82 Injury Prevention 2000;6:82–90

HADDON MEMORIAL LECTURE Inj Prev: first published as 10.1136/ip.6.2.82 on 1 June 2000. Downloaded from

Risky business: safety regulations, compensation, and individual behavior

James Hedlund

Editors comment: We are Government regulations and industry practices paper, behavioral adaptation describes all be- proud to be able to bring to constrain our behavior in many ways in an havioral change in response to perceived our readers this full text version of the Haddon attempt to reduce injuries. Safety features are changes in risk and risk compensation describes Memorial Lecture delivered designed into products we use: cars now have the special case of behavior change in response at the recent Fifth World airbags; medicine bottles have “childproof” to laws and regulations. The distinction Conference on Injury Pre- vention and Control in caps. Laws require us to act in a safe manner: we becomes murky at times: if a new safety feature New Delhi, India. James must wear seat belts while driving and hard hats appears on all chain saws, any behavioral reac- Hedlund oVers a brilliant in construction areas. But do these measures tion won’t depend on whether the feature is review of one of the most important areas of debate influence our behavior in other ways? Risk com- required by government regulation or adopted in the entire field of injury pensation theory hypothesizes that they do, that voluntarily by all manufacturers. The risk control. This is the most we “use up” the additional safety though more compensation definition adopted here focuses complete, most perceptive, and well balanced apprais- risky actions. on the injury prevention strategies of greatest als of this complex issue I This paper surveys risk compensation by controversy, where government attempts to have ever read. Take the reviewing its history, discussing its theoretical increase safety by law or regulation. time to digest it completely. Our thanks to the Insur- foundations, outlining evidence for and against ance Institute for its claims, and providing the author’s own views. Safety for agreeing to per- It concludes by discussing the relevance of risk A brief history of risk compensation mit us to publish it. We all change our behavior in response to some http://injuryprevention.bmj.com/ compensation for injury prevention workers changes in perceived injury risk. Most obvi- who seek to reduce unintentional injuries. ously, we may take additional precautions if we believe our risk has increased. When and The setting: injury prevention strategies are icy, we may walk more carefully and risk compensation for fear of falling and we may drive more slowly Injury prevention as a discipline began when to be sure that we can stop safely. But it is not injuries were understood to be both predictable at all obvious that we change our behavior in and preventable. Most injuries are the unin- response to every increase or decrease in risk. tended consequences of individual actions in a The heart of the risk compensation debate lies risky environment; they are not due to fate or to in determining which risk changes will produce problem behavior. This understanding led to compensating behavioral change. three fundamental injury prevention strategies, The early risk compensation literature deals as described in the comprehensive report with safety, as traYc crashes have been the on September 28, 2021 by guest. Protected copyright. Injury in America1: largest cause of accidental injury and death in x Persuade persons at risk to change their motorized countries and consequently pro- behavior, duced extensive safety regulations. By mid- century behavioral adaptation had been recog- x Require behavior change by law or adminis- 2 trative rule, nized but not seriously studied. x Provide automatic protection through prod- uct and environmental design. RISK COMPENSATION AND ECONOMICS: Injury prevention policymakers and workers PELTZMAN’S EVALUATION OF MOTOR VEHICLE generally agreed on the relative priorities of SAFETY STANDARDS these strategies. As Injury in America again Risk compensation in response to government reports: “Each of these general strategies has a actions became a public issue with University role in any comprehensive injury-control pro- of Chicago economist Sam Peltzman’s 1975 gram; however a basic finding from research is paper “The eVects of automobile safety that the second strategy—requiring behavior regulation”.3 Nine years earlier, in 1966, the change—will generally be more eVective than United States Congress established the Na- the first, and that the third—providing auto- tional Highway TraYc Safety Administration’s 1 Highway Safety North, matic protection—will be the most eVective”. predecessor, appointed William Haddon as its 110 Homestead Road, The favored strategies implicitly assume that first administrator, and directed it to improve Ithaca, NY 14850–6216, people will not react to safety laws or safer motor vehicle safety. By the end of 1968 the USA products in ways that would reduce or agency had issued 29 Federal Motor Vehicle eliminate their eVect. But of course they may. Safety Standards (FMVSS) applying to new Correspondence to: Dr Hedlund (e-mail: This behavioral reaction in response to safety motor vehicles. Some FMVSS sought to [email protected]) measures has been called many things. In this prevent crashes by setting standards for brakes, Haddon Memorial Lecture 83

tires, and mirrors. The majority sought to With his 1982 paper, “The theory Risk Analysis Inj Prev: first published as 10.1136/ip.6.2.82 on 1 June 2000. Downloaded from reduce crash injuries by requiring features such of risk homeostasis: implications for safety and as seat belts, shatterproof windshields, and health”,22 its four accompanying comment- energy-absorbing steering columns. aries,23–26 and Wilde’s response,27 his ideas Peltzman evaluated the eVects of the FMVSS: attracted substantial attention. had they in fact improved safety as anticipated? In Wilde’s view, risk is an inherent part of He began with the assumption that we are our psychological makeup. Not only can we not rational economic consumers who act in our avoid risk, we need risk. Wilde hypothesizes own best interests. If we have more of something that we each have a “target level of risk” and than we want, we will (if we can) exchange it for measure risk on our own “risk thermostat”. If something else that we desire. Peltzman consid- the perceived risk of a situation exceeds our ered safety (or risk) an economic good that we target level, we will act to reduce it. And if the will trade with other goods in the same way. If perceived risk is lower than our target level, we our car is safer than we want it to be, then we will will attempt to increase our risk back to our drive faster, trading safety for time. In his words, target level through more dangerous actions. we will trade safety for “driving intensity”. Wilde’s name for this process is risk Peltzman tested his idea with an econometric homeostasis, by analogy with the self- analysis of FMVSS eVects. He concluded that regulatory and unconscious manner in which the standards were ineVective: they had no we maintain our body’s temperature. Risk eVect on overall traYc fatalities; they may have homeostasis is then an extreme form of behav- saved some auto occupants’ lives while increas- ioral adaptation: not only do we modify our ing pedestrian deaths.3 behavior in response to external changes This conclusion startled the road safety designed to make us more or less safe, but we community and challenged the role of govern- seek to counteract these changes completely ment in attempting to improve safety through and return to our desired risk level. Because the regulation. In Peltzman’s view, government risk in our environment constantly changes, we regulation was useless and perhaps even coun- constantly are forced away from our target risk terproductive. level, but we always move back toward it, Peltzman’s paper prompted a lively debate always countering external influences (such as over the next 20 years. Some papers criticized or injury prevention measures) that attempt to supported Peltzman’s study; others oVered new decrease or increase our risk. analyses.4–19 The debate centered on Peltzman’s Wilde developed a mathematical formula- statistical analyses. Issues included the variable tion of risk homeostasis in road safety: the used to measure FMVSS eVects, the time period “accident rate per time unity of driver exposure http://injuryprevention.bmj.com/ analyzed, the regression equation’s functional is invariant regardless of road geometry” (or, form, what other factors should be included in for that matter, regardless of anything else).22 In the model as controls and how they are best other words, my accident rate per hour of travel measured, and how to account for motorcycles, on a high-speed divided motorway is the same trucks, and other vehicles not regulated in the as my rate per hour on a low-speed neighbor- same way as passenger vehicles. Each choice hood . Wilde described his theory in may aVect the results substantially. For example, catchy language: “the sum of the sins is Graham and Garber recalculated Peltzman’s constant”. Others called it even more pictur- regression estimates using absolute instead of esquely “the law of conservation of misery”. logarithmic variables. Their estimates suggested Wilde extended risk homeostasis beyond road that regulation prevented roughly 5000 fatalities safety: “Risk homeostasis may thus apply not

between 1966 and 1972, rather than causing only to road use, but also to industrial safety, on September 28, 2021 by guest. Protected copyright. about 10 000 deaths as Peltzman concluded.15 sports, making love, smoking, drinking, doing Blomquist reviewed 11 of these studies.20 His home repairs, climbing ladders, physical exer- summary of their aggregate evidence is that the cise, investing money, gambling, and who FMVSS increased safety for passenger car knows how many other activities”.28 occupants, but probably not as much as had Risk homeostasis challenges the foundations been predicted, and reduced safety for non- of injury prevention strategy. It holds that the occupants, but not enough to oVset the benefits only eVective safety measures are those that to occupants. In a recent critical review, Levy alter my desired risk level. Anything that merely and Miller agree that the FMVSS improved modifies the environment or that regulates my occupant safety but find only weak support for behavior without aVecting my target risk level any eVect on non-occupants.21 is useless. This debate did not aVect the FMVSS: they Wilde’s risk homeostasis theory was chal- remained in force, without serious challenge lenged by Leonard Evans, Frank McKenna, and (indeed, manufacturers had adopted many of others. Wilde published extensively over the fol- the FMVSS voluntarily before they were lowing 10 years, engaging his critics in running required by regulation). But Peltzman’s paper debates.29–42 Wilde’s 1994 book Target Risk sum- introduced risk compensation as a serious road marizes his views for a general audience.28 safety issue. RISK COMPENSATION AND PUBLIC POLICY: ADAMS’ RISK COMPENSATION AND PSYCHOLOGY: WILDE’S CAMPAIGN AGAINST USE LAWS RISK HOMEOSTASIS THEORY John Adams (University College, London) GJSWilde (Queen’s University, Ontario) began investigating the eVects of seat belt use considered risk compensation from a psycho- laws in his work on transportation planning, in logical rather than economic point of view. which he opposed policies that increased the 84 Hedlund

number of private cars. He concluded that seat Evidence for and against risk Inj Prev: first published as 10.1136/ip.6.2.82 on 1 June 2000. Downloaded from belt laws were not eVective. In fact, as Peltzman compensation concluded for the FMVSS, Adams believed that The evidence falls into two broad categories: belt laws (and other vehicle safety measures) evaluations and experiments. reduced risk for passenger car occupants but increased risk for pedestrians and cyclists. He adopted much of Wilde’s risk homeostasis as the EVALUATIONS Risk compensation occurs if people react to a behavioral basis for his findings. safety law or regulation by acting less safely. It Adams’ goal was to influence public policy. can be evaluated either by examining individu- He opposed seat belt laws while they were als to observe if their actions have changed or being debated in the British Parliament. His by examining aggregate data to measure the primary method for estimating belt law eVects law’s or regulation’s eVects. was to compare overall road fatality trends in countries with and without belt laws. His results were easily understood by newspaper (1) Individual actions readers and politicians alike. He also stated his Was explicit compensating behavior observed? thoughts clearly, succinctly, and controver- This apparently logical way to evaluate risk sially: “protecting motorists from the conse- compensation is virtually impossible to carry quences of bad driving encourages bad out satisfactorily in practice, for two reasons. driving”.43 His 1995 book Risk/John Adams First, risk compensation predicts that behavior summarizes and extends his ideas in a broad will change but does not predict how it will discussion of risk in society.44 change, so we don’t know what to observe. Adams’ views and analyses on belt law eVec- Behavior may change in ways that are not at all tiveness were countered forcefully by many, obvious. Wilde suggests that measures to notably Murray Mackay. Critics argued that reduce drunk driving may in fact have a road Adams’ methods were suspect. In particular, safety benefit but also may cause those who overall road fatality trends are aVected by many would have driven drunk to act in more risky factors and a detailed statistical analysis with ways when not on the road.28 Or, compensating good data and appropriate controls is needed behavior may take place well after the fact. to evaluate belt law eVects. Adams suggests that laws requiring traYcto After extensive debate, Parliament adopted a stop when children are entering or leaving a belt use law for front seat occupants eVective in school bus may encourage children to be care- January 1983 which would lapse after three less, so that in later years they are in danger years unless Parliament voted again to continue when entering or leaving a transit bus.56 No http://injuryprevention.bmj.com/ it. Evaluations of data from the first post-law study can examine all possible ways in which year by Mackay,45 the British Department of compensating behavior might occur. Second, Transport,46 and statisticians from the London behavior change is diYcult to measure. We may School of Economics46 all concluded that be able to measure large changes such as passenger car front seat occupant casualties performing a task more quickly but usually decreased substantially. Adams countered by cannot measure more subtle changes such as claiming that the observed decrease was due to a increased carelessness. well-established downward trend in casualties A few studies have examined driving behav- and to a campaign to reduce drunk driving.43 44 ior changes after various road safety measures. In 1986, Parliament voted to retain the seat They typically find no eVects for measures to belt use law. Irwin provides a thoughtful view of protect occupants in the event of a crash (such

scientific and political issues throughout the as seat belts) but may find eVects for measures on September 28, 2021 by guest. Protected copyright. British seat belt law debate.47 Adams48 and that attempt to prevent crashes by improving Mackay45 give contemporary views from both vehicle performance (such as better brakes or sides. Evans49 and Adams44 analyze the issues tires). For example, O’Neill et al studied drivers and evaluate the results from a longer perspec- in Canada and England after seat belt use laws tive. were implemented.57 They examined travel speeds and following headways and reported RISK COMPENSATION IN 2000 no evidence of riskier behavior due to the belt Peltzman’s, Wilde’s, and Adams’ papers use laws. Sagberg et al observed travel speeds prompted debate in journal, at conferences, and following headways for Oslo taxi drivers and in books.28 44 49–54 In terms of ideas and with and without airbags and antilock brakes.58 controversy, though, little has changed in the They reported shorter headways for cars with past decade. An Organization for Economic antilock brakes but no significant diVerence for Co-operation and Development (OECD) re- cars with airbags. Wilde reports on a study of port, prepared by an international working Munich taxicabs with and without antilock group, gives an excellent, thorough, and brakes.28 The study found that drivers with dispassionate summary of risk compensation antilock brakes changed their behavior by driv- theories and evidence. The report concludes ing faster and braking harder than before. that “ ... behavioural adaptation to road safety Most important, even if observations show programmes does occur although not consist- that behavior has changed, the eVect on crashes ently. ... behavioural adaptation generally does or injuries is unclear. Faster driving may not not eliminate the safety gains from pro- necessarily lead to increased crash risk; grammes, but tends to reduce the size of the diVerent methods of sawing wood may not lead expected eVects”.55 The extreme views of risk to increased injuries. Risk compensation is homeostasis have attracted little support. relevant only if a safety measure produces Haddon Memorial Lecture 85

behavior change that in turn increases risk. The evaluations dis- Vehicle safety standards— Inj Prev: first published as 10.1136/ip.6.2.82 on 1 June 2000. Downloaded from This must be evaluated with aggregate data. cussed previously typically examine data on all road fatalities, thus including all reasonable (2) Aggregate data system eVects. Several evaluations also com- Were injuries reduced as intended? The pare results with predictions. Blomquist’s sum- question requires all the standards of a good mary suggests that the FMVSS did make evaluation: a sound experimental design, good occupants safer, but less than expected; that data, good controls for other factors, and they may have increased risk for non- appropriate statistical analyses. Even well done occupants; and that they increased overall road safety measure evaluations may fall short. safety: in short, they may have produced some While overall injury counts are high, individual risk compensation, but their overall eVect was injuries typically are rare events. Data on which positive.20 Levy and Miller’s review questions to base an evaluation frequently are inaccurate the increased risk for non-occupants.21 or imprecise. EVects may be small: most traYc Seat belt use laws—The British seat belt law safety measures do well to reduce casualties by studies discussed previously consider system 10%. Safety measures seldom are implemented eVects on all road users. The results also are in controlled experimental conditions but are consistent with the well-established protective put in place in the real world, where many benefits of belts in a crash and with the observed other changing factors can aVect the results. increases in belt use resulting from the law.49 As The literature contains tens of thousands of noted above, Adams challenged these findings.44 studies evaluating safety measures. For example, Evans compared United States seat belt law a literature search produced 54 078 abstracts or results with predictions.49 He found observed titles that might be relevant to nine motor fatality reductions close to, but typically less vehicle injury prevention strategies.59 But good than, predicted reductions. He attributes this studies are rare. For instance, a meta-analysis of not to risk compensation but to two other drunk driving prevention and control literature factors: that belted drivers are generally safer from 1960 through 1991 identified 6500 docu- drivers than unbelted, and that belt use rates are ments, of which only 125 passed minimal stand- lower at night than during the day (when belt ards of scientific rigor and quality.60 use observations typically are taken). To test for risk compensation a high quality Many other belt use law evaluations have study should do two additional things. First, it been published with far less conclusive results. should compare the eVect with what was Most do not consider eVects on other road predicted in the absence of any compensating users. Levy and Miller reviewed a few studies behavior change. Predictions are of course that do examine possible risk compensation.21 http://injuryprevention.bmj.com/ imprecise. However, results falling far short of They conclude that belt use laws generally predictions may suggest risk compensation. improved vehicle occupant safety but that sev- Second, the study should examine system eral studies reported evidence on non- eVects. Safety measure evaluations frequently occupant casualties consistent with compen- fail to look beyond the population directly sating behavior. aVected by the measure: for example, seat belt Motorcycle helmet use laws—These state laws evaluations often consider consequences only in the United States provide an excellent test of to vehicle occupants. risk compensation for several reasons. Helmets Evans provides a good sampling of actual clearly reduce head injury in a crash, and head and predicted eVects.49 61 He examined 24 injury is the leading cause of death for motor- studies from the road safety literature and cyclists. Motorcyclists are very aware of the

compared the eVects predicted and actually protection oVered by a helmet, but some on September 28, 2021 by guest. Protected copyright. realized. For measures designed to increase motorcyclists dislike helmets for various safety he found examples where safety in- reasons. The potential risk compensation creased even more than expected, about as mechanism is obvious: a motorcyclist who expected, less than expected, where the meas- would not choose to wear a helmet may drive ure did not change safety at all, and where the more recklessly if helmet use is required by law. measure actually decreased safety—a perverse Helmet use depends strongly on the presence eVect. Similarly, for measures expected to of a law: about half of all motorcyclists wear decrease safety, he found the same range of helmets if there is no law but almost all wear eVects, from a decrease greater than expected helmets if there is a law. States have enacted, to an actual increase in safety—an equally per- repealed, and re-enacted helmet laws many verse eVect. Evans concluded that behavioral times over the past 35 years, providing many adaptation to traYc safety measures is wide- opportunities to measure law eVects. Finally, spread, that the eVects can vary widely, and system eVects from helmet use laws are that there is no evidence for the complete com- negligible, since motorcycle crashes very rarely pensation predicted by risk homeostasis. cause serious injury to anyone other than Many of Evans’ measures are not laws or motorcyclists themselves. regulations, so his conclusions on behavioral At the request of Congress, in 1991 the adaptation cannot be applied immediately to United States General Accounting OYce re- risk compensation. His studies also do not viewed all 46 available helmet law eVectiveness report on system eVects. To examine both studies.62 It concluded that helmet laws reduced issues, consider three road safety areas where motorcyclist fatalities by 20% to 40%, a result risk compensation issues have been raised in reasonable agreement with helmet eVective- frequently: vehicle safety standards, seat belt ness in a crash. Evans reaches similar laws, and motorcycle helmet laws. conclusions,49 as do more recent evaluations.63–65 86 Hedlund

From this evidence, motorcycle helmet laws and carelessly. The equipment changed the Inj Prev: first published as 10.1136/ip.6.2.82 on 1 June 2000. Downloaded from have produced no detectable risk compensa- injury distribution, reducing injuries to areas tion. directly protected by the equipment but Risk compensation in other settings—Several increasing other injuries. Overall injury rates recent studies examine risk compensation in decreased somewhat. response to both aggregate and specific con- sumer product and workplace safety regula- (3) Summary tions. The following examples give a flavor of This brief review of the evaluation evidence the results. Each study discusses relevant strongly suggests that various amounts of risk earlier research. compensation have occurred in response to Consumer product regulations: Since its some safety measures but not in response to establishment in 1972, the United States Con- others. The review also illustrates the diYculty sumer Product Safety Commission (CPSC) of accurately establishing or refuting, much less has issued many regulations aVecting diVerent measuring, risk compensation. Risk compensa- products. Early studies typically concluded that tion proponents acknowledge this dilemma: “ CPSC’s regulations did not reduce consumer ... the multi-dimensionality of risk and all the accidents.66 Recent studies of some specific problems of measuring it discussed earlier, standards have found safety benefits. preclude the possibility of devising any conclu- Bicycles: Magat and Moore concluded that sive statistical tests of the [risk compensation] bicycle safety standards in the United King- hypothesis”.44 dom and the United States have reduced bicy- 67 cle accidents. EXPERIMENTS Child-resistant medicine bottle caps: Rodg- Controlled experiments in laboratory settings ers found “persuasive and robust evidence of eliminate much of the messiness and variability the eVectiveness of child-resistant packaging of real world evaluation. In these studies, for oral prescription drugs” in the United subjects typically perform some task for a States.68 In Sweden, Assargård and Sjöberg reward that depends on their performance. concluded that the caps have been highly eVec- They also face penalties for an “accident”. The tive in reducing accidental poisoning from liq- experimenter varies the reward and the acci- uid paracetamol among children.69 dent risk and observes changes in the subject’s Cigarette lighters: Viscusi and Cavallo con- performance. Two examples illustrate these cluded that lighters with child-resistant fea- experiments. For a summary of many experi- tures cause some consumers to reduce the care mental studies on risk compensation, see they take with lighters but that their overall Glendon et al.76 http://injuryprevention.bmj.com/ eVect has been to improve safety.70 Jackson and Blackman report on a study in Power lawn mowers: Beginning in 1982, all which subjects “drove” a driving simulator walk-behind mowers sold in the United States through a specified route.77 Subjects received a were required to have a safety shield and a monetary reward for completing the route “deadman” clutch that stops the motor when quickly. They were penalized for “accidents” or the handle is released. Alexander concluded for being caught speeding. The and that the regulations increased injuries per the penalties for speeding and accidents were population.71 Moore and Magat, on the other varied. The authors found that, “consistent hand, found evidence suggesting that the regu- with risk homeostasis theory, increased speed lations reduced injuries.72 limit and reduced speeding fine significantly Workplace safety regulations: Beginning in increased driving speed but had no eVect on

the early 1970s, the United States Occupa- accident frequency. Moreover, increased acci- on September 28, 2021 by guest. Protected copyright. tional Safety and Health Administration dent cost caused large and significant reduc- (OSHA) has set numerous safety standards for tions in accident frequency but no change in workplace equipment design (such as width speed choice”. and spacing requirements for handrails) and Wilde reports on several experiments in Tar- performance (strength requirements for ladder get Risk.28 In a typical experiment, a subject is rungs) as well as worker conduct (speed limits seated at a computer screen in which a large while driving a forklift). OSHA enforces these square appears at random times. The subject’s standards through inspections and penalties for objective is to press a button as close to but no violations. Early analyses found little or no sooner than 1.5 seconds after the square eVect on accidents.73 74 Lanoie’s review of more appears. The subject’s reward increases the recent research concluded that the standards closer the response is to 1.5 seconds. Re- have reduced injuries of certain types and also sponses sooner than 1.5 seconds receive no reduced overall injuries in firms that were reward and may impose a penalty. In this and inspected.74 Using data from 1973–83, Viscusi many similar experiments, Wilde finds evi- found a “modest” eVect on injuries resulting in dence for risk compensation: as the penalty for lost workdays but not on overall injuries.73 responding too quickly increases, response Protective equipment for loggers: Klen stud- times also increase so that the number of ied both individual behavior and overall responses drawing penalties decreases. accident rates after protective equipment (such These experiments show clearly that people as helmets, eye protectors, safety gloves and modify their behavior in response to changes in boots) was required or used voluntarily by the reward and penalty structure of their envi- Finnish loggers.75 He concluded that most log- ronment. This is hardly news: behavioral gers felt safer with this equipment and nearly change in response to reward and risk changes half reported that they worked more quickly has been observed in decisions to invest, to buy Haddon Memorial Lecture 87

insurance, and in many other ways. But this is only way I learn about them is from manufac- Inj Prev: first published as 10.1136/ip.6.2.82 on 1 June 2000. Downloaded from far from relevant to injury prevention. Labora- turer or media information. Many features to tory experiments carry no risk of injury or reduce or prevent injuries to vehicle occupants, death. Performance in the laboratory likely has such as side door beams or penetration- little or no relation to risk compensation. Risk resistant windshields, are invisible for all prac- compensation proponents recognize this: “Re- tical purposes. sorting to laboratory and simulation studies Laws and regulations restricting my behavior may be methodologically pleasing (and morally can be very obvious, if advertised and enforced innocuous), but it is doubtful that the theory in vigorously. If not, they too may be invisible. question [risk compensation] can ever be Rule 1: If I don’t know it’s there, I won’t cogently tested under such contrived condi- compensate for a safety measure. tions. ... In other words: simulation of risk, like a sham duplicating the real thing, is a (2) EFFECT contradiction in terms”.27 How does the change aVect me, both physically and mentally? THEORY This factor has several dimensions. First, If experiments cannot provide useful evidence, how does the change aVect my physical and if evaluations are contaminated by poor performance of the task, through direct sensory data and uncontrolled factors, we are left with feedback or otherwise? Is it annoying, like theory. Are we economic beings who constantly child-resistant medicine caps that too often are balance costs and benefits in deciding whether adult-resistant as well? Is it physically uncom- to speed up a bit on this road (knowing that our fortable, as helmets are for some motorcyclists? airbag will help protect us if worst comes to Does it make the task easier, like improved worst)? Are we constantly motivated by our vehicle handling or brakes? Or more diYcult, desire for risk? Both common sense and the like the lawnmower deadman switch that evidence reviewed above suggest that these requires me to hold the handle constantly? factors influence our actions but do not deter- Second, how does the change aVect my atti- mine them absolutely. How much will they tude? Does it annoy me, like a requirement to influence our response to a safety measure? For wear seat belts may for a libertarian? Or do I this we must be able to predict our responses welcome it, like a guardrail added to a danger- and measure the results, which leads right back ous curve? These two dimensions clearly inter- to evaluation. act, as changes aVecting my performance also may aVect my attitude. Finally, how does the change aVect my http://injuryprevention.bmj.com/ A personal view of risk compensation perception of risk? Do I feel safer because I am Behavioral adaptation and risk compensation wearing a ? Do I feel that risk has clearly occur in some situations. We react to been eliminated, as the Titanic’s passengers and changed conditions; we are famous for not crew may have believed? Or do I think the always doing as we are told or as is expected of change has little or no eVect on my risk, us. On the other hand, I believe the evidence is because I felt there was no risk in the first place, overwhelming that every safety law or regula- because I believe the measure is ineVective, or tion is not counterbalanced by compensating because I don’t know that anything has behavior. changed? Thus the issue becomes not yes or no, but Rule 2: If it doesn’t aVect me, I won’t when and how much. When may compensa- compensate for a safety measure. tion occur in response to a safety measure? on September 28, 2021 by guest. Protected copyright. How likely is it to occur? What are its possible (3) MOTIVATION consequences, both direct and indirect? What influences my behavior? What is my I suggest that four factors influence adapta- motivation in doing the task? What is my eco- tion and compensation in response to safety nomic utility function? What are my psycho- measures. Each factor has several aspects, and logical needs? the factors interact with each other. Each has This factor is key to most risk compensation been suggested previously, for example by theory discussions. Economists hold that I am OECD.55 The four factors lead to overall guid- influenced by economic goals. If I am driving, ance and to principles for action. they believe my only goal is to be transported in the shortest time and to avoid the economic (1) VISIBILITY costs of crashes and injuries. So if my car How obvious is the change produced by the becomes safer, either because it is less likely to safety measure? Do I even know there has been crash or less likely to injure me if it does, and if a change? the additional safety is not useful to me, then I Some changes are very obvious, especially will drive faster. The same reasoning applies if those that aVect performance through direct I am required to use safety equipment such as feedback: vehicle brakes and studded tires, seat belts or cycle helmets. In the workplace, if child-resistant caps on medicine bottles, pro- my salary depends on my output, my goal is to tective equipment for athletes or workers. maximize my production while keeping my Other changes are apparent but easily over- injury risk below an acceptable level. looked. I “know” there’s a smoke detector out- Risk homeostasis theorists, on the other side my study, but I think about it only when I hand, hold that my basic goal is to maintain my replace its battery. Finally, some changes may desired risk level: “It is primarily risk to self be completely or psychologically invisible. The that governs behaviour on the road”.44 88 Hedlund

Both views are simplistic. I am motivated by As all factors reach moderate levels, the like- x Inj Prev: first published as 10.1136/ip.6.2.82 on 1 June 2000. Downloaded from many factors, both economic and behavioral. lihood of compensation increases. On the road I want to get from here to there x If each is high, behavioral compensation is while avoiding both personal injury and likely: the measure is highly visible, aVects crashes. I may or may not care about saving me substantially, I have good reasons to time: while late for an appointment, I may change behavior and I have the freedom to cherish every second; on a casual trip, I may do so. Then I will consume at least some of decide to take a longer and slower route my increased safety as performance. And because I enjoy the scenery. even then it’s unclear whether the overall I also am motivated by habit and by my eVect will be to compensate partially, desire to simplify decisions. While I may make completely, or more than completely for the more or less rational decisions in an unfamiliar safety measure. situation, I quickly fall into habits and put many daily operations on “automatic pilot”. So I don’t think each time about how fast I drive Implications for injury prevention down my neighborhood street, I do it just as I Risk compensation is important for everyone always have done. Once accustomed to wearing who plans and implements injury prevention a seat belt in a car or a hard hat on a construc- measures. In summary: tion site, many of us do it every time without x Risk compensation can occur —people are not thinking. machines. We all change our behavior in These factors all influence my motivation to response to changes in our environment. compensate for safety changes. If I am Safety measures change our environment, so motivated to change behavior, I may well com- we may change our behavior in response to pensate. But if there is no motivation for them. Many rational and behavioral factors behavior change, I won’t. influence whether and how our behavior will Rule 3: If I have no reason to change my change. Never assume that behavior will not behavior, I won’t compensate for a safety change. measure. x Four factors influence risk compensation— visibility, eVect, motivation, and control. Risk (4) CONTROL compensation occurs only in certain circum- How much do I control the situation? Can I stances. The four factors and the compensa- change my actions even if I want to? tion index help analyze an injury prevention Workplace situations frequently are tightly

measure to estimate whether risk compensa- http://injuryprevention.bmj.com/ controlled by rules, supervisors, and the physi- tion is likely or not. cal environment, so may allow little oppor- x To reduce or eliminate risk compensation, use tunity for behavioral change. Piecework set- measures rating low on at least one factor. Pre- tings provide more flexibility and freedom. fer measures that are invisible to people, or Driving allows considerable freedom: traYc that do not aVect their actions or attitudes, laws provide nominal control, but since most or for which they have no motivation or free- laws are not enforced rigorously, individual dom to change behavior. drivers have considerable latitude for their x Consider system eVects. Injury prevention actions. Household settings allow virtually measures may have eVects beyond the complete control. individual actions they influence directly. Sport provides interesting examples of the These eVects may be harmful or helpful. interplay between injury prevention, compen- Always consider potential system eVects.

sation, and control. In many sports, such as ice on September 28, 2021 by guest. Protected copyright. Many injury preven- hockey and American football, players are x Don’t over-predict benefits. tion measures promise more benefits than required to wear protective equipment. Some they deliver, due to bad science, political players have compensated by acting more violently within the confines of the rules. In pressures, or failure to consider risk compen- some instances this has led to rules changes to sation or system eVects. While calm and real- control player actions more tightly.78 istic benefit estimates are diYcult to produce, unduly optimistic predictions will hamper Rule 4: If my behavior is tightly controlled, I won’t compensate for a safety measure. injury prevention eVorts in the long run. x Trading safety for performance isn’t necessarily Safety isn’t society’s only goal. All A COMPENSATION INDEX bad. Each of these four factors—visibility, eVect, action produces risk. As society and as indi- motivation, and control—is far more complex viduals we constantly balance performance than this simple discussion suggests. But and risk (in many dimensions of each). If together they provide a useful framework for some safety benefits predicted for an injury considering potential risk compensation in prevention measure become performance response to a safety measure: a highly imprecise improvements instead, society overall may compensation index. Assess each factor subjec- benefit. tively, from “no, not at all, zero” to “maybe, moderate, some” to “yes, strong, a lot”. As a This work was supported by the Insurance Institute for Highway first approximation: Safety. Its opinions, findings, and conclusions are those of the author and do not necessarily reflect the views of the Insurance x Compensation is unlikely if any of the four is Institute for Highway Safety. This paper is based on the Haddon zero: the measure is invisible, doesn’t aVect Memorial Lecture presented to the Fifth World Conference on Injury Prevention and Control, New Delhi, in March 2000 and me, or I have no motivation or no freedom to published in Injury Prevention and Control (D Mohan, G Tiwari, change my behavior. eds; London: Taylor & Francis) with permission. Haddon Memorial Lecture 89

1 Committee on Trauma Research, Commission on Life Sci- 41 Wilde GJS. Risk homeostasis theory: an overview. Inj Prev ences, National Research Council, and the Institute of 1998;4:89–91. Inj Prev: first published as 10.1136/ip.6.2.82 on 1 June 2000. Downloaded from Medicine. Injury in America: a continuing public health prob- 42 O’Neill B, Williams A. Risk homeostasis hypothesis: a lem. Washington, DC: National Academy Press, 1985. rebuttal. Inj Prev 1998;4:92–3. 2 Smeed R. Some statistical aspects of road safety research. 43 Adams JGU. Risk and freedom: the record of road safety regula- Journal of the Royal Statistical Society Series A 1949;112:1– tion. CardiV, Wales: Transport Publishing Projects, 1985. 34. 44 Adams JGU. Risk/John Adams. London: UCL Press, 1995. 3 Peltzman S. The eVects of automobile safety regulation. 45 Mackay M. Seat belt use under voluntary and mandatory Journal of Political Economy 1975;83:677–725. conditions and its eVect on casualties. In: L Evans, RC 4 Joksch HC. Critique of Sam Peltzman’s study: the eVects of Schwing, eds. Human behavior and traYc safety. New York, automobile safety regulation. Accid Anal Prev 1976;8:129– NY: Plenum, 1985: 259–78. 37. 46 Department of Transport. Compulsory seat belt wearing. 5 ]Peltzman S. The eVects of automobile safety regulation: a London: HMSO, 1985. reply. Accid Anal Prev 1976;8:139–42. 47 Irwin A. Technical expertise and risk conflict—an institu- 6 Joksch HC. The eVects of automobile safety regulation: tional study of the British compulsory seat-belt debate. comments on Peltzman’s reply. Accid Anal Prev 1976;8: Policy Sciences 1987;20:339–64. 213–14. 48 Adams JGU. Smeed’s Law, seat belts and the emperor’s new 7 Robertson LS. A critical analysis of Peltzman’s “The eVects clothes. In: Evans L, Schwing RC, eds. Human behavior and of automobile safety regulation”. Journal of Economic Issues traYc safety. New York, NY: Plenum, 1985: 193–248. 1977;11:586–600. 49 Evans L. TraYc safety and the driver. New York, NY: Van 8 Peltzman S. A reply to Robertson. Journal of Economic Issues Nostrand Reinhold, 1991. 1977;11:672–8. 50 Evans L, Schwing RC, eds. Human behavior and traYc safety. 9 Robertson LS. Rejoinder to Peltzman. Journal of Economic New York, NY: Plenum, 1985. Issues 1977;11:679–83. 51 Brown I, Janssen W, eds. Risky decision-making in transport 10 Robertson LS. State and federal new-car safety regulation: operations. Ergonomics 1988;31:405–669. eVects on fatality rates. Accid Anal Prev 1977;9:151–6. 11 Robertson LS. Automobile safety regulations and death 52 Trimpop RM, Wilde GJS, eds. Challenges to accident . Groningen, reductions in the United States. Am J Public Health prevention: the issue of risk compensation behavior 1981;71:818–22. Netherlands: Styx Publications, 1994. 12 Robertson LS. Automobile safety regulation: rebuttal and 53 Stanton N, Glendon I, eds. Risk homeostasis and risk new data. Am J Public Health 1984;74:1390–4. assessment. Safety Science 1996;22:1–262. 13 Robertson LS. Reducing death on the road: the eVects of 54 Viscusi WK, ed. Product safety and managerial decisions. minimum safety standards, publicized crash tests, seat Managerial and Decision Economics 1996;17:437–530. belts, and alcohol. Am J Public Health 1996;86:31–4. 55 Organization for Economic Co-operation and Develop- 14 Crandall RW, Graham JD. Automobile safety regulation and ment. Behavioural adaptations to changes in the road transport oVsetting behavior: some new empirical estimates. Ameri- system. Paris: OECD, 1990. can Economic Review 1984;74:328–31. 56 Adams JGU. Risk homeostasis and the purpose of safety 15 Graham JD, Garber S. Evaluating the eVects of automobile regulation. Ergonomics 1988;31:407–28. safety regulation. Journal of Policy Analysis and Management 57 O’Neill B, Lund AK, Zador P, et al. Mandatory belt use and 1984;3:206–24. driver risk taking: an empirical evaluation of the risk com- 16 Orr LD. The eVectiveness of automobile safety regulation: pensation hypothesis. In: Evans L, Schwing RC, eds. evidence from the FARS data. Am J Public Health 1984;74: Human behavior and traYc safety. New York, NY: Plenum, 1384–9. 1984: 93–107. 17 Zlatoper TJ. Regression analysis of time series data on 58 Sagberg R, Fosser S, Saetermo I-AF. An investigation of motor vehicle deaths in the United States. Journal of Trans- behavioural adaptation to airbags and antilock brakes port Economics and Policy 1984;18:263–74. among taxi drivers. Accid Anal Prev 1997;29:293–302. 18 Garbarcz C. How eVective is automobile safety regulation? 59 Rivara FP, Thompson DC, Beahler C, et al. Systematic Applied Economics 1990;22:1705–14. reviews of strategies to prevent motor vehicle injuries. Am J 19 Chirinko RS, Harper EP Jr. Buckle up or slow down? New Prev Med 1999;16:1–5. estimates of oVsetting behavior and their implications for 60 Wagenaar AC. Importance of systematic reviews and meta- http://injuryprevention.bmj.com/ automobile safety regulation. Journal of Policy Analysis and analyses for research and practice. Am J Prev Med 1999;16: Management 1993;12:270–96. 9–11. 20 Blomquist GC. The regulation of motor vehicle and traYc 61 Evans L. Human behavior feedback and traYc safety. safety. Boston, MA: Klewer Academic Publishers, 1988. Human Factors 1985;27:555–76. 21 Levy DT, Miller T.Review: risk compensation literature, the 62 General Accounting OYce. Highway safety: motorcycle helmet theory and evidence. Journal of Crash Prevention and Injury laws save lives and reduce costs to society. Washington, DC: Control 1999 (in press). US General Accounting OYce, 1991. 22 Wilde GJS. The theory of risk homeostasis: implication for 63 Muelleman RL, Mlinek EJ, Collicot PE. Motorcycle crash safety and health. Risk Analysis 1982;2:209–25. injuries and costs: eVect of a re-enacted comprehensive 23 Slovic P, FischhoV B. Targeting . Risk Analysis 1982; helmet use law. Ann Emerg Med 1991;21:266–72. 2:227–34. 64 Fleming HS, Becker ER. The impact of the Texas 1989 24 Graham JD. On Wilde’s theory of risk homeostasis. Risk motorcycle helmet law on total and head-related fatalities, Analysis 1982;2:235–7. severe injuries, and overall injuries. Medical Care 1992;30: 25 Orr LD. Goals, risks, and choices. Risk Analysis 1982;2:239– 832–45. 42. 65 Kraus JF, Peek C, McArthur DL, et al. The eVect of the 26 Cole GA, Withey SB. The risk of aggregation. Risk Analysis 1992 California motorcycle helmet usage law on motorcy- 1982;2:243–7. cle crash fatalities and injuries. JAMA 1994;272:1506–11. 27 Wilde GJS. Critical issues in risk homeostasis theory. Risk 66 Viscusi WK. Consumer behavior and the safety eVects of Analysis 1982;2:249–58. product safety regulation. Journal of Law and Economics on September 28, 2021 by guest. Protected copyright. 28 Wilde GJS. Target risk: dealing with the danger of death, disease 1985;28:527–54. and damage in everyday decisions. Toronto: Castor and 67 Magat WA, Moore MJ. Consumer product safety regulation Columba, 1994. in the United States and the United Kingdom: the case of 29 McKenna FP. The human factor in driving accidents: an bicycles. Rand Journal of Economics 1996;27:148–64. overview of approaches and problems. Ergonomics 1982;25: 68 Rodgers GB. The safety eVects of child-resistant packaging 867–77. for oral prescription drugs. JAMA 1996;275:1661–5. 30 Wilde GJS. Evidence refuting the theory of risk homeosta- 69 Assargård U, Sjöberg G. The successful introduction of sis? A rejoinder to Frank P McKenna. Ergonomics 1984;27: child resistant closures for liquid paracetamol preparations. 297–304. Safety Science 1995;21:87–91. 31 McKenna FP. Do safety measures really work? An examina- 70 Viscusi WK, Cavallo G. Safety behavior and consumer tion of risk homeostasis theory. Ergonomics 1985;28:489– responses to cigarette lighter safety mechanisms. Manage- 98. rial and Decision Economics 1996;17:441–57. 32 Wilde GJS. Assumptions necessary and unnecessary to risk 71 Alexander DL. An empirical investigation of lawn-mower homeostasis. Ergonomics 1985;28:1531–8. safety regulation. Applied Economics 1990;22:795–804. 33 McKenna FP. Evidence and assumptions relevant to risk 72 Moore MJ, Magat WA. Labeling and performance stand- homeostasis. Ergonomics 1985;28:1539–41. ards for product safety: the case of CPSC’s lawn mower 34 McKenna FP.Behavioural compensation and safety. Journal standards. Managerial and Decision Economics 1996;17: of Occupational Accidents 1987;9:107–21. 509–16. 35 McKenna FP. What role should the concept of risk play in 73 Viscusi WK. The impact of occupational safety and health theories of accident involvement? Ergonomics 1988;31:469– regulation, 1973–1983. Rand Jornal of Economics 1986;17: 84. 567–80. 36 Wilde GJS. Accident countermeasures and behavioral 74 Lanoie P. Government intervention in occupational safety— compensation—the position of risk homeostasis theory. lessons from the American and Canadian experience. Journal of Occupational Accidents 1989;10:267–92. Canadian Public Policy 1992;18:62–75. 37 McKenna FP. In defense of conventional safety measures: a 75 Klen T. Personal protectors and working behavior of reply to GJS Wilde. Journal of Occupational Accidents 1990; loggers. Safety Science 1997;25:89–103. 11:171–81. 76 Glendon AI, Hoyes TW, Haigney DE, et al. A review of risk 38 Evans L. Risk homeostasis theory and traYc accident data. homeostasis theory in simulated environments. Safety Risk Analysis 1986;6:81–94. Science 1996;22:15–25. 39 Wilde GJS. Notes on the interpretation of traYc accident 77 Jackson JSH, Blackman R. A driving-simulator test of data and of risk homeostasis theory: a reply to L Evans. Wilde’s risk homeostasis theory. Journal of Applied Psychol- Risk Analysis 1986;6:95–101. ogy 1994;79:950–8. 40 Evans L. Comments on Wilde’s notes on “Risk homeostasis 78 Tenner E. Why things bite back: technology and the revenge of theory and traYc accident data.” Risk Analysis 1986;6:103– . New York, NY: Random House, 7. 1996.