How Incentives Drive Behavior — and Why They Sometimes Produce the Opposite

In April 1902, French colonial administrators in Hanoi had a rat problem of their own making. The construction of a modern sewer system, a point of pride for Governor-General Paul Doumer, had inadvertently created one of the largest rat habitats in urban Asia. The sewers filled rapidly with rodents that carried bubonic plague. The solution, announced that spring, seemed straightforward: pay a one-cent bounty for every dead rat, with proof required in the form of a severed tail.

For a few weeks, the program worked. Rat catchers and city workers brought in thousands of tails. Then the numbers kept climbing, far past what any reasonable estimate of the wild rat population would predict. Colonial inspectors began noticing rats in the streets missing their tails. Tailless, and alive. Citizens had discovered that they could sever a rat's tail to collect the bounty, then release the rat to breed more future bounties. A smuggling network developed to import rats from surrounding provinces into the city.

Historian Michael Vann, in his archival research on the episode published in French Colonial History (2003)¹, documents the paper trail from the colonial administration's own records. The program was canceled in September 1902. The rat population, swollen by months of inadvertent farming, was larger than when the bounty had begun.

This is not a legend or an economic parable. It is one of the most precisely documented examples in the historical record of what economists call a perverse incentive: a reward or penalty that produces the exact opposite of its intended effect.

How incentive designers think (and where they go wrong)

Standard incentive design rests on a clean premise: if you want more of something, pay for it; if you want less, impose a cost. The logic flows from basic price theory. Raise the cost of speeding and fewer people speed. Pay teachers for student test scores and teachers teach to the test, or, in more extreme cases, change the scores. Pay bounties for dead rats and people kill more rats.

The Hanoi case reveals the gap between the intended behavior (rat killing) and the behavior that was actually incentivized (tail production). These happened to be correlated until they weren't, until the incentive was large enough relative to the cost of breeding rats that the optimal response diverged sharply from the one the administrators imagined.

Economists recognize two channels through which incentives work, only one of which is usually modeled.

The price channel is the intended channel: a reward raises the return to the desired behavior; a penalty raises the cost of the undesired one. This is deterrence logic. It works reliably in simple, well-monitored situations where the scope for gaming is limited.

The information channel is subtler and more treacherous: a reward or penalty communicates something about what kind of relationship is in play, what behavior is expected, and what it means to comply or not. When the information channel sends an unintended signal, the behavioral response can be the opposite of what was designed.

The Haifa daycare study, conducted by economists Uri Gneezy and Aldo Rustichini and published in the Journal of Legal Studies in 2000², is the canonical modern demonstration of this dynamic.

The daycare fine: a clean natural experiment

In January 1998, administrators at ten Israeli daycare centers faced a common frustration: parents were regularly arriving late to pick up their children, forcing teachers to remain on the clock without pay. The instinctive response was a fine, a small monetary penalty for each late pickup.

Six of the ten centers introduced the fine. Four served as an unwitting control group.

Gneezy and Rustichini tracked weekly late pickups across both groups for seventeen weeks. The prediction from deterrence theory was clear: the fine should reduce lateness. Parents would weigh the penalty against the convenience of arriving late and adjust their behavior accordingly.

What happened was the opposite. Within the six centers with fines, the rate of late pickups approximately doubled. Within the four centers without fines, there was no significant change.

The mechanism, as the authors explain, was the information channel. Before the fine, arriving late carried a social cost: guilt, discomfort, the sense of imposing on a teacher who had somewhere to be. That social pressure was doing quiet regulatory work. The fine converted the interaction from a social obligation ("I owe it to the teachers to arrive on time") into a commercial transaction: "I can purchase an extra thirty minutes by paying the fee." Once the fine made tardiness purchasable, the social norm was displaced. The fine had not raised the cost of being late; it had cleared the conscience of being late.

In week 17, the fine was removed. Lateness did not return to its pre-fine baseline. It held at the elevated level established during the fine period. The social norm, once displaced, did not reconstitute itself. The incentive had produced a permanent deterioration in the baseline behavior it was designed to correct.

Moral hazard: when insulation from consequences changes choices

A related but distinct phenomenon runs through insurance markets, banking, and government policy: moral hazard. The term comes from the early insurance industry, where underwriters observed that insured property was more likely to burn down than uninsured property, not necessarily through fraud, but because owners of insured property had less reason to be careful. When someone else absorbs your downside risk, your incentive to manage that risk diminishes.

As the Library of Economics and Liberty's treatment of moral hazard in health insurance explains⁴, once you have comprehensive coverage for a medical cost, you will use more of the covered service than you would if you were paying out of pocket, because the marginal cost to you at the point of use has fallen to near zero. This is not fraud; it is rational price-responsive behavior. But it produces overconsumption of medical services relative to the socially optimal level.

The moral hazard problem becomes most acute when the insulation from consequences is provided by a third party who has no ability to price or monitor the individual risks being taken. A bank whose losses will be covered by a government backstop has weaker incentives to avoid risky lending than a bank that bears its own losses fully. This dynamic played out visibly in the lead-up to the 2008 financial crisis, where institutions extended mortgage credit they would not have extended if they had retained the risk themselves, rather than originating loans and selling them to investors who then sold them again, diffusing accountability across a chain where no individual party bore the full consequences of a default.

Mexico City's driving restriction: good design, perverse outcome

The Hoy No Circula program, launched in Mexico City in 1989, was designed to reduce vehicle emissions by restricting each car from being driven one weekday per week, based on the last digit of its license plate. At maximum penetration, the policy should have reduced the vehicle fleet on any given day by 20 percent.

Researcher Lucas Davis, in a study examining the program's air quality effects and published in the Journal of Political Economy (2008)³, found no improvement in air quality across pollutants. The reason was behavioral: many households acquired a second vehicle with a different license plate ending so they could drive every day of the week.

The newly purchased vehicles were disproportionately older, used models imported from other parts of Mexico or from the United States, vehicles that lacked modern emissions controls and that emitted substantially more pollution per mile than the cars they were meant to offset. The net effect, documented through pollution monitoring data, was no improvement and possible deterioration relative to the counterfactual.⁶

The Hoy No Circula case illustrates a third failure mode in incentive design: the substitution effect. When a policy imposes a cost on one behavior, rational actors find substitutes. If the substitutes are worse, in this case more polluting, the intended improvement is partially or fully canceled. Policy designers who model only the targeted behavior and not the substitution landscape routinely produce underwhelming results.

What makes incentives work

The Hanoi bounty, the Haifa fine, and the Mexico City driving restriction are not indictments of incentive-based policy. They are case studies in where incentive design fails and why. The lessons are specific enough to be actionable.

Match the incentive to the behavior you can actually measure. The Hanoi program incentivized tail production, not rat elimination. Any incentive that rewards a measurable proxy for the desired behavior is vulnerable to gaming if the proxy can be gamed at lower cost than the real behavior. Tighter monitoring, with inspectors who verified that tails came from rats at known locations, would have substantially reduced the exploitation.

Audit the information channel before deploying the price channel. When the target behavior is currently regulated by a social norm, reciprocity, or professional ethic, introducing a price incentive can crowd out the existing mechanism rather than reinforce it. The Haifa administrators could have achieved their goal with sharper social reinforcement, thanking on-time parents publicly, making the teacher's situation more visible, rather than a fine that converted the relationship into commerce.

Test at small scale before locking in. The daycare fine was rolled out across six centers simultaneously. If one center had piloted the fine for two or three weeks with careful tracking, the perverse outcome would have been visible before it became entrenched. Small-scale iteration is the most reliable way to detect information-channel effects before they produce durable behavioral shifts.

Model substitution, not just the targeted behavior. Every policy that restricts or taxes one option creates incentives to use other options. If the substitutes are worse on the dimensions that matter, the policy can produce a net negative outcome despite achieving its narrow behavioral target. Mexico City reduced driving of certain cars on certain days; it did not reduce total vehicle miles traveled or total emissions.

Incentives are the essential vocabulary of economics. As the NBER paper by Benabou and Tirole formalizes in a theoretical framework for prosocial behavior⁵, the interaction between extrinsic incentives and intrinsic motivation is not additive: it is sometimes competitive. Introducing a payment for something people were already doing voluntarily can reduce the total amount of the behavior if the payment crowds out the reputational or self-image value of doing it for free.

The implication for policy, management, and personal decision-making is the same: before you design an incentive, map the current motivational landscape. Know what is already working, and what your incentive might inadvertently displace.