Intended for healthcare professionals

Education And Debate

For and againstDoes risk homoeostasis theory have implications for road safetyForAgainst

BMJ 2002; 324 doi: https://doi.org/10.1136/bmj.324.7346.1149 (Published 11 May 2002) Cite this as: BMJ 2002;324:1149

Does risk homoeostasis theory have implications for road safety

Risk homoeostasis (also called risk compensation) theory predicts that, as safety features are added to vehicles and roads, drivers tend to increase their exposure to collision risk because they feel better protected. Gerald Wilde provides evidence for it and suggests that it should be used to inform road safety strategies. Leon Robertson and Barry Pless, however, argue that the evidence is deeply flawed and that the theory is little better than an excuse for doing nothing

For

  1. Gerald J S Wilde (wildeg{at}psyc.queensu.ca), professor emeritus of psychology
  1. Department of Psychology, Queen's University, Kingston, Ontario, Canada
  2. a 11 Dixon CT, Nogales, AZ 85621, USA,
  3. b Montreal Children's Hospital, C538 2300 Tupper, Montreal QC H3H 1P3, Canada

    Anyone wishing to reduce the risk of misfortune on the road to zero can do so by never using the roads, but that person would also miss all the benefits accruing from road travel and thus live a greatly diminished life. Suboptimal risk taking also occurs if a person underestimates or overestimates the danger of a given activity, because that person would either take too much risk or too little for greatest net benefit. A person learns to assess risk by perceiving the outcomes of decisions. Our intuitive assessment of risk is honed by our experience and that of others, sometimes communicated through the mass media. This feedback will thus confirm or correct a person's perception of the size of the four utility factors that determine the optimal (or target) level of risk (see box).

    Theory of risk homoeostasis

    While some actions entail more danger (probability×magnitude of loss) than others, there is no behaviour without some risk. The challenge, therefore, is to optimise rather than eliminate risk. This optimal, or target, level of risk is that which maximises the overall benefit (probability×amount). Four utility factors determine the target level of risk:

    • The expected benefits of risky (comparatively dangerous) behaviour options

    • The expected costs of comparatively cautious behaviour options

    • The expected benefits of comparatively cautious behaviour options

    • The expected costs of risky behaviour options.

    The first two factors increase the target level of risk, whereas the last two diminish it. Thus, a rational person should opt for the behaviour option (or set of alternatives) that is perceived as most likely to deliver the greatest net benefit.

    The figure depicts the dynamic of this closed-loop control process for driving a road vehicle. A driver compares the perceived risk of current action (box b) with the target level of risk (box a). When there is a discrepancy the person wishes to reduce it (box c) and takes some adjustment action (box d). All actions taken by all drivers in a jurisdiction over a year determine that year's loss from road vehicle crashes in that jurisdiction (box e). So, the actions determine the amount of loss, but the loss in turn affects the levels of risk perceived by drivers. This, in turn, influences their subsequent decision making and thus the subsequent crash loss, and so forth in an ongoing circular process: behaviour determines the amount of loss, and the amount of loss determines behaviour.1

    Evidence for risk homoeostasis in road crashes

    The Swedish experience with the change from left hand to right hand traffic in 1967 offers a telling example. Contrary to government and safety experts' expectations, the change was followed by an immediate and major reduction in the per capita traffic death and injury rates, which, however, returned to original levels within two years. These findings were explained as follows.2 The change led to a sudden surge in the level of perceived risk; this was now much higher than the level of risk accepted, and people drove more cautiously. This increased caution reduced the rate of road crashes. After some time, people discovered, through their own and others' experience and through the mass media, that the roads were less dangerous than they had thought. Now, the risk was not as high as was tolerated, and people became less cautious in their actions, causing the crash rate to return to “normal.” When Iceland changed over to right hand traffic in 1968 it experienced a similar short term fluctuation in the rate of road crashes.

    This homoeostatic mechanism may be compared to a thermostat. The thermostat determines the action of the heating system and thus the temperature in the house, while the temperature in the house controls the actions of the thermostat. Fluctuations may occur, but the time averaged temperature remains the same, unless the desired (target) temperature is set at a different level. Analogously, major and lasting reductions in the rate of death, injury, or other damage due to driving, smoking, being overweight, terrorism, extreme sports, etc, can be obtained by interventions that effectively lower the target level of risk.1

    The figure places people's perceptual, decision making, and vehicle handling skills outside the closed loop. Similarly, the sensitivity of the thermometer (analogous to box 4), the inertia of the thermostat switch (box 2), and the heating capacity of the furnace (box 3) do not affect the time averaged temperature, although they will affect the wavelength and amplitude of short term temperature fluctuations. Indeed, improving driving skills through advanced driving courses has not been found to lead to greater road safety. This is attributed to the fact that, although these courses enhance driving skill, they also lead to greater confidence, with confidence growing faster than skill. The result is overconfidence, and thus a reduced level of perceived risk (box b); consequently, perceived risk will less often exceed target risk, which leads to less caution in driving.3

    In the course of the 20th century, the death rate from road vehicle crashes dropped substantially for distance driven. This drop is generally attributed to the construction of more forgiving roads and more crashworthy vehicles. However, as risk homoeostasis theory would predict, the improvements do not imply a similar reduction in the per capita death rate, because greater safety per kilometre allows more and faster mobility while the per annum risk of death on the road remains the same. In the absence of changes in the target level of risk, and in contrast to the mortality per unit distance driven, no reduction in the annual per capita mortality should occur. Longitudinal data from North America show about the same per capita traffic death rate in 1996 as in 1923; the distance travelled per capita increased by about the same factor (roughly 10) as the death rate per unit distance decreased.1

    Figure1

    Homoeostatic model relating the rate of road vehicle crashes in a jurisdiction to the level of caution in road user behaviour and vice versa, with the average target level of risk as the controlling variable1

    In the United States, as in other countries, there have been major fluctuations over time in the per capita traffic mortality. These can be explained as a function of the business cycle and its effect on the target level of risk. During economic upswings the first utility factor listed in the box (expected benefits of risky behaviour) is enhanced and the fourth (expected costs of risky behaviour) is reduced relative to disposable income. Annual variations in the traffic death rate, over periods ranging from 18 to 39 years in eight different countries, have been found to strongly correlate with economic prosperity, the coefficients varying between r=0.68 and 0.96 (median r=0.87).1

    Cross sectional data show that on roads and streets where the rate of crashes is comparatively low, drivers move commensurably faster4 and thereby keep the crash rate per hour of exposure virtually at the same level regardless of where the driving is done.1

    The intensity of law enforcement has not been shown to affect the per capita crash rate.5 Crackdowns on speeding or driving while intoxicated may reduce speed related or alcohol related death rates, but they do not reduce the overall death rate, 6 7 because of the phenomena of “accident migration” and “accident metamorphosis,”1 meaning that accidents move from one location to another or their immediate causes change.

    Mandatory standards in vehicle manufacturing, such as the installation of air bags, do not reduce death and injury rates in car drivers—because of the added feeling of safety they provide, they lead to more “aggressive” driving.8 Real life experiments on antilock braking systems (ABS) and seatbelt wearing have shown that drivers compensate for the potential safety benefit by driving faster and following other vehicles more closely. 9 10 Computer simulation of risk taking behaviour suggests that people, when given feedback on their decisions, adapt to changes in external danger11 and quickly learn to optimise the risk they incur.12

    Reducing the target level of risk

    There are four possible approaches to reducing the target level of risk for drivers—rewarding particular safe behaviours, rewarding drivers who have not crashed, punishing particular unsafe behaviours, and punishing drivers for having crashed. The third option, which is used predominantly in most societies, has not been found to be particularly effective. 13 14

    However, incentives for crash-free driving (the second approach) have been shown to be particularly useful. The promise of free renewal of the driver's licence in California in return for crash-free driving was followed by a 22% reduction in crashes in the first year and by a 33% reduction in the second.15 Norwegian beginner drivers, who were promised by their insurance company complete reimbursement plus interest of the insurance surcharge for young drivers, responded by a 35% reduction in crashes.16 German truck drivers who were given the prospect of a cash bonus for every half year of crash-free driving subsequently caused many fewer crashes; this incentive programme has been in effect for over 30 years without any dwindling in its power to prevent accidents.17

    These examples show that the “risk thermostat” can be reset. The success of incentive programmes1 compares favourably with the more traditional countermeasures. The incentive approach would therefore seem to warrant much wider implementation in the promotion of safety and lifestyle dependent health.—Gerald J S Wilde

    Footnotes

    • Competing interests None declared.

    References

    1. 1.
    2. 2.
    3. 3.
    4. 4.
    5. 5.
    6. 6.
    7. 7.
    8. 8.
    9. 9.
    10. 10.
    11. 11.
    12. 12.
    13. 13.
    14. 14.
    15. 15.
    16. 16.
    17. 17.

    Against

    1. Leon S Robertson, retired lecturer in epidemiology and public health, Yale Universitya,
    2. I Barry Pless, editor, Injury Preventionb
    1. Department of Psychology, Queen's University, Kingston, Ontario, Canada
    2. a 11 Dixon CT, Nogales, AZ 85621, USA,
    3. b Montreal Children's Hospital, C538 2300 Tupper, Montreal QC H3H 1P3, Canada

      Occasional change in driver behaviour as a result of perceived change in risk gives some credence to risk homoeostasis theory. But the key words in that sentence are “occasional” and “perceived.” For example, some drivers sometimes slow down in rain, but that does not mean that they do so systematically or that they know exactly how much to slow down to maintain constancy of risk. The notion that people have a constant point of acceptable risk, pay sufficient attention to risk, or have the knowledge and ability to constantly adjust their behaviour to achieve so called risk homoeostasis is ludicrous in view of what is known about human limitations. It also makes little sense in light of the sharp and consistent decline in death rates from vehicle crashes and, indeed, all other sources of injury.

      Driver reaction time varies from 0.5 seconds to more than 2 seconds.1 Thus, when driving at 24 km/h, the fastest reactors require 8.2 metres to stop a car with perfect brakes, whereas the slowest reactors require 18.3 m. A pedestrian who suddenly steps into even slow moving traffic close to a car will be injured despite the best effort of the driver to stop. In experiments where participants had to judge their speeds and the speeds of oncoming vehicles in order to decide whether to overtake a slower vehicle, when driving at 81 km/h, the drivers' average estimated distance needed to pass a vehicle and not be struck by oncoming vehicles was half the distance necessary.1

      Only about 4% of licensed drivers in most countries are involved in a crash reported to police in a given year. Thus, perception of risk cannot be based on frequent experience. Widespread participation in lotteries and casino gambling reveals how poorly the public understands the laws of probability. The only limitless human capacity is the capacity for self delusion, one that proponents of risk homoeostasis possess to an awesome degree.

      In even a short drive a driver is likely to pass within two or three metres of a driver who is depressed, drugged, drunk, distracted, or just stupid, none of which is conducive to constant risk assessment. Therefore, it is remarkable that the crash rate is not many times worse than it now is.

      Modern neoclassic economists claim credit for risk homoeostasis theory,2 but it was used in the 19th century to oppose the mandatory requirement of safety equipment on trains.3 It is almost invariably trotted out today to bolster the views of opponents of many safety measures, such as the compulsory wearing of bicycle helmets. Unfortunately, economic research is too often anti-regulation ideology cloaked in esoteric mathematical formulae to give it the appearance of genuine science.

      Flaws in the evidence

      The primary issue for road safety is the extent, if any, that risk homoeostasis offsets the effectiveness of vehicle and road modifications to reduce injury incidence or severity. Most supposed evidence for such an offsetting effect is based on faulty research and is countered by overwhelming contrary findings.4

      Typical studies claiming a risk compensation effect examine trends in mortality relative to changes in vehicle modifications or laws requiring safety measures such as seat belt use or motorcycle helmet use. One of the early studies, for example, projected expected death rates for vehicle occupants and “pedestrians” before and after motor vehicle safety standards were inaugurated in the United States.5 The author, Peltzman, said that reductions in occupant fatalities were offset by increases in “pedestrian” fatalities because drivers, protected by the safety standards, drove more riskily. A simple examination of the data indicated that Peltzman had counted motorcyclists as “pedestrians,” at a time when registrations of these most dangerous vehicles were doubling every five years. Nor did he attempt to disaggregate regulated from non-regulated vehicles to assure that the regulated vehicles were over-involved in actual pedestrian deaths. Finally, the statistical model used to project death rates did not accurately project the rates before regulation. This simple test of the model was not included in the original study.6 Flaws of this kind are evident in most studies used to support risk homoeostasis.

      Subsequent analysis of disaggregated data found that regulated vehicles had 40% lower occupant mortality and were less likely to have struck pedestrians, bicyclists, or motorcyclists compared with non-regulated vehicles.7 This is not surprising since the standards included requirements for side running lights to make vehicles more visible, redundant braking systems to make them more reliable, and reduced glare in windscreens to help drivers see other vehicles and pedestrians.

      Despite the obvious need to use disaggregated data to reveal who hit whom, studies using aggregated data continued. One, for example, attempted to correlate deaths of vehicle occupants and other road users among US states to claimed use of vehicle safety belts and factors such as vehicle miles driven, percentage urban population, average speed on rural roads, alcohol sales per capita, percentage youth in the population, and income per capita. After statistical adjustment for these factors, it was claimed that higher fatalities among non-occupants occurred in states with higher claimed use of safety belts.8 The study relied on self reported belt use in the Centers for Disease Control's behavioural risk factor survey. It has been known for decades that self reported belt use is not valid, particularly so in the Centers for Disease Control's data.9 And the non-occupants in the study included motorcyclists, almost half of whose deaths do not involve collisions with other vehicles.

      Even more compelling than disaggregated crash data are observations of the driving behaviours of users and non-users of safety belts before and after laws on wearing safety belts were introduced. In Newfoundland, for example, there was no adverse change before and after the law was enforced in four important risk related driving behaviours—speed, stops at intersections during the amber phase of traffic lights, turning left in front of oncoming traffic, and the following distance from the vehicle in front. Observed belt use increased from 16% to 77%. The only changed outcome was slower speeds on freeways. In Nova Scotia, which lacked a seat belt law and acted as a control province, there were no changes in the same behaviours, again contrary to risk homoeostasis.10

      Another test of the hypothesis that increased protection is offset by riskier behaviour was provided by the introduction of air bags. If drivers of cars equipped with air bags wished to keep their risk constant they would have to reduce their use of seat belts, but they did not. Observed belt use in cars with air bags remained the same as in those without.11 It is noteworthy that the main proponent of risk homoeostasis uses a single study of the introduction of air bags to support the hypothesis, whereas critics cite six others that reach a contrary conclusion.4

      The most compelling argument against risk homoeostasis is the observation that occupant death rates in passenger cars per distance travelled fell by nearly two thirds in the United States from 1964 to 1990. A comprehensive study of the effect of vehicle modifications, laws on use of seat belts, and reductions in drunk-driving indicates that about 90% of this reduction was due to vehicle modifications. Moreover, there is no association of these reductions with any increases in fatalities to occupants of cars in collision with the safer vehicles or any other road users struck by them.12

      Conclusion

      Road builders, vehicle manufacturers, and policy makers can be assured that they can improve road safety standards without having them offset by drivers attempting to adjust their alleged risk thermostats.—Leon S Robertson, I Barry Pless

      Footnotes

      • Competing interests None declared.

      References

      1. 1.
      2. 2.
      3. 3.
      4. 4.
      5. 5.
      6. 6.
      7. 7.
      8. 8.
      9. 9.
      10. 10.
      11. 11.
      12. 12.