Peltzman Effect: When a Safety Measure Changes How People Act

The argument starts from an assumption about people rather than about cars. Treat risk as something a driver chooses rather than something that happens to them: speed, following distance and attention are all adjustable, and each adjustment buys time or convenience at the price of danger. Make the car safer and the price of driving fast has fallen, so a driver who was content with their old level of danger can take some of the gain in speed instead of in safety. That is Sam Peltzman's mechanism, and it is a straightforward piece of economics rather than a quirk of psychology.

The strong form of the idea goes further. Gerald Wilde's risk homeostasis theory, from the 1980s, proposed that people hold a target level of risk and adjust until they reach it, which would mean that safety measures can never reduce harm on average. That version has not held up: the seat belt evidence below shows a real reduction in deaths that behavior did not claw back.

The reason the debate ran for decades is that the offsetting behavior is hard to see. Deaths are counted carefully, but following distances and attention are not, so researchers test the theory indirectly by watching the people who get no protection from the device. If belted drivers really drive worse, deaths among pedestrians, cyclists and motorcyclists should rise with belt use. Cohen and Einav ran exactly that test on 15 years of state data, and the estimate came out statistically indistinguishable from zero, which is the main evidence against the complete offset.

So the honest summary has two parts. Behavioral adaptation to safety equipment is real and has been observed directly, as with the Oslo taxi drivers who followed more closely in cars with antilock brakes. The claim that adaptation cancels the benefit is not supported, and the version of the Peltzman effect that circulates as an argument against safety rules is the version the data rejected. Using it that way is the most common error on this topic.

Sam Peltzman published "The effects of automobile safety regulation" in the Journal of Political Economy in 1975. The subject was the federal standards that began to apply to US cars from the 1966 model year: seat belts, padded instrument panels, dual braking systems and energy-absorbing steering columns. Engineering studies of those devices implied that annual highway deaths would have been about 20 percent higher without them.

Peltzman's objection was that those studies hold driver behavior fixed. He modeled the demand for driving intensity, estimated the relationships from time-series and cross-section data on fatality rates, and concluded that the offsets from non-regulatory demand for safety and from driver response were "virtually complete," so that the regulation had not reduced highway deaths. His time-series estimates, though not his cross-section ones, suggested that occupant lives were being saved at the cost of more deaths among pedestrians and cyclists and more nonfatal accidents, which is the pattern his model predicted.

The strong conclusion was contested almost at once. Leon Robertson's 1977 critique in the Journal of Economic Issues reported errors in the data and argued that Peltzman's model failed to reproduce fatality rates from before the regulation, and that evidence he had passed over showed the standards reducing deaths.

The best-identified test came in 2003, when Alma Cohen and Liran Einav published "The effects of mandatory seat belt laws on driving behavior and traffic fatalities" in The Review of Economics and Statistics. Their panel covered all 50 states and the District of Columbia from 1983 to 1997, with measured belt usage rather than law dummies alone, and they instrumented usage with the laws and their enforcement type to deal with the fact that cautious drivers buckle up more. Laws with secondary enforcement raised usage by about 11 percentage points and laws with primary enforcement by about 22. A 1-percentage-point rise in usage was associated with about 136 fewer deaths a year, an effect less than half the size the National Highway Traffic Safety Administration was using at the time. Usage had no statistically significant effect on deaths among people outside cars, which is the compensating-behavior prediction, and the result held when the test was narrowed to pedestrians.

Risk compensation is worth looking for whenever a protection is added to an activity people control. The rows below are illustrations, not findings from the studies cited on this page.

The useful moves are about keeping the safety gain in the safety column instead of spending it.

• Name the behavior the device makes cheaper. A system that makes braking safer makes tailgating cheaper; once that is written down, it can be watched.
• Fix the limit before the equipment arrives. A following distance in seconds, a time in the sun, a maximum speed on a route: a rule set in advance is not renegotiated by the feeling of being protected.
• Count the people who are not protected. If a measure helps those inside and the harm moves outside, the total has not improved. This is the test that settled the seat belt argument.
• Do not treat partial adaptation as a reason to skip the protection. The measured adaptations are much smaller than the gains, so the device still wins.
• When you hear the Peltzman effect used to argue against a safety rule, ask for the number. The complete offset is the claim that failed; the size of any partial offset is an empirical question with an answer for each measure.
• For designers: avoid signaling the protection louder than the protection itself. A device that advertises invulnerability invites exactly the behavior that eats the benefit.

The Peltzman effect sits inside a wider family of ideas about safety and behavior.

• It is not the same as risk compensation in general, which is the broad finding that people adjust behavior when their perceived risk changes. The Peltzman effect is the specific economic claim about safety regulation, and the specific strong version of it, a near-complete offset, is the part the evidence rejected.
• It is not zero-risk bias, the preference for eliminating one risk completely over reducing several by more in total. That is a choice between options; the Peltzman effect is about what happens after an option is in place.
• It is not the illusion of control, which is overestimating your influence over an outcome that is mostly chance. A driver who shortens a following distance because the brakes are better is responding to a real change in control, not an imagined one.