Marginal deterrence is the principle that the law should punish worse offenses more severely than lesser ones by a wide enough margin that an offender already committing a crime still has a reason not to escalate. The idea goes back to Jeremy Bentham in 1789 and was formalized in economic terms by George Stigler in 1970. It explains why stealing $100,000 should not carry the same penalty as stealing $1,000, why armed robbery should be punished more harshly than unarmed robbery, and why a tax fraud penalty should sit well above a penalty for simple negligence. When the gaps between punishments collapse, the law loses its grip on the offender’s next decision.
The Bentham Rule
Bentham stated the principle as a rule for legislators: where two offenses compete for an offender’s attention, the punishment for the greater one must be high enough to steer the person toward the lesser. He put the point in arithmetic. If someone who steals ten shillings is punished the same as someone who steals five, then taking the extra five shillings is effectively free. “As often as a man gives you five blows,” Bentham wrote, “he will be sure to give you five more, since he may have the pleasure of giving you these five for nothing.”
Nearly two centuries later, Stigler made the argument in the language of economics. Uniform penalties, he warned, destroy marginal deterrence entirely: “If the offender will be executed for a minor assault and for a murder, there is no marginal deterrence to murder. If the thief has his hand cut off for taking five dollars, he had just as well take $5,000.”1National Bureau of Economic Research. The Optimum Enforcement of Laws The lesson for any legal system is that penalties need gaps between them, and those gaps need to be large enough that an upgrade in misconduct brings a real jump in consequences.
How the Escalation Moment Works
The practical force of marginal deterrence shows up at the point where an offender has already committed one crime and is deciding whether to commit a worse one. A burglar choosing whether to arm himself. A drug dealer choosing whether to silence a witness. A robber confronted with resistance. At that moment, the difference in penalties between the crime already underway and the worse crime under consideration is the only thing the law can offer as a reason to stop.
When the difference is real, it acts as a brake. A five-year sentence for robbery against a twenty-five-year sentence for armed robbery gives a robber a concrete reason to leave the gun at home. But when the gap shrinks or disappears, the brake fails. If armed robbery and murder both carry life imprisonment, the armed robber who encounters resistance has no penalty-based reason to avoid killing. The marginal cost of the worse act drops to zero.
The same logic runs up and down the whole criminal code. Minor property crimes carry fines. Serious property crimes carry short jail terms. Violent crimes carry prison time that scales with the harm inflicted. Each step up the ladder costs more, and the cost differential is what keeps people from climbing higher.
The Expected Cost of a Crime
Gary Becker’s 1968 model defines the expected cost of an offense as a function of two variables: the probability of being caught and punished, and the severity of the punishment itself. Raising either one raises the price of the crime.2National Bureau of Economic Research. Crime and Punishment: An Economic Approach
That framework explains why fines for certain white-collar offenses look enormous on their face. If the chance of being caught for insider trading or environmental dumping is low, the penalty has to be proportionally higher to keep the expected cost meaningful. A 10 percent detection rate paired with a fine equal to the illicit gain means the expected cost is only a tenth of the gain. The fine must be set at a multiple of the gain to compensate.
Marginal deterrence adds a second requirement. It is not enough for each crime to carry a high expected cost in isolation. The expected cost has to rise as the crime gets worse. If a jurisdiction hikes the penalty for a minor offense until it matches the penalty for a major one, it may deter the minor crime while destroying the incentive to avoid the major one. The whole schedule has to be internally consistent, each rung priced above the one below after adjusting for detection probability.
When Marginal Deterrence Collapses
Three-Strikes Laws
The clearest real-world failure comes from three-strikes sentencing. These statutes impose lengthy mandatory sentences on repeat offenders, sometimes life imprisonment on a third felony conviction. The structural problem is that once an offender is already facing a potential life sentence, the gap between the current crime and a much worse one effectively vanishes. A two-strike offender caught committing a third felony has a rational incentive to eliminate witnesses, because the punishment for murder is no worse than what already awaits.
Thomas Marvell and Carlisle Moody tested this prediction across the 24 states that had enacted three-strikes laws. They reported a 10 to 12 percent increase in homicides in the short run and a 23 to 29 percent increase in the long run, with “little evidence that the laws have any compensating crime reduction impact through deterrence or incapacitation.”3The Journal of Legal Studies. The Lethal Effects of Three-Strikes Laws Laws intended to suppress repeat offending appear to have encouraged the worst possible escalation.
Mandatory Minimums
Mandatory minimums produce a related breakdown. A floor on punishment set without regard to circumstances means trivial differences in conduct can trigger massive differences in consequences, while large differences in culpability get flattened under the same minimum. A low-level courier and the head of a distribution network can face the same number. The research on whether mandatory minimums actually reduce crime is not encouraging; studies have largely failed to show a deterrent effect, and some suggest these sentencing floors generate more serious crime by eliminating the penalty gaps that would otherwise discourage escalation.
Why Rational-Actor Assumptions Only Go So Far
Marginal deterrence assumes people run a cost-benefit calculation before acting. The theory works best when the actor is informed, rational, and forward-looking. Many people committing crimes are none of those things at the critical moment.
Behavioral economics has identified several patterns that undercut the model. Offenders tend to be overoptimistic about avoiding detection. They weigh immediate rewards far more heavily than future punishment, a pattern called hyperbolic discounting. Even someone who knows a crime is a bad long-term bet may act on impulse when the short-term payoff is right in front of them.
The discounting problem is especially damaging to marginal deterrence at the high end of the scale. Research on how offenders value future punishment suggests the disutility of each additional year of imprisonment drops sharply. One study estimated that, relative to the first year inside, the marginal disutility of an added year falls to about 30 percent after five years, 7 percent after ten, and just 1 percent after fifteen.4Institute of Labor Economics. Criminal Discount Factors and Deterrence Raising a sentence from 20 to 25 years barely registers. Marginal deterrence does its best work at the lower end of the penalty scale, where the jump from a fine to a jail term, or from one year to five, represents a qualitative change in the offender’s life.
That does not make graduated sanctions pointless. It means the theory’s strongest predictions hold for calculated, economically motivated conduct and weaken for impulsive crime. A corporate executive weighing environmental noncompliance sits much closer to Becker’s rational actor than a person committing assault in a moment of rage.
Graduated Penalties in Federal Sentencing
The federal sentencing guidelines are one of the most detailed working implementations of the idea. Each offense has a base level, and specific aggravating facts add defined numbers of levels. Higher levels map to longer recommended prison terms through the sentencing table.
Robbery illustrates the design. The base offense level is 20. If the offender brandished or possessed a firearm, five levels are added. If a firearm was used more actively, six levels are added. If a firearm was discharged, the enhancement is seven levels.5United States Sentencing Commission. Primer on Robbery Offenses A separate enhancement of up to five levels applies for bodily injury, though the combined increase from weapons and injury cannot exceed eleven levels. A robber using intimidation alone faces a meaningfully lower offense level than one who brings a weapon, and one who brings a weapon faces less time than one who fires it. Each step in violence carries a specific, predictable cost.
Where It Works Best: Regulatory Ladders
Marginal deterrence may work better in regulatory contexts than in criminal law, because regulated entities are closer to the informed, calculating actors the theory assumes. Businesses track compliance costs, weigh penalties against profits, and respond to incentive structures in reasonably predictable ways.
Workplace Safety
OSHA’s penalty structure is a textbook ladder. A serious workplace safety violation currently carries a maximum penalty of $16,550 per violation. A willful or repeated violation carries a maximum of $165,514 per violation, roughly ten times more.6Occupational Safety and Health Administration. OSHA Penalties An employer who knows about a hazard and ignores it faces consequences an order of magnitude worse than one whose violation was inadvertent.
Tax Penalties
The Internal Revenue Code builds a similar ladder. A taxpayer who underpays due to negligence or a substantial understatement faces a 20 percent accuracy-related penalty on the underpaid amount.7Office of the Law Revision Counsel. 26 USC 6662 – Imposition of Accuracy-Related Penalty on Underpayments A taxpayer who underpays due to fraud faces a 75 percent penalty.8Office of the Law Revision Counsel. 26 USC 6663 – Imposition of Fraud Penalty A taxpayer who has made an honest mistake has little reason to compound it by fabricating records, because the penalty nearly quadruples once intent enters the picture.
Environmental Enforcement
The EPA follows the same pattern. The agency uses penalty matrices that scale fines with the duration and severity of noncompliance, so a company that quickly corrects a problem faces a substantially lower hit than one that lets the violation run.9Environmental Protection Agency. Guidance: Penalty Matrix for RCRA Section 7003 Civil Penalty Policy Each additional day of noncompliance adds cost, applying the same logic to the time dimension.
What Makes the System Work
Graduated sanctions depend on three conditions. First, the offender has to know the penalty gap exists. Someone who does not know that armed robbery carries a longer sentence than unarmed robbery cannot be deterred by the difference. This is where regulated businesses have an advantage over street offenders; compliance officers study penalty schedules, while most violent offenders do not.
Second, the spacing between tiers has to be meaningful. If the gap is too small, it will not register in a person’s decision-making. Stigler’s point was that marginal costs need to be right-sized for each tier, not merely present.1National Bureau of Economic Research. The Optimum Enforcement of Laws
Third, enforcement has to be credible. A steep penalty that is almost never imposed has little deterrent effect, because expected cost is penalty multiplied by probability. A $165,000 OSHA fine means little if inspections are rare enough that employers treat them as hypothetical.
Marginal deterrence tells you what a sensible penalty schedule should look like. It does not guarantee that anyone is paying attention to it. The strongest case for graduated sanctions sits in tax compliance, securities regulation, and workplace safety, where actors are informed and enforcement is consistent. The weakest case is impulsive street crime, where offenders are poorly informed, heavily discounting the future, and often acting under the influence of substances or emotion.