Do safety features make people take more risks?
The idea
In 1975 the economist Sam Peltzman published a long paper in the Journal of Political Economy asking whether the wave of American car safety rules of the 1960s, which mandated devices such as seat belts, had actually saved lives. Engineering studies had predicted that without these devices annual highway deaths would be about 20 percent higher. Peltzman argued those studies missed something: drivers are not passive. If a crash becomes less dangerous to you, the cost of driving a little faster or a little less carefully falls, and some people will buy that extra speed with their new margin of safety.
His conclusion was blunt. Using aggregate national time seriestime series A sequence of measurements of the same quantity taken at regular intervals over time., he found that some deaths of people inside cars seemed to have been avoided, but that this came with more pedestrian deaths and more nonfatal crashes, so that the offsets were, in his words, “virtually complete.” The general idea, now called risk compensationrisk compensation The idea that people respond to a safety improvement by taking more risk, using up some of the protection it gives. or the Peltzman effect, is that people partly spend a safety improvement on taking more risk.
A concrete version: a driver who has just started wearing a belt feels a little more protected and takes a bend a little faster. The belt still protects them if they crash. But the pedestrian on the bend has no belt, so any added risk lands on someone the device does nothing for.
Why it might change a decision
The idea gives a specific, checkable prediction. If a safety device mainly changes behaviour, then harm to the people it does not protect should rise after it spreads. That changes what you would measure when judging any safety measure: not only the outcomes for the protected group, but also outcomes for bystanders, and the rate of minor accidents, which more careless behaviour should increase even when injuries do not.
How solid is this?
Peltzman’s original claim rests on one country’s aggregate time series, which cannot easily separate the effect of regulation from everything else changing on American roads in the 1960s. It has been heavily tested since, and the best-known direct test did not support it.
Alma Cohen and Liran Einav, in the Review of Economics and Statistics in 2003, used state-level panel datapanel data Measurements of the same set of units, such as states or people, repeated over several years. covering all 50 states and Washington DC from 1983 to 1997. Their trick was to split deaths into car occupants, whom belts protect directly, and pedestrians and cyclists, whom belts can only affect through how people drive. If belted drivers drove worse, more belt use should go with more deaths outside the car. Once they treated belt use as something the laws caused, using the timing of state laws as an instrumental variableinstrumental variable Something that shifts the thing being studied but has no other route to the outcome, used to tell cause apart from mere association., that relationship was not statistically significant. Belts did save lives: they estimated about 136 lives a year for each percentage pointpercentage point The plain difference between two percentages, so a move from 40 to 45 percent is a rise of 5 percentage points. rise in usage. That was well short of the official figure, though. Their estimates were less than half of the federal government’s 5,536 lives a year.
A null result is not proof that no compensation happens; it means the data could not detect it at the scale examined. Small offsets could still hide inside the estimates.
A second case shows how hard even a narrow version is to pin down. Ian Walker, a psychologist, used a bicycle instrumented to record how close vehicles came when passing it on roads in Salisbury and Bristol, with a single male experimenter riding. In a 2007 paper in Accident Analysis and Prevention he reported that drivers passed significantly closer when the rider wore a helmet, and gave more room when he wore a long wig and looked female from behind. This is compensation by other people: drivers perhaps judging a helmeted rider as more experienced or predictable. Walker himself framed it as drivers reading appearance, not as a measured increase in danger.
In 2013 Jake Olivier and Scott Walter reanalysed the same 2,355 passing events in PLOS ONE. Counting only passes closer than one metre, a common safety rule, they found no association with helmet wearing, while distance from the kerb, vehicle size and city did matter. They suggested the original result reflected a sample so large that trivial differences became significant. Walker and Dorothy Robinson replied in 2019 that splitting distances at one metre discards information and lacks statistical powerstatistical power The chance that a study will detect an effect of a given size if that effect is really there., and that the whole distribution of passing distances moved modestly closer for helmeted riders. Both sides have a point, and neither study can get past the fact that every pass involved the same rider.
The strongest competing explanation for apparent compensation is selection: the people who adopt safety gear, or the places that adopt safety rules, may differ in ways that also affect risk. The popular version of the Peltzman effect, that safety measures are cancelled out, outruns the data. The narrower claim, that behaviour can shift somewhat in response to safety, remains plausible and hard to measure.
Sources
- Peltzman, S. (1975). The effects of automobile safety regulation. Journal of Political Economy, 83(4), 677-725. doi:10.1086/260352
- Cohen, A., & Einav, L. (2003). The effects of mandatory seat belt laws on driving behavior and traffic fatalities. Review of Economics and Statistics, 85(4), 828-843. doi:10.1162/003465303772815754
- Walker, I. (2007). Drivers overtaking bicyclists: Objective data on the effects of riding position, helmet use, vehicle type and apparent gender. Accident Analysis and Prevention, 39(2), 417-425. doi:10.1016/j.aap.2006.08.010
- Olivier, J., & Walter, S. R. (2013). Bicycle helmet wearing is not associated with close motor vehicle passing: A re-analysis of Walker, 2007. PLOS ONE, 8(9), e75424. doi:10.1371/journal.pone.0075424
- Walker, I., & Robinson, D. L. (2019). Bicycle helmet wearing is associated with closer overtaking by drivers: A response to Olivier and Walter, 2013. Accident Analysis and Prevention, 123, 107-113. doi:10.1016/j.aap.2018.11.015