Risk Compensation: Why Feeling Safer Changes How You Drive

Risk compensation is the general finding that people respond to feeling safer by accepting a little more risk, and this page is about that finding in behavior: how drivers, cyclists and workers adjust when their protection improves. The related Peltzman effect is the specific economic claim, made about US car safety regulation in 1975, that the adjustment cancels the regulation's benefit; that claim has its own page and its own evidence.

The mechanism is ordinary. A driver chooses a speed and a following distance by weighing how fast the trip goes against how likely a crash feels. Better brakes lower the felt likelihood, and the trade-off resettles at a slightly faster speed or a shorter gap. Nothing about this is irrational in itself. The problem is in the accounting: engineers and regulators estimate a measure's benefit from the physics, as if behavior would hold still, and behavior does not hold still.

The strongest version of the idea is Gerald Wilde's theory of risk homeostasis, published in Risk Analysis in 1982. Wilde argued that each population has a target level of risk it is willing to accept and that the accident rate is governed by that target alone: make the road or the car safer and people adjust until the accident rate returns to where it was. That is a claim of complete offset, and it has not held up. Leonard Evans examined traffic fatality data in 1986 and found every data set he looked at incompatible with the theory; fatalities per mile differ several-fold between road types and have fallen many-fold over decades, which a constant target could not produce. Wilde replied the same year, but the weaker claim is the one that survived: people compensate, the amount varies by measure and setting, and it has to be measured rather than assumed.

Compensation is largest when the protection is noticeable and under the person's control. Antilock brakes announce themselves every time they engage, and a driver can use them to brake later. A stronger roof pillar is invisible and offers nothing to spend, so little compensation is expected.

The Munich taxi study is the field experiment most often cited for risk compensation, because its design removed the usual objection that cautious drivers are the ones who buy safety equipment. Bernhard Biehl, Michael Aschenbrenner and Gerhard Wurm, in a project for the German federal road research institute, arranged for a Munich taxi company to fit antilock brakes to part of its fleet and to assign drivers to cars at random, so that drivers did not choose whether they drove with ABS. Their report appeared in 1987 in the institute's series Unfall- und Sicherheitsforschung Strassenverkehr; Aschenbrenner and Biehl summarized it in English in 1994 in the volume Challenges to Accident Prevention, edited by Rudiger Trimpop and Gerald Wilde.

The accident analysis covered 1981 to 1983 and compared 21 ABS-equipped taxis with 500 random samples of taxis without ABS drawn from the same fleet. The ABS taxis were not involved in fewer accidents than the comparison vehicles, and the report concluded that under big-city traffic conditions the system showed no safety gain. A second phase, in 1985 and 1986, looked at behavior. Trained observers rode as passengers on 113 drives without the drivers knowing their driving was being rated, and the observers did not know which cars had ABS. The ABS drivers cut curves more, held their lane less precisely, looked less far ahead in traffic, merged with poorer adjustment to other vehicles and created more conflict situations. The 1994 chapter named six mechanisms by which the compensation could happen, including inattentive driving that follows from feeling safer and overestimation of what the system can do in some conditions.

The study has limits worth stating plainly. Twenty-one cars is a small treatment group, the result is about taxis in one city, and the behavioral observations came two years after the accident data. The reason it matters is the random assignment: the drivers who drove more aggressively had not chosen the safer car, so the car changed the driving. That is the pattern risk compensation names, measured in traffic rather than in a questionnaire.

The pattern appears wherever a safety measure is visible to the person it protects. The rows below are illustrations, not findings from the cited studies.

Risk compensation cannot be switched off, because it is a reasonable response to a real change. What can be managed is how much of a safety gain gets spent.

• Separate the decision from the equipment. Choose your speed, your following distance and your route as a rule for the road, and let the protection be a margin on top of the rule rather than a license to move it.
• Notice the sentence "it can handle it." When the justification for doing something riskier is the equipment rather than the conditions, that is compensation talking.
• Prefer protection that offers nothing to spend. Measures people cannot feel, such as a stronger structure or a better-designed intersection, tend to deliver their full engineering benefit.
• Measure after the change. For a fleet or a workplace, the question is not whether the device works in a test but what the accident and near-miss record looks like six months after people have adjusted to it.
• Keep the limits in view. The compensation happens most easily when the protection is felt and its limits are not; a driver who knows that antilock brakes help only when the gap in front is kept is in a position to keep the gap.

Risk compensation sits in a family of ideas about safety and behavior, and the neighbors differ in what they claim.

• It is not the Peltzman effect, the specific 1975 claim that US car safety regulation produced no net reduction in deaths because drivers offset the gain. The Peltzman effect is risk compensation applied to regulation and pushed to a complete offset, and that strong version is the part the later data rejected. This page is about the general finding and the partial adjustments it describes.
• It is not optimism bias, the tendency to rate your own chance of a bad outcome below other people's. A compensating driver may be accurate about the risk and still choose to spend part of the new margin on speed; the adjustment is a preference, not a misestimate.
• It is not the illusion of control, which is overestimating your influence over an outcome governed by chance. A driver who follows closer because the brakes are better is responding to a real change in control; the error, if there is one, is in how large the change is and in forgetting that other road users got none of it.