Those who have decided to swim against the current should not expect it to change direction. Stanislav J. Lec[1]

3. Toward a compact theory of risk taking

At the time, with the assistance of a doctoral student, I was preparing a literature review of psychological factors in accident causation for the Canadian federal government. This was an interesting exercise. We knew, of course, that "Traffic, like God, Football and Politics, belongs to that select group of subjects on which everyone, when the spirit seizes him, instinctively feels that he can speak with overriding authority and conviction".[2] What we didn't know, since we were novices in the scientific study of the field, was that the available literature was, and still is, extremely disconnected, fragmented and beset with a multitude of narrow views and pet solutions to the problem. Accidents have been associated with everything from mud flaps to poor eyesight, low barometric pressure, anti-social tendencies, narrow roads, driving too fast, driving too slow, alcohol use, abstinence, being young, being old, bad weather, good weather, being left-handed or in the process of getting a divorce.

Many of these factors are blessed with empirical evidence for their support, but that this blessing is rather mixed becomes apparent if one considers that the resources for the development of accident countermeasures and research are limited. If it is agreed that nothing is as practical as a good theory, one must regret the sparsity of comprehensive theories relative to the available body of findings. How can governments, social agencies and citizen groups decide where to focus their efforts in research and countermeasure development when so many divergent directions for action appear to present themselves? And what can you, as an individual, do to reduce your risk of accident? What is needed is a theory which compresses a lot of experience in a format that is concise enough to guide remedial action.

Having to write a global and coherent review forced us to seek the nature of the forest rather than the identity of the individual trees. Entering a new field while having to scan its full expanse can provide useful preventative medicine against developing that peculiar disease of experts--knowing progressively more and more about less and less, until one eventually knows everything about nothing.

So, we were looking for the big waves, not the minor ripples. The biggest wave, we discovered, was the one described below, and it has ever since been one of the strong influences in directing my thoughts on safety and lifestyle-dependent health.

3.1 Taylor's study

As we all know, anxiety increases perspiration. Fortunately for those of us who are concerned about social composure, most perspiration is not visible to the naked eye, but perspiration also increases the electrical conductivity of the skin, and even minor variations in perspiration alter this resistance and are "no sweat" to be picked up by sensitive equipment that is specially designed for its measurement.

Galvanic Skin Response (GSR) is the term psychologists and physiologists use to describe this phenomenon, which is named after its discoverer, Luigi Galvani. For anybody interested in the history of science, it is a truly galvanizing experience to see his sculpture at the entrance of the University of Bologna in Italy. This is the oldest university in the western world and was established in the 11th century. By the 13th century it had some 10,000 students.

The Galvanic Skin Response can be expressed as a percentage change in skin conductance and offers a quantitative measure of the degree of fear or risk--or other arousal--experienced by a person in reaction to some event. A driver approaching a traffic light and seeing it change from green to amber may show a small GSR, but the sudden discovery of another car approaching in the same traffic lane is likely to produce a major GSR. In Taylor's study, three important variables were measured for each of forty different road sections:

(a) The Accident Rate per Vehicle-Mile. This is calculated by dividing the record of accidents over the past two years (as documented by the police) by the number of passing vehicles, and then dividing this ratio by the length of the section measured in miles. This variable represents the spatial and objective accident risk per vehicle per mile of movement through specific road sections.

(b) The GSR Rate per Mile. Total GSR activity (the combination of the number and the size of responses) is divided by the length of the section. This gives a variable which represents the spatial and subjective accident risk per driver per mile of movement through specific road sections.

(c) Average Speed. The average moving speed, in each separate section, for all drivers in the study.

The degree of association between these three variables was expressed in terms of correlation coefficients. Whenever two variables are perfectly and positively related, the correlation coefficient equals +1; when the association is negative and perfect, the correlation coefficient equals -1, and it will be zero when there is no association at all. For example, the correlation between body height and body weight is typically in the order of r = + 0.7. Taylor found the following correlation coefficients (abbreviated as r):

So, what is the significance of these findings (apart from the fact that their likelihood of having occurred by chance was less than one in a thousand)? The positive association between Accident Rate per Vehicle-Mile and the GSR Rate per Mile can be interpreted to indicate that the drivers experienced more subjective risk in road sections that were also marked by high rates of accidents in the historical records. Apparently, then, drivers on average are sensitive to conditions in which many accidents happen, and react with increased fear.

And what did they do in these conditions? The observed negative correlation between the Accident Rate per Vehicle-Mile and Average Speed indicates that in these conditions they slowed down, while they moved faster in road sections with a low accident rate per vehicle-mile.

Finally, and most interestingly, the negative correlation between Average Speed and GSR Rate per Mile in the forty road sections indicate that the drivers kept the amount of risk they experienced relatively stable over time as they drove through the various road sections. In those road sections where they experienced a lot of risk, they slowed down and thus spent more time in these locations, thereby spreading the GSR activity out over a longer period of time. In contrast, where the GSR Rate per mile was low, they moved at higher speeds, so whatever GSR activity there was occurred in a shorter time frame. The end result was that GSR activity, or subjective risk experienced per time unit of travel, appeared to be relatively stable and independent of the particular road sections in which the driving was done, and thus independent of the accident rates per vehicle-mile of these road sections. To quote Taylor:

With subjective and objective risk thus defined [i.e., not per mile or km, but per time unit of exposure to risk], the conclusion from the present data is that they are both independent of what we normally mean by variations in road conditions. A possible reason why the subjective risk should be distributed thus may be found in the driver's ability to vary his performance. To some extent at least, he can voluntarily influence the risks taken (for example, by accepting or not accepting opportunities to overtake, or simply by going slowly or fast). Provided that he has this control, there would usually be no reason why he should wish to engage in more risk on one part of the road than on another, and in fact he could be said to be performing a self-paced task. When it is considered that the major restriction on his speed is due to other vehicles, the drivers of which may be expected to be behaving in much the same way, driving could be called a "collectively self-paced task".[3]

3.2 Some possible consequences

Allow yourself a moment to go beyond Taylor's article, as I did at the time, and imagine what far-reaching implications this interpretation might have. Engineering improvements of motor vehicles and the road environment can reduce the objective spatial risk, but what happens to objective temporal risk? If drivers adjust their behaviour in response to these improvements, it is no longer logical to presume that the accident risk per time unit of driving will also be reduced. In fact, if the total amount of time people spend on the roads is not affected, the accident rate per head of population would not change at all. And if the engineering improvements make driving more attractive so that people spend more time on the roads, the accident rate per head of population will even increase. All this is possible despite, or rather because of, engineering improvements that led to a reduction in the accident rate per km driven.

Accordingly, the prospect for greater public safety is unlikely to be found in a "technological fix" because of the way people respond to such fixes. Instead, the prospect for safety is inside the human being, not in the human-made machine or human-made physical environment. Where it is located in the human being and how it may be influenced will be proposed in the chapters that follow.

3.3 Replicating Taylor's findings

Excited and bewildered as I was by these rather unconventional but potentially very significant speculations, I grabbed the first possible opportunity to visit the author of the study that had triggered them. In this day and age of academic pressure to publish or perish, and the consequent threats to the quality and dependability of what appears in the scientific press, one wishes to check that one is relying on a serious person, and I myself have many times been subjected to inspection for that very purpose. Years earlier, a professor of surgery at the University of Amsterdam had told me that he went to scientific conferences, not primarily for the purpose of receiving the latest bit of information, but to get a personal impression of the reliability of authors whose publications he had read.

The second thing one of my students and I did was to see if we could replicate Taylor's findings while using a somewhat different method of data collection on drivers in locations as far afield as Windsor in England and Kingston in Ontario. We asked our sample of drivers to give a continuous indication of the amount of risk they perceived on a rating scale from one to ten. The results of these verbal ratings were the same as the earlier findings regarding GSR: the reported amount of risk per time unit of driving was essentially independent of where the driving was done and, just as Taylor had found, less experienced drivers gave higher risk ratings than old hands at the task of driving.[5] This corresponds neatly with the fact that inexperienced drivers are more likely to have accidents (see Chapter 10).

Years later, another student[6] produced another replication and extension of the preceding studies and found the following correlation coefficients:

Compare these findings with those in Section 3.1 above and you will see that they amply support the earlier ones. Moreover, the eleven individual drivers in this study showed considerable agreement with one another regarding the average perceived riskiness per km of the ten different road sections that were included in the study. The inter-rater reliability amounted to r = + 0.83. So, drivers are not only sensitive to conditions of different accident histories, but also show marked similarity in their risk perceptions. We return to this issue in Section 10.4.

3.4 The French connection

The formulation of what would eventually become Risk Homeostasis Theory (RHT) has taken a long time. A major step forward was made in the early seventies during a sabbatical leave, that leisure of the theory class, at the then National Institute of Road Safety in Montlhéry near Paris. It was then that I had an opportunity to try to accommodate the many factors known to be associated with accident likelihood in a single comprehensive model. For that purpose, I had written the elements I felt had to be entered in such a model on separate filing cards and arranged them in various different flow diagrams in hopes that one of these diagrams might be a reasonable representation of reality. One morning, while organizing the cards, I was suddenly hit by a two-pronged idea. The first prong was the notion of a closed-loop control process between accident occurrence and driver adjustment action. Thus was born the idea that changes in driver behaviour influence the accident rate (which is obvious) and that changes in the accident rate influence driver behaviour (which may be less obvious). The second prong was the concept of a target level of risk, which ultimately controls the accident rate, as the only important causal factor.

Eureka, everything on the filing cards seemed to fall into place. But the next moment I was filled with doubt. The notion of homeostasis appeared too much of an "idealization" to account for the complexities of human behaviour and "too rational", so to speak, to hold in a world where people only have incomplete information. Could the multitude of factors that play a part in accident occurrence really be grasped in such a comparatively simple model? Weren't there numerous ways in which such a theory could go wrong? People being so different from each other in skill and motivation, each having faulty perceptions of risk, and nobody--not even the experts--really knowing the precise dimensions of the accident toll,[7] there seemed to be plenty of reasons to question the idea. On the one hand, it seemed to me that it had to be true and on the other, that it couldn't.

Wasn't it heresy to propose a theory that might suggest that obvious advances in engineering, education, legislation and medicine have failed to reduce the rate at which people die as a result of accidents?

Making a first attempt at finding supporting evidence, I discovered some World Health Organization statistics that seemed to indicate that, although the per capita rate of traffic fatalities increased over this century, the total mortality rate due to all violent death remained very much the same. As it turns out, later studies would shore this up more firmly (see Chapter 12).

For a long, long time I have swayed between the feeling of jubilation and the fear of making a fool of myself. In fact, there have been several critics who felt that I had been quite successful in doing the latter. While some made congratulatory remarks,[8,9] the idea of risk homeostasis has been mocked as a Freudian and pseudo-scientific belief held by a morally and religiously prejudiced Dutch-Calvinist preacher (bien étonnés de se trouver ensemble!) who is blazing the notion of a perverse death wish.[10 ]Other colourful reactions referred to the theory as "Wilde's law of the conservation of misery",[11] "the devil's idea to some in the safety community",[12] or wondered if it is as difficult to prove as the existence of God[13] (which it isn't, thank God) or wrote in the process of quoting RHT that it wasn't worth quoting![14]

In another statement, the theory was rejected because "...the claim that risk per unit time is a constant is no more a theory than the claim that all people are the same height, or think they are the same height".[15] This criticism reveals a lack of understanding of homeostasis and attacks RHT on a position that is not even held by that theory. Homeostasis does not mean constancy (see Section 2.2). Another comment in the literature says: "In my view, a sufficient argument against the validity of risk homeostasis is provided by the incoherance [sic] of its "theoretical formulation";[16] unfortunately, this critic does not explain his reasons for the "incoherance" allegation.

Others have blamed RHT for being "negative" or "pessimistic" with respect to the potential for accident prevention.[17,18] Neither accusation seems to make much sense: there is nothing negative in saying that the sun does not revolve around the earth, as people have believed for centuries and some authorities even longer. How could being negative, in the sense of saying that something isn't true or doesn't work, ever be a scientific vice?

There is nothing pessimistic in RHT with respect to the potential for accident prevention. In the first place, it does acknowledge that the accident rate per kilometre of travel can be brought down. Secondly, it states that some accident countermeasures are unable to reduce the per capita accident rate, but also spells out how the accident rate per person in the population can be reduced by accident countermeasures of a different nature. Thus, proponents of RHT are no more pessimistic than physicians who tell their patients that a strep throat cannot be cured with bloodletting, while at the same time handing over a prescription for antibiotics.

Apart from a not-so-subtle pressure on journal editors to refrain from any further publication of the theory or even a bibliographical reference to it, perhaps the most extraordinary, and indeed bizarre reaction, was that of an editor who had invited me to write a chapter for a study guide. He did not like a large part of what I had written and replaced it with his own views. And then, unbeknownst to me, he published the text under my name. Plagiarism in reverse! Evidently, even within the academic world there are forces that would like to see the ivory tower lean in their preferred direction, the ivory becoming tainted in the process of their pushing. In case you, reader, are experiencing similar objections from your peers, here is my consolation: if your ideas are immediately acceptable to most of your colleagues, they are unlikely to be of much interest. Stanislav Lec offers another word of comfort: "People's understanding may be slow to kindle, but it will catch on by the next generation."[19]

But these unfavourable reactions to RHT were not to occur until several years later. I return to the time when I was still in sweet and wonderful France and comfortably close to the perfectly straight Eiffel tower. In addition to Taylor's work and my much earlier exposure to cybernetics, there was another publication that I had read within weeks, if not days, of that morning in the hills of Montlhéry, and it would give me a measure of confidence in my feedback model that tries to explain the accident rate.

That publication was the Report of the Club of Rome,[19] which was widely discussed at the time and has added a major momentum to concerns about world-wide environmental degradation and pollution. It had just appeared as a book entitled The Limits to Growth and it contained many examples of rather surprising feedback effects of technological innovations upon social and economic variables. One of these innovations, the so-called Green Revolution--a "technological fix" combining new seed varieties, fertilizers and pesticides--is a telling example of unforeseen and undesirable side-effects. This "revolution" produced greater rural prosperity in some countries but greater poverty in others, because of feedback mechanisms that are easily understood after the fact, but were apparently more difficult to predict in advance. In areas of the world where an equitable system of land distribution was in existence, the number of agricultural jobs grew faster than the population and real income rose. But in other areas, where most people worked very small lots and the vast majority of the land was owned by very few land-owners, the inequality was accentuated by the Green Revolution. The big-scale farmers adopted the innovation first, made profits and bought the landholdings of the small farmers. Increased unemployment among the latter, poverty, and migration to the cities was the result.

The Green Revolution produced a greater harvest yield per unit of arable soil, but not necessarily a greater prosperity per head of population. The effect of the technological innovation depended upon the nature of the human condition that adopted it. Because of the powerful impact that Limits to Growth had on me, and the obvious parallel between that book and the theory I am describing here, I have been tempted to call this book of mine The Limits to Safety.

At any rate, my mind was made ready to conceive and even to advance, albeit at first with hesitation, the idea that safety measures that reduce the accident rate per km driven do not necessarily enhance safety per head of population and may even diminish it.

The hesitation originated not only from the fear of being plainly wrong or the threat of ridicule, but also from my strong allegiance to the applied mission of ergonomics, that is, the notion that by "fitting the task to the operator", gains can be made in productivity, safety, health, comfort and satisfaction with the task, and all this by altering the physical features of the task environment rather than by interventions that try to change the operator. At first it seemed to me, as it must have to others, that by proposing the new views, I was guilty of disloyalty to the professed goals and methods of ergonomics. There is no need for such misgivings, because, depending upon societal goals, it can indeed make good sense to modify roads and cars in ways that reduce the accident rate per km driven, even if such interventions do not reduce the accident rate per time unit of road use (see Chapter 5).

The theory that resulted from this turmoil of conflicting considerations and inclinations I originally labeled "risk compensation theory." And in choosing that label I clearly made a mistake.[20,21,22] The term "risk compensation" is, strictly speaking, incorrect because, according to the theory, road users are not expected to compensate for risk in such a way as to reduce it to zero, but instead to show some form of behavioural adjustment in response to what might be called "changes in intrinsic risk". These are the changes in risk that would theoretically occur under the condition that road users would not alter their behaviour in the face of interventions, for instance, if they did not decide to drive faster when cars are made more crashworthy and roads are widened.

However, risk compensation theory says that they will alter their behaviour. Thus, labels such as "conservation of risk" or "safety compensation" might have been more appropriate, but unfortunately, these do not clearly point at the mechanism of homeostasis. Quite a number of authors have referred to my work using the terms "risk compensation" or "danger compensation", and some have made a distinction between "risk compensation" and "risk homeostasis" as if compensation were a soft-pedalling or watered-down version of homeostasis. They suggest that compensation might be partial and fall short of homeostasis, that is, complete compensation. That is not what I meant. Despite the relative unfamiliarity of the word "homeostasis", the term "risk homeostasis" seems to be preferable to "risk compensation", "risk conservation" or "safety compensation". Another possible label would be "the theory of behavioural compensation in response to changes introduced in intrinsic risk". But, although correct, this title is rather awkward. At any rate, all four labels are merely different names for the same fare.

[TABLE OF CONTENTS] [PREVIOUS CHAPTER] [NEXT CHAPTER]