-neighbor that lacks it, and watch the verdict flip.
The Four Elements
Credible. The threat must be believable to the person being coerced. The mugger with the loaded gun clears this bar. A child who says “If you don’t give me candy, I’ll turn you into a frog” does not: no rational adult assigns the transformation any chance of happening. Notice that the frog threat has everything else — it is conditional, it names a consequence, it aims at extracting candy. What it lacks is believability, and without believability the threat exerts no pull on anyone’s decision. There is nothing to comply about.
Threat. The harm must be conditional: if you do not comply, harm will follow; if you do, it won’t. “If you don’t sign this contract, we’ll destroy your reputation” is a threat — the harm is avoidable through compliance, which is precisely what gives it leverage. “I’m going to fire you” is not a threat in this sense; it is a statement of fact, a prediction of unconditional harm. It becomes coercion the moment an escape hatch is attached: “…unless you work this weekend.” The conditionality is the mechanism. A harm you cannot avoid by complying cannot be bargaining with your agency.
Actual harm. The threatened consequence must be something the coerced agent has genuine reason to avoid. “Pay your taxes or you’ll be imprisoned” qualifies — imprisonment is harm by any account. “If you don’t agree, I’ll be disappointed” does not, unless the disappointment is backed by material consequence. Emotional discomfort, withheld approval, the prospect of someone thinking less of you: these can influence behavior, but they are not actual harm, and calling them coercion inflates the concept until it covers all social life. Exactly what harm is gets its own definition in What Counts as Harm; for now the working test is that the consequence must be a real setback, not a mere displeasure.
To gain compliance. The purpose of the threat must be to induce specific behavior. “Hand over the documents or I’ll leak your private messages” is aimed at altering what you do next. “You’re going to jail because you broke the law” is not — it is punishment, delivered after the fact, with no behavior left to alter. Punishment can be unjust, disproportionate, or cruel, but it is a different act with a different analysis. (The advance threat of punishment — pay your taxes or else — is exactly how law coerces; the retrospective sentence is not.)
Put the elements back together and run a composite case: “If you don’t testify in court, we’ll expose your immigration status to the authorities.” Credible? Yes. Conditional threat? Yes. Actual harm? Yes. Aimed at compliance? Yes. Coercion — and notice that no one has been touched.
The four elements also draw the boundary against coercion’s neighbors, which is half their value. Persuasion changes your mind with reasons; there is no harm in the picture. Bribery changes your payoffs by adding to them; there is an offer but no threat. Force skips the conditional entirely; the harm is not threatened but already being inflicted. Coercion lives in the shadow of violence while preserving the illusion of choice. It is an attempt to shape agency through fear — and that formulation, “shape agency,” is doing precise work, as the last section will show.
Credibility Is Credence
The first element needs cashing out. “Believable to the person being coerced” — what exactly is believability?
The framework already owns the right instrument. Credence — the subjective probability an agent assigns to a proposition, developed in Measure and Credence — quantifies an agent’s degree of confidence that something is true or will happen. Credibility is just Credence viewed from the threat’s side of the table:
A threat is credible if and only if the targeted agent rationally assigns a high enough Credence to the realization of the threatened harm for it to influence their decision-making.
The consequence falls out immediately: credibility is subjective and agent-relative. It is not an absolute feature of the threat itself — not a fact about the gun, the muscles, or the leaked documents — but a fact about the epistemic state of the agent receiving it.
The test case: a mugger implies he has a gun concealed in his pocket. In reality it is a toy, incapable of harming anyone. But the victim, reading the context, the behavior, and the implication, rationally assigns a high Credence to a real firearm. The threat is credible, and the scenario is coercion — even though no genuine harm could ever have been delivered. Coercion does not require the capacity to harm. It requires only the rational expectation of harm, because it is the expectation, not the capacity, that reaches into the victim’s decision and bends it.
The relativity cuts both ways. A perfectly serious threat, backed by real capability, fails to coerce an agent who assigns it low Credence — the mugger with a real gun that his victim mistakes for a toy is attempting coercion and failing at it. And a seemingly minor threat can be profoundly coercive if it commands high Credence in its particular target. This is why coercion cannot be read off the threatener’s resources; you have to look at the victim’s rational epistemic state.
The word rationally is holding the other flank. Agent-relative is not agent-arbitrary: the Credence must be the one the agent’s evidence actually supports. Someone who hears a death threat in the weather report has not been coerced by the forecaster. Between the toy gun that rationally reads as real and the forecast that only delusion reads as a threat lies the entire domain of credibility, and Credence is the instrument that maps it.
Deleting Branches, Poisoning Branches
Coercion’s nearest neighbor deserves more than a boundary line, because the two are constantly blurred together — coercion treated as merely another form of violence, or both conflated with speech. They are distinct modes of agency violation, and they operate at different levels of the choice landscape.
Violence is the direct application of physical force that causes bodily harm, damage, or destruction: punching, stabbing, shooting, arson, smashing property. Its effect on agency is collapse. Violence removes or cripples the capacity to act in the most literal sense — by injuring, destroying, or eliminating the agent or the resources agency runs on.
Coercion is the conditional attack. “Hand over your wallet or I’ll stab you” leaves the victim’s options physically intact — comply, resist, escape are all still there. What the threat does is revalue them. The alternatives to compliance are poisoned with extreme negative expected consequences, so that only compliance appears rationally accessible. Nothing has been removed from the choice landscape; the landscape has been distorted until every path but one runs uphill into fear.
In the Quantum Branching Universe (QBU) — the Everettian picture of reality as branching timelines, whose formal construction belongs to another volume — the distinction becomes exact. Violence deletes branches: death, injury, and destroyed resources prune entire sets of possible futures permanently from the tree. Coercion revalues branches: none are removed, but most are loaded with such negative value that the agent is funneled toward one distorted path. Violence is brute force in action; coercion is violence-in-potential, harnessed strategically.
The distinction is normative as well as analytic. Defensive violence can be justified — self-defense is the countering of force with force. Coercion is harder to justify, because it does not merely counter force; it instrumentalizes another’s agency, conscripting the victim’s own rationality as the delivery mechanism of the attack. Where the boundaries of justified coercion actually lie — the pre-consented, the defensive, the compensatory — is the work of The Boundaries of Force.
Speech at the Boundary
One application matters enough to state here, briefly. Speech is not violence: words, however harsh or inflammatory, do not collapse anyone’s choice space, and violence requires physical force or its direct equivalent. But sometimes speech is coercion: when words constitute a credible threat of actual harm to gain compliance, they are not expression about a coercive act — they are the coercive act itself. “We should rise up against the government” is protected; “Do what I say or I’ll burn your house down” is not. Incitement is not coercion: calls to action are permissible, threats are not. That single line marks the precise boundary at which speech becomes an agency-violating act, and it lets you protect all speech short of coercion without pretending that threats are opinions or that opinions are violence. The full defense of that position — why speech is not violence, why incitement is not coercion, and the case for free speech itself — belongs to a later volume.
The Shape of the Threat
Step back and look at what the definition buys. With four elements in hand, you can distinguish legitimate defense from domination, voluntary agreement from coerced compliance, ethical boundaries from brute power — and you can do it case by case, element by element, instead of by rhetorical temperature. Every claim about rights inherits this precision, because rights are enforced preferences and enforcement is coercion all the way down.
And the branch framing leaves a promissory note. Violence is harm actualized: branches deleted. Coercion is harm suspended: branches revalued, the harm hanging over the tree rather than falling on it. But a harm hanging over your future with high Credence is not nothing — it is already reshaping what your future is worth, before anything happens and whether or not the threat is ever executed. Follow that thought to its end and the wall between threatened harm and actual harm starts to look thinner than this chapter has treated it. Tearing it down — showing that imposed risk is not the possibility of harm but a form of harm itself — is the radicalization that arrives in Risk Is Harm.