En attendant Godot.
Waiting for Godot.

Samuel Beckett[1]

7. Remedy by engineering?

In order to show what may happen when an engineering measure that is unlikely to reduce the target level of risk is introduced, let us first consider the findings of a real-life study recently conducted in Germany.[2] This study deserves special interest, because it was commissioned by the federal ministry of transport in that country for the purpose of testing some empirical implications of risk homeostasis theory under well-controlled conditions of investigation. That explicit purpose was not present in several other studies that will be discussed later in this chapter.

7.1 The Munich taxicab experiment

Part of a taxi fleet in Munich was equipped with an anti-lock brake system--also known as ABS. This type of brake system prevents the wheels from locking up under extreme braking conditions. It offers the advantage of improved steering control over the vehicle during rapid deceleration, especially on slippery road surfaces. The system makes it possible to change the direction of the car and abruptly reduce speed at the same time, at a considerably reduced risk of losing control. ABS brakes offer a perfect example of what was called a change in intrinsic risk towards the end of Section 3.4--a change in the objective accident loss expected if drivers don't change their behaviour when a "safer vehicle" is made available. However, according to risk homeostasis theory, drivers are expected to change their behaviour and to maintain their accident likelihood per hour of driving as long as the target level of risk is not altered.

The cars with and without ABS in that Munich taxi fleet were of the same make and identical in all other respects. The majority of cab drivers were randomly assigned to one or the other of the two types of cars and the remaining drivers rotated between driving one type or the other. The exposure to traffic of each of the ABS taxicabs was carefully matched with cabs with traditional brakes over a period that lasted three years. Due to the matching procedure there was no difference in the time of day, the day of the week, the seasons, and the weather conditions in which both types of cabs were in operation.

Among a total of 747 accidents incurred by the company's taxis during that period, the involvement rate of the ABS vehicles was not lower, but slightly higher, although not significantly so in a statistical sense. These vehicles were somewhat under-represented in the sub-category of accidents in which the cab driver was judged to be culpable, but clearly over-represented in accidents in which the driver was not at fault. Accident severity was independent of the presence or absence of ABS.

In another part of their investigation, the researchers installed accelerometers in ten ABS and ten non-ABS cars, without the drivers' knowledge. These sensors measured the g-force of acceleration and deceleration once every ten milliseconds for a total of 3276 hours of driving. It was found that extreme deceleration, that is, extremely hard braking, occurred more often in the vehicles with ABS.

A third part of the study consisted of observations of driving style. Observers were trained in the systematic observation of a person's driving style and in recording their evaluations on rating scales. They were then instructed to call a taxi and to observe the traffic manners of the driver while they were passengers. A total of 113 such trips were made, 57 in cabs with ABS and 56 in cabs without. All trips covered the same 18 km route. Speed measurements were taken at four predetermined points of this route.

The drivers were not aware that their driver behaviour was being observed and the observers did not know whether they were in a taxi with ABS or without. The drivers did, of course, know whether or not they were operating an ABS cab, because of their familiarity with the car they were driving.

Subsequent analysis of the rating scales showed that drivers of cabs with ABS made sharper turns in curves, were less accurate in their lane-holding behaviour, proceeded at a shorter forward sight distance, made more poorly adjusted merging manoeuvres and created more "traffic conflicts". This is a technical term for a situation in which one or more traffic participants have to take swift action to avoid a collision with another road user.[3] Finally, as compared with the non-ABS cabs, the ABS cabs were driven faster at one of the four measuring points along the route. All these differences were significant.

In a further extension of their study, the researchers analysed the accidents recorded by the same taxi company during an additional year. No difference in accident or severity rate between ABS and non-ABS vehicles was observed, but ABS taxis had more accidents under slippery driving conditions than the comparison vehicles. A major drop, however, in the overall accident rate occurred in the fourth year as compared with the earlier three-year period. The researchers attributed this to the fact that the taxi company, in an effort to reduce the accident rate, had made the drivers responsible for paying part of the costs of vehicle repairs, and threatened them with dismissal if they accumulated a particularly bad accident record.

To sum up, in response to the installation of ABS brakes, drivers changed their driver behaviour. First, they utilized ABS to their advantage, but no improvement in the accident loss per time unit of exposure to traffic could be seen. Second, regardless of whether they were driving with or without ABS, a reduction in the accident rate did occur when the drivers' target level of risk was reduced by increasing their expected cost of risky behaviour.

The Munich taxicab experiment attracted a great deal of attention, not only in the professional circles, but also in the popular press. Newspapers carried articles about it and Bavarian Television wanted to show the viewers what had happened. As the experiment had already been completed, they decided to re-enact the experimental manipulation and the way the drivers had responded. Airing of this documentary added further to the popular debate. The results of this experiment were also discussed by a group of international experts from the Organisation for Economic Co-operation and Development, commonly abbreviated as OECD. In their final report, these experts from sixteen different countries stated that: "Behavioural adaptations of road users which may occur following the introduction of safety measures in the transport system are of particular concern to road authorities, regulatory bodies and motor vehicle manufacturers, particularly in cases where such adaptations may decrease the expected safety benefit."[4]

Another federal government wanted confirmation of the idea that drivers show an adaptation effect in response to ABS. The Canadian Ministry of Transport asked 81 drivers, selected from the general population, to perform a set of tasks while driving a car equipped with ABS which could be turned on or off at the turn of a switch. These tasks were carried out at the Transport Canada Test Centre in Blainville, Québec. They involved braking at a stop sign, accelerating to 70 km/h, emergency braking in a straight line, curve following, and emergency stopping in a curve. There was no interaction with other traffic on the test track. All drivers were informed of the features of ABS and told when they were driving with ABS on and when they would have to rely on standard brakes. Some drivers were given an opportunity to practise hard braking with ABS, while others were not.[5]

The most interesting results include the finding that, with the use of ABS, driving speeds and pressure exerted on the brake pedal were higher when drivers knew they were driving with the ABS system turned on. Further, higher maximum speed was observed in drivers who had experienced emergency braking with ABS as compared with those who had not. Most important, however, was the observation that the stopping distances during the braking manoeuvres were not any shorter in the presence of ABS than with standard brakes. They would have been shorter had the only driver response been to brake harder, without an increase in speed. Thus, the potential occurrence of shorter braking distances did not materialize. It was lost due to the fact that drivers utilized the more sophisticated brakes for higher speeds and harder braking, not for greater safety. This supports another statement in the above-mentioned OECD report: "An important conclusion of the Scientific Expert Group is that behavioural adaptation exists, and does have an effect on the safety benefits achieved through road safety programmes."[6]

Antilock brakes are a recent addition to a long string of supposed safety measures and they have been welcomed by many with great fanfare and heralded as a true life-saving device. It is interesting to note, however, that the limitations of better brakes in providing greater safety were already suspected in a footnote to a paper published almost 60 years ago, but the authors failed to identify motivational factors as the dominant determinants of the accident rate[7]: "...more efficient brakes on an automobile will not in themselves make driving the automobile any safer. Better brakes will reduce the absolute size of the minimum stopping zone, it is true, but the driver soon learns this new zone and...." You will have no difficulty in guessing the gist of the remainder of the sentence, but will you be able to resist the temptation of believing in the safety benefits promised by the next technological innovation, the next "technological fix", so to speak? Or will you still be waiting for Godot and willing to believe in the "unprecedented safety" heralded by the "high-tech" developments that are supposed to result in an "intelligent vehicle and highway system"?[8] By "safety benefits", we mean, of course, more safety per head of population, not per kilometre driven.

More difficult to understand are the first few words of the cited sentence, the first part that I left out, "Except for emergencies", and these words just do not seem to make sense. Maybe the authors were not quite certain that better brakes would fail to add to safety. What is certain is that the authors did not emphasize motivation or risk acceptance as the main determinant of safety. The quote comes from a mere footnote in their publication which, in general, emphasizes the importance of skill instead: "Safe and efficient driving is a matter of living up to the psychological laws of locomotion in a spatial field." As an aside, we may also note that psychological laws are supposed to describe the behaviour of any human being--including your behaviour and mine--whether we like these laws or not; we have no choice. Legal laws are different: we have the choice between compliance and going against them, and whether we do or not depends on psychological laws.

More safety per kilometre driven may well be expected from antilock brakes. Under dry pavement conditions, these brakes offer greater braking opportunity than standard brakes, because they are more likely to be activated with maximum foot pressure. Under slippery conditions, they offer the advantage of being able to brake and change direction at a lesser risk of skidding. Therefore, under both types of conditions, they offer an opportunity for greater speed without adding to unsafety. Driving faster means being able to drive a greater distance in the same amount of time. Drivers will thus have an opportunity for more mobility per time unit of driving. If greater mobility is attractive to them, they will drive more, with the end result that the accident loss per head of population does not change, while the accident rate per km driven is favourably affected. There is progress in the sense that people are given the opportunity of driving more kilometres per road accident. But at the same time there is stagnation in the sense that there's no reduction in the accident rate per person (see Equation 3 in Table 5.1).

7.2 The wheels of misfortune

"The wheels of misfortune" is the title of a report that describes the impact of engineering and policy interventions on the number of bicycle accidents on the campus of the University of California at Santa Barbara over a period of nearly seven years.[9] When classes were in session, some 10,500 bicycles per day entered the campus, which had an average yearly population of about 17,400 people. The use of bicycles was thus very common.

Common, too, were bicycle accidents. The Student Health Service on campus recorded an average of 249 bicycle accidents per year. All of these were accidents with personal injury, and more than one-tenth involved serious head injury, which occasionally led to withdrawal from university. Some of the more serious accidents have given rise to lawsuits against the university.

In order to reduce the number of bicycle accidents, various countermeasures were undertaken. Some of these were of the engineering kind; others were administrative in nature. The engineering interventions consisted of constructing bicycle paths that separated bicycles from both motor vehicles and pedestrians. In addition, a "bicycle traffic circle" was built that allowed two or more lanes of bicycle traffic to meet at an intersection of several bicycle paths without slowing down. Another, already existing, bicycle pathway was widened, and a large area was closed to bicycle traffic because it had been the scene of numerous accidents during periods of congestion.

One of the administrative measures consisted of refusing an automobile parking permit to students who lived within one mile (1.61 km) of the campus, and this was aimed at reducing parking problems, not bicycle accidents. The administrative measure that was meant to reduce bicycle accidents involved the removal of unsafely parked bicycles to an impoundment lot.

The researchers conducted a time-series analysis in which each of the above interventions was inspected for its effect on the daily bicycle accident rate. They found evidence that most of the engineering measures had the effect of increasing accidents, as did the limitation on automobile parking permits. The installation of the "bicycle round-about", when considered alone, also appeared to have increased the accident rate. The authors of "The wheels of misfortune" note that "it is hard to immediately know what to make of this finding. Perhaps the round-about made bike riding more hazardous or, alternatively, bike riders took advantage of the improved traffic flow to increase their speed. The latter is the interpretation apparently favoured by University engineers."

Of the two administrative measures (restricting automobile parking permits and impounding unsafely parked bicycles), only the first had an effect on the daily bicycle accident rate. A significant increase was the result, which is not surprising because a considerable increase in bicycle use was to be expected as a consequence. Unfortunately, however, this variable was not systematically monitored in this study and the same holds for many other features, as the authors themselves pointed out: "For example, increases in bike use unexplained by [seasonal variations in the size of] our student population variables may have swamped beneficial intervention effects."

This highlights a problem which is, alas, very common in traffic safety research. Some accident countermeasure is introduced and some yardstick of safety is assessed, but little information is gathered as to what happens in between. Why, and how, did the intervention achieve, or fail to achieve, a certain effect? Did people's behaviour change in response to the safety measure and, if so, in what way? Does the change in behaviour or the lack of it explain the subsequent accident rate? The Munich taxicab experiment referred to above is a model example in the sense that it was designed to provide answers to these questions, but such experiments are rare.

At any rate, in the case of the Santa Barbara bicycle accident study, it would seem fair to conclude (as the authors did) that the engineering and impounding interventions failed to reduce the number of bicycle accidents per time-unit of personal-injury accident recording.

7.3 Traffic lights

Since about one-half of all accidents in urban areas occur at intersections, many efforts have been made to make these traffic locations safer, for instance, by yield signs, stop signs and traffic lights. If we suppose that risk homeostasis theory is valid, what results would be expected when traffic lights are installed at these locations? Because this technical measure would not be likely to influence the amount of accident risk people are willing to take, it would not be expected to lower the accident rate.

This does not mean that traffic lights are not useful for other purposes. They provide an orderly assignment of right-of-way for different streams of traffic at different moments of time, so that the overall flow of traffic may be improved. They may also make it easier for pedestrians to decide when and where to cross.

Traffic lights are not to be condemned, but--contrary to naive opinion among some professionals and the general public--they serve no safety purpose, not even in the intersections proper or in their immediate vicinity. Numerous studies on the effect of traffic lights on accidents have compared the numbers of accidents at intersections before and after installation. The effect is that fewer right-angle accidents happen, but more rear-end accidents, as well as left-turn and side-swipe collisions, occur, and the total frequency remains roughly the same.[10,11, 12]The latter is also true for the average severity of intersection accidents. Although driver actions are drastically altered by these devices, accident loss is not, and the risk remains the same.

A common weakness of before-after comparisons is that traffic light installations are not made at a random selection of intersections. They are installed because of some peculiar characteristics of the intersections in question, for instance, their past accident rate. Accident rates at any specific road location vary from year to year, just as the weather does on the first of May. So if, in a given year, the accident rate at a given location is exceptionally high, one has reason to expect that it will be lower in the next year, even in the absence of intervention. Similarly, if the accident rate was particularly low in a given year, a rise in the next would be expected. This is what statisticians call the "regression to the mean" effect.

Moreover, the possibility that drivers may change their routes as a consequence is often not considered, and counts of passing traffic before and after the installation of lights are often not available. Thus, it is difficult to estimate what the accident loss at the intersection would have been had no traffic light been installed.

This is why a cross-sectional study of accidents at 137 intersections in the city of Kitchener, Ontario, is of special interest.[13] There were four types of traffic control at these intersections: traffic lights, a stop sign, a yield sign, or nothing at all (meaning that the basic rule that traffic coming from the right has the right of way, applied).

No change in traffic control at these intersections was carried out. The investigator proceeded as follows. He collected data on the geometry of the intersections, collisions between vehicles, and traffic volumes for each intersection for one calender year. On the basis of the geometry and volume data, an index of potential hazard was calculated for each intersection. This index is the number of accidents that would have happened theoretically if no driver made any effort to avoid a collision. As you can imagine, this essentially amounts to the number of cars approaching the intersection on one street, multiplied by the number of cars on the crossing street, the product being divided by the amount of space in the intersection.

The index of potential hazard was then compared with the number of accidents that had actually occurred at each street crossing. An index of safety for each crossing was obtained by dividing the number of potential accidents by the actual number of accidents. Thus, the smaller the actual accident rate per intersection relative to its potential hazard, the greater the safety index was. It indicates, therefore, how effective drivers were in avoiding potential accidents.

Finally, the four types of intersections, each categorized according to the type of traffic control in place, were inspected on whether there were any differences in this safety index. No differences were found. Driver effectiveness in avoiding collisions with other vehicles at intersections was not helped by yield signs, stop signs or traffic lights. The author concluded that the four types of "intersection control devices do not affect the total number of collisions between vehicles".

He noted, too, that the concentration of accidents at intersections has led to countless public demands on city engineering departments for corrective action at these locations, and that at the time of the study some 60% of all intersections in urban areas in Ontario had been equipped with some control device. At that very same period of time, less than 5% of the intersections in Oslo, the capital city of Norway, were controlled by signs or signals. Yet, within the city boundaries of Oslo, about one-half of all accidents happened at intersections, just as they did in Ontario cities. So, as far as yield or stop signs and traffic lights are concerned, whether or not an intersection control device is installed doesn't seem to make much difference to collision frequency, nor does the type of traffic control installed. What does reduce the accident rate is discussed in Chapter 11.

Cross-sectional studies, like before-after comparisons and longitudinal studies over extended periods of time, also present their problems of interpretation. One cannot be sure that drivers or their personal characteristics are the same from one intersection to another or from one time frame to another, let alone from one jurisdiction to another, such as Norway and Ontario. What we can say, however, is that the available evidence from both before-after comparisons and cross-sectional studies does not support the notion that traffic control installations at intersections have a beneficial effect on safety. Drivers do behave differently at intersections with different control signs and signals, but once again we see no change in risk.

Engineering modifications of the roadside at locations other than intersections bring about rather similar changes in driver behaviour, according to the OECD report we mentioned above in Section 7.1. Here are some illustrations. Increases in lane width of two-lane highways in New South Wales, Australia, have been found to be associated with higher driving speeds: a speed increase by 3.2 km/h for every 30 cm additional lane width for passenger cars, and for trucks an increase of about 2 km/h for every 30 cm in lane width. An American study dealing with the effects of lane-width reduction found that drivers familiar with the road reduced their speed by 4.6 km/h and those unfamiliar by 6.7 km/h. In Ontario it was found that speeds decreased by about 1.7 km/h for each 30 cm of reduction in lane width. In Texas, roads with paved shoulders, as compared to unpaved shoulders, were found to be associated with speeds at least 10% higher. Drivers have generally been found to move at a higher speed when driving at night on roads with clearly painted edge markings.

Recently, a Finnish study investigated the effect of installing reflector posts along highways with an 80 km/h speed limit. A total of 548 km of randomly selected road sections were equipped with these posts and compared with 586 km that were not. The fact that the installation of reflector posts increased speed in darkness will no longer come as a surprise. There was not even the slightest indication that it reduced the accident rate per km driven on these roads; if anything, the opposite happened.[14]

By now, what may be surprising instead is that the OECD report, although it frequently refers to changes in the accident rate in relation to some modification in engineering, rarely spells out the denominator of the accident rate under consideration, such as per km driven, per hour of driving or road use, or per head of population. This is puzzling because the report was prepared by an international committee of experts to deal with the challenge posed by risk homeostasis theory, among other views. In that theory, the distinction between the accident rate per km driven and per hour of road use is as essential as the distinction between death per cigarette smoked and death per cigarette smoker, and between the proverbial apples and oranges.

7.4 Motor-vehicle manufacturing regulations

A major package of legislative regulation concerning the "safe" design of new passenger vehicles to be sold in the USA came into effect in 1966. This included the obligatory installation of seatbelts for all vehicle seats, a steering column that would collapse in a crash instead of piercing the driver's chest, penetration-resistant windshields, a dual braking system, and padded dashboards. The effect of these mandatory construction features upon subsequent accidents was studied by an economist at the University of Chicago.[15]

Comparing the pre-regulation period 1947-1965 with the 1966-1972 period, in which there were more and more regulated vehicles in use, he came to the conclusion that the newly-legislated vehicle-manufacturing standards had not led to a reduction in the number of fatalities per km driven. While the legislation may have brought about a reduction in fatal accidents to car occupants per km of mobility, it did not reduce the total death rate so defined. It may, in fact, have led to an increase in the death rate of non-occupants, such as bicyclists and pedestrians, per motor-vehicle distance of mobility. A similar shift in risk from drivers to pedestrians has been reported in Australia.[16]

The Chicago study was published in 1975, and has been attacked ever since by many other authors who maintained that the vehicle-manufacturing standards have had a reducing effect upon the traffic death rate per unit distance driven by motor vehicles. There have been others who found evidence supporting the controversial findings.[17] In fact, in 1989, the issue was still not settled.[18] You may already have concluded, and correctly so, that this debate is, at best, only marginally relevant to the question of the validity of risk homeostasis theory. There is nothing in that theory that says that the accident rate per km driven cannot be reduced by technological interventions, regardless of whether they are mandated or not. What we are interested in is the accident rate per hour of exposure to the roads and per head of population. As regards the post-regulatory years 1966-1972, one definitely cannot detect in Figure 5.5 a lower per capita traffic death rate than in the preceding years 1947-1956. On the contrary, it was noticeably higher.

What you can see in Figure 5.5 is that the increase in the traffic death rate per capita from 1961 to 1965 did not continue between 1966 and 1972. Was this due to the vehicle-manufacturing standards that came into effect in 1966? This would seem rather unlikely. Note that the period 1966-1972 falls within the 1960-1982 time frame that has shown a high correlation between the rate of employment and traffic fatalities, as was discussed in Section 5.4. That correlation was very high and leaves little room for anything else that may have exerted an independent influence. The effect of the 1966 legislation on the per capita death rate in traffic, if it occurred at all, must have been quite marginal.


Copyright © 1994 Gerald J. S. Wilde, Ph.D.
since FEB-10-96.