The unformatted htm text below is from

Copyright © 2004 by Leonard Evans

Information about

How to purchase

Table of Contents

Text

Reviews

The text below is © Leonard Evans.  Portions totaling not more than 400 words my be freely used with appropriate citation.
For more extensive use, permission in writing must be first obtained by emailing the author 

9 Driver behavior

This  html version contains only the text (no figures, tables, equations, or summary and conclusions) To check printed book appearance see pdf version of Chapter 1 or pdf version of Chapter 16.

Introduction
     It is crucial to distinguish between driver performance and driver behavior. Not differentiating between them has caused, and continues to cause, confusion. The two concepts are:
Driver performance - what the driver CAN do.
Driver behavior - what the driver DOES do.
Driver performance relates to the driver's knowledge, skill, perceptual and cognitive abilities, as discussed in Chapter 8. Driver behavior is what the driver chooses to do with these attributes. The example in Fig. 8-2 (p. 183) showed that the probability and severity of a crash depends on driver reaction time, a driver performance attribute. However, the outcome also depends on the speed of the vehicle. The reason why drivers choose different speeds is not conveyed by Fig. 8-2. The ability to judge speed, and the capability to control the vehicle at that speed, are aspects of driver performance. The speed chosen is at the core of driver behavior.
     As driver performance focuses on capabilities and skills, it can be investigated by many methods, including experiments using laboratory equipment, driving simulators, and instrumented vehicles traveling on test tracks. As driver behavior is what drivers actually do, it cannot be investigated by such methods. As a consequence, we have less solid quantitative information about driver behavior than about driver performance. Particularly important, but difficult to quantify, are relationships between driver behavior and crash risk.
Normal driving is a self-paced task
     
The distinction between performance and behavior is central to traffic safety because normal driving is a self-paced task. That is, drivers choose their own desired level of task difficulty. Increased skill can be used for many purposes. A likely use is to choose a different level of task difficulty. In Chapter 5 we found that a technology that improved vehicle braking, antilock braking systems (ABS), was associated with a large increase in rollover risk, and did not enhance safety in the way that a naïve interpretation of impressive test track results indicated. In the same way, an increase in driving skill may not increase safety because it can be used for such purposes as increasing speed, overtaking in tighter situations, or performing more secondary tasks, like talking on
cell phones. While crash rates of older drivers have been shown to increase with declining performance, there is a conspicuous absence of convincing evidence of any relationship between crash risks and performance measures for drivers who are not old.
A distinguished sea captain commented:
A superior seaman uses his superior judgment to keep out of situations requiring his superior skills.
Drivers behaving in such a way would enhance safety. However, a superior driver who has learned skid control or advanced braking techniques may well seek opportunities in traffic to exhibit these skills.
     When task difficulty is maintained constant, higher skill has been shown to increase safety. Crash rates of Helsinki bus and streetcar drivers were found to be strongly correlated with a series of performance-measuring tests. In addition, crash rates were stable over long periods when monitored from 1947-1973. The basic distinction between these drivers, who were employed to perform specific tasks, and drivers in general, is that their schedules, and other aspects of driving behavior, were specified. Higher skill could not be used to increase speed - indeed, keeping closely on schedule would relate to skill. More recent research has developed screening procedures that identify commercial fleet drivers with higher than average probability of crashing. Such procedures are not available for normal driving.
Driving in a regulated structured environment has features in common with piloting a commercial airliner, so that increased skill, knowledge and performance are expected to increase safety in both situations. However, what is crucial is that this does not apply to normal self-paced driving. Misguided policy has often resulted from a mistaken belief that safe driving is primarily a perceptual-motor skill, and that measures that increase driving skill will improve safety.
Racing drivers compared to average drivers
    
The belief that increasing skill would reduce crash rates has seemed to many too obvious to be worth investigating. Such a belief reinforces the view that driver education must increase safety, even in the face of so much evidence that it does not (Chapter 8). It is widely held by driving aficionados that high-skill drivers are inherently safe drivers.
This was examined directly by comparing the on-the-road driving records of unusually skilled drivers to the records of average drivers. The investigators obtained the names and addresses of national competition license holders from the Sports Car Club of America. They compared the on-the-road driving records of these license holders (referred to in their paper as racing drivers) in Florida, New York, and Texas, to comparison groups of drivers in the same states matched in such characteristics as gender and age.
     The results of the study are summarized in Fig. 9-1, which displays the violation and crash rates for the racing drivers divided by the corresponding rates for the comparison drivers. If there were no differences between the groups of drivers, these ratios would all be close to one, whereas if the racing drivers had lower rates, the ratios would be less than one. What is found is that in all 12 combinations examined, the rates for the racing drivers exceeded those for the comparison drivers, in most cases by considerable amounts. Thus, on a per year basis, the racing drivers not only had substantially more violations, especially speeding violations, but also more crashes.

Figure 9-1. The number of incidents per year for racing drivers compared to the values of the same quantities for comparison average drivers matched in gender and age.5

    It could be claimed that higher speeds by more skillful drivers do not cause harm if the person driving faster is more able to avoid crashes. What is most telling in this study is not the higher violation rates, but the higher crash rates. Self-reported estimates of distance of travel indicated that the racing drivers traveled more than the comparison drivers.   However, additional analyses found that this does not explain all of the observed differences.In interpreting the difference between the driving records of the racing drivers and the comparison drivers, it is not possible to determine whether the effect flows from the use of the additional skill acquired by the drivers to drive more aggressively, or whether it is simply high risk drivers who are attracted to racing. It is possible that without the additional skills acquired in pursuit of their advanced license they might have had yet higher crash rates, although I find this improbable. The study does show that higher skill levels are not associated with lower crash rates.
Effect of speed on risk
    
Quantitative relationships between crash risk and pre-crash travel speed are difficult to obtain because the speed that a vehicle was traveling prior to its crash is not generally known. Pre-crash speeds are rarely recorded, and it is an involved and uncertain process to infer speeds from such physical evidence as might be produced by the crash. Skid marks provide some information, but antilock brakes eliminate these by maintaining wheel rotation under maximum braking. When skid marks are present they still provide no information about how much speed reduction occurred when the vehicle was not skidding. The detailed post-crash examination required to estimate speed precludes its inclusion in large data files like FARS.
Urban speeds and injury crashes - case-control method
    
The effect of travel speed on crash risk was investigated by applying a case-control study design to data collected on urban streets in the Adelaide (Australia) metropolitan area. The case-control method involves comparing the speeds of cars involved in crashes, or case cars, to the speeds of control cars not involved in crashes. The case sample included only cars involved in casualty crashes in which someone was transported to a hospital. All case drivers had zero blood alcohol concentration. Pre-crash speeds of case cars were determined using crash reconstruction techniques involving detailed investigations of the crash scenes. Features of the crash, such as skid marks, impact points, final positions of cars, damage to cars, and participant and witness statements contributed to the reconstruction process. The control cars matched the case cars in direction of travel, roadway location, time of day, day of week, and time of year. Control-car speeds were measured by a laser speed meter. All data were for roads with 60 km/h speed limits.
     The data led to the risk ratios plotted in Fig. 9-2. All values are relative to the base case risk of traveling at the 60 km/h speed limit. The error limits are one standard error, as determined by the number of case and control cars at each speed interval and at 60 km/h.
     Risk of involvement in a casualty crash increases steeply as speeds exceed the 60 km/h speed limit, approximately doubling for each additional 5 km/h. The increase is so steep that a logarithmic scale is used in Fig. 9-2. The risk at 80 km/h is 30 times the risk when driving at the speed limit.
     There is no clear indication that risk decreases for travel speeds below the speed limit. There are other data associating speeds lower than the average with increased risk. However, such findings underline some of the most difficult problems encountered in traffic safety research. Observing that slow drivers have higher crash risk does not imply that driving slower increases risk. The reason they are driving slower could be in response to some other risk-increasing condition (fatigue, poor vision, major distraction, vehicle problems). Similarly, the higher speed drivers might have higher risks even when they drive slower. There is little opportunity for traffic safety studies to provide results that can be interpreted without some element of judgment regarding plausible and implausible explanations. Notwithstanding the uncertainties that are always present, the data in Fig. 9-2 provide clear evidence that risk of involvement in a casualty crash increases steeply as a function of how much a driver exceeds the speed limit.
Rural speeds and injury crashes - case-control method
     
The approach above was applied to rural roads in South Australia with speed limits of 80 km/h or higher. Because of smaller sample sizes, different speed limits, and more varying driving conditions, the results were expressed in terms of deviations from average speeds of control cars rather than departures from speed limits. As with the urban study, all cars contributing to the study were traveling at free speeds, meaning that they were not overtaking, following, accelerating, slowing, etc.
The authors summarized the results as
9-1
where dv = speed above the average speed (km/h), and
R = risk at speed dv above average compared to risk at average speed.
Eqn 9-1 estimates that a car traveling 10 km/h faster than the average speed in a rural area will be 2.2 times as likely to be involved in a casualty crash as a car traveling at the average speed.
Speed limit changes and casualty risk
    
Another approach to examining how speed affects risk is to relate changes in aggregate casualties on selected roads to changes in the average speed of the traffic on these roads. This approach provides relationships without the need to know how fast specific crash-involved vehicles were traveling prior to their crashes, which, as in the case-control studies described above, is difficult and expensive to obtain.
     Claims in the literature that speed limits have had no effect on safety have led to some confusion. What is written in a legal statute or on a road sign does not, by itself, influence safety. If all 40 mph signs were replaced by 35 mph signs in the middle of the night, it is unlikely that many commuters would notice the change the next morning. It is travel speed, not speed limits, that affects safety. For speed limits to affect travel speed they must be known to motorists, and, for many motorists, they must also be perceived to be enforced.
     Relationships between crash risk and travel speed were determined by examining casualty rates before and after changes in speed limits on many (mainly higher speed limit) roads in Sweden. In many cases, speeds on roads with changed speed limits were compared to speeds on similar roads that did not have their limits changed. The results were found to fit the power function
9-2
which, for small percent changes in speed, can be approximated by
9-3
P is a power that depends on the severity of the crash:
for any crash P = 2
for serious-injury crash P = 3
for fatal crash P = 4
These results and the increasing value of P with increasing severity supports the pattern that speed increases lead to increases in the:
- Risk of crashing
- Risk of serious injury, if a crash occurs
- Risk of death, if a serious injury occurs
     While these relationships estimate substantial increases in risk with increasing speed, the increases are not as large as found in the case-control studies. For example, Eqn 9-2 estimates traveling at 90 km/h compared to 80 km/h increases serious injury risk by a factor (90/80)3 = 1.42. This 42% increase, while substantial, falls short of the 120% increase estimated by Eqn 9-1 for a vehicle traveling 10 km/h faster than the average speed. Fitting Eqn 9-2 to values computed using Eqn 9-1 gives P = 6.6, and the fit to the urban data in Fig. 9-2 gives P = 12. The differences could reflect that in the case-control studies, the speeding case drivers might be less safe than average drivers even when they were not speeding, because other risk-increasing behaviors are related to speeding.
    A British study reports a value of P = 2.5 for all crashes involving injury, based on comparing rates on different roads with similar features but different observed average speeds. Another British study reports that for every 1 mph rise in the mean traffic speed, crashes rise by about five percent, equivalent to P = 5 for all crashes.
A study in Israel found that increasing speed limits from 90 km/h to 100 km/h on three inter-urban highways led to speed increases of about 7%. Traffic deaths increased by 15%, but the authors claim that this observed increase would have been larger had it not been for other concurrent countermeasures. They found that the number of deaths per serious injury increased by 38%. If the P value for fatalities in Eqn 9-2 is PF, and that for injuries is PI, then the ratio of these gives approximately the increase in the risk of death in an injury crash as PF/PI " (38%)/(7%) " 5. This is consistent with high P values. This study finds increased casualties on roads with unchanged speed limits, providing further evidence of spillover effects.
     The studies measuring mean speed cannot identify which individual vehicles contribute most to overall risk. However, the results from the case-control studies, which did measure speeds of individual vehicles, support that the major contribution to aggregate risk is from the few fastest vehicles. If we assume that the variance to mean ratio remains constant when a change in average speed occurs, then the fastest vehicles will increase their speeds by more than the increase in average speed. It is the fastest drivers who make the main contribution to risk before the speed increase, and it is their larger than average speed increase after the change that makes the major contribution to the aggregate risk increases observed after increases in average speed.
Lowering of speed limits on US rural Interstates and fatality rates. A national maximum speed limit of 55 mph became law throughout the US on 1 January 1974. It was introduced in response to the October 1973 Arab oil embargo in order to reduce fuel use - safety was not a consideration. However, the speed limit change provided a unique opportunity to evaluate how travel speed affects fatality risk.
     Prior to the change, the speed limits on rural Interstates were nearly all 70 mph, with a few being higher. Average travel speed was observed to be 63.4 mph. After imposition of the 55 mph limit, the observed speed declined to 57.6 mph, a seemingly modest 5.8 mph drop compared to the 15 mph decrease in the speed limit. However, a sharp drop in the fatality rate occurred (Fig. 9-3). The change in the speed limit was one of many changes from 1973 and 1974 related to the oil embargo. Reduced travel was another. This and other changes should not directly affect the number of deaths for the same distance of travel, the rate plotted.

Figure 9-3. Fatalities per billion km on the US rural Interstate system. Until 1973 maximum speed limits were 70 mph (in some cases greater). From 1974 to 1987 there was a nationwide 55 mph speed limit. In 1987 some states increased limits on some portions to 65 mph. Based on Ref. , p. 340.

     Substituting the after and before speeds into Eqn 9-2, with P = 4 for fatality crash risk, gives (57.6/63.4)4 = 0.68, thus estimating a 32% decrease in fatality rate. The change in the fatality rates plotted between 1973 and 1974 is a 34% decrease. A small part of the decline is likely due to an ongoing trend (as expected, see Chapter 3), but even after taking this into account, the observed decline supports that P = 4 is an appropriate choice in Eqn 9-2 for fatalities.
Raising of speed limits on US rural Interstates and fatality rates. On 28 Nov-ember 1995 the Federal Government terminated its involvement in speed limits, returning the responsibility entirely to the individual states. Before that, in April 1987, Congress had passed legislation allowing states to increase speed limits to 65 mph on some sections of rural Interstates. The introduction of the national maximum speed limit occurred at one time, and applied nationwide, which led to the large easily observed change in Fig. 9-3. When the limit was made more flexible, and later ended, states were not required to make any changes. Some did, some did not, and the states that made changes did not make identical changes, or change at the same time. As a result, estimating how increases in the speed limits affected fatalities cannot be performed by a simple calculation at the national level. Many evaluations were performed, including those by individual states, giving a consistent pattern that the states that increased their limits from 55 mph to 65 mph experience about a 10% increase in fatality rates.
     Within a year after the Federal Government ended any role in setting speed limits, 23 states raised their rural Interstate speed limits to 70 or 75 mph. Many studies showed that this increased fatalities. One study compared changes in the number of fatalities for the same distance of travel for states that increased their limits to states that kept their limits at the then prior limit of 65 mph. The states that increased speed limits to 70 mph experienced a 35% increase in fatality rate, and the states that increased speed limits to 75 mph a 38% increase. Data are unavailable for actual changes in speed in response to the speed limit changes.
Speed and speed change - v and Dv
It is not travel speed, v, that plays the direct role in injuries, but the change in speed, Dv, the vehicle undergoes as a result of the crash (Chapter 2). Analysis of data for crashed cars showed that, when a crash occurs: ,
9-4
9-5
     The fatality relationship, Eqn 9-4, was introduced earlier and used to produce Fig. 8-2, p. 183. In general Dv is smaller, usually much smaller, than v. However, the simple (but rare) case of a vehicle striking an immovable object without braking gives Dv = v (assuming no bounce-back). If the risk of striking the object increases as v2 (P = 2 for crash risk), then fatality crash risk will increase as v(2 + 3.54) = v5.54 giving, in terms of Eqn 9-2, the values P = 5.54 for fatalities and, similarly, P = 4.22 for injuries. When there is braking, Dv will increase more steeply than proportional to v (Fig. 8-2), so the values P = 5.54 and P = 4.22 may be low. This very approximate calculation supports the higher P values reported from the case-control studies.
German Autobahns
     
Opponents of speed limits often claim that German Autobahns have no speed limits yet low casualties. The claim of no speed limits is countered by temporary speed limits on most of Germany's highways. Signs suggesting a recommended speed limit of 130 km/h (80 mph) are posted along most Autobahns, while urban sections and a few dangerous stretches sometimes have posted speed limits of 100 km/h. Congestion, which reduces fatality risk by reducing speeds (Chapter 5), is common.
Surprisingly, there does not appear to be any technical study comparing safety on Autobahns to safety on freeways in other countries. Data in a column listed killed per 1 billion km - motorways shows that the German rate, while 27% lower than the US rate, is 48% higher than the Swedish rate and 76% higher than the British rate (see Chapter 15 on failure of US safety policy).
Speed variance
Variability around average speeds, or speed variance, has been discussed as a safety factor, especially since observations associated higher risks with speeds below as well as above the average.7, An analysis interpreting crash risk as proportional to the number of times a vehicle is overtaken or overtakes concludes similarly that traveling below or above the average speed increases crash risk. The common policy of setting lower speed limits for trucks than for other vehicles would seem to imply rejecting the notion that driving slower than the average speed increases risk.
From time to time there are claims that speed variance is as important as speed. The most extreme view claims that speed, as such, does not affect safety and that the safety goal should be to require all traffic to move at the same (presumably high) speed. After all, vehicles traveling at identical speeds in the same direction cannot crash into each other, and will never overtake. While there are indications (not supported in Fig. 9-2) that vehicles traveling slower than average have above-average crash rates, the reason they are traveling slower is because the driver chooses to do so, likely in response to known driver (or vehicle) inadequacies. If such drivers were encouraged or compelled to speed up to the average speed, an increase in crash risk seems more likely than a decrease. Even if slower-than-average drivers have crash rates similar to faster-than-average drivers, their injury and fatality rates will still be substantially less.
     Higher speed variance would appear to increase the risk of rear-end crashes. However, only 5.0% of occupant fatalities in FARS 2002 occurred in rear-impacted vehicles (initial impact point 5, 6 or 7 o'clock). Single-vehicle crashes, for which the concept of speed variance is hardly relevant, account for 49% of driver fatalities (Table 3-3, p. 48). Thus speed variance can make no more than a modest contribution to fatalities, especially when one additionally considers how steeply the risk of death increases with speed. Traffic in which all vehicles traveled at 40 mph would, according to some definitions, have zero speed variance. Yet a head-on crash between two vehicles traveling at 40 mph is far more lethal to both drivers than one between a 40 mph and a 20 mph vehicle. The speed variance concept has been applied to vehicles traveling on different roads, not merely in opposite directions on the same road! Multivariate analyses that find speed variance a prime factor in determining fatalities simply provide another example of how such methods can be coaxed to support any conclusion. Even if speed variance were all that important, the way to reduce it is to reduce the speed of the fastest vehicles, which is already the current focus of enforcement.
Comments on relationship between speed and risk
   
An enormous body of evidence consistently supports that the risk of crashing, being injured, or being killed, increases with increasing speed. There is no doubt that the dose-response curve is very steep - a little extra speed generates a lot more harm. How much more harm per increment of speed increase is not known to much precision. One source of uncertainty is variability in results due in part to methodological differences. However, it is also likely that different relationships apply to different roads, circumstances, and driver populations. For fatalities, a one percent increase in speed appears to increase fatality risk somewhere in the range 4% to 12%, with stronger support for a value towards the low end of the range. The lower value indicates that a 3% speed reduction reduces risk by 13%. This is larger than the reduction from frontal airbags. A 1% to 2% nationwide reduction in speed provides a nationwide reduction in traffic deaths similar to that provided by airbags. Property damage, a major cost of traffic crashes (Chapter 2), is reduced by speed reductions, but increased by airbags, which must be replaced after deployment (more on airbags in Chapter 12).
     Lower speed limits reduce the efficiency of travel. The safest traffic system consists of gargantuan vehicles all of which are nearly stationary. Setting speed limits involves trade-offs. Studies taking into account delay, injury, fuel use, emissions, and other costs have estimated socially optimum speeds of about 50 km/h for urban streets, and 85 to 105 km/h for rural roads. Actual speed limits (which drivers routinely violate) are found to be generally higher than the estimated social optimum.
     Reducing speed limits is almost certain to increase the number of law violators. Opponents of lower speed limits often make a big issue out of the fact that more drivers violated the US national maximum speed limit of 55 mph speed limit than the prior 70 mph limit. A 150 mph speed limit would produce many deaths, but few speeding tickets (although there are 27 car models available in the US claiming top speeds in the range 150 to 240 mph).
Demographic factors related to risk taking in traffic
     The empirical studies showing that speed choice (a driver characteristic) has a large influence on crash risk were not based on measuring a characteristic of the driver, but a characteristic of the vehicle (its speed). In most of the studies, information about the driver traveling at the measured speed was not available. Obtaining driver information would require major increases in experimental difficulty and cost. Various methods have been used to relate aggregate driver characteristics to speeding, and, more generally, risk taking in traffic. No method is without some weaknesses, but as different methods have different weaknesses, they collectively generate a fairly consistent picture that provides insight into why so many crash rates have the clear gender and age dependencies reported in Chapter 7.
Traffic tickets
Table 9-1 shows the percent of residents of North Carolina with driver licenses who received a traffic-law violation citation (ticket) in 1998. Violations related to crashes are excluded.

Table 9-1. The percent of North Carolina residents with driver licenses who received traffic citations (those involving traffic crashes excluded) in 1998.26

For each race and age comparison, the male rate far exceeds the female rate (by factors between 2.0 and 3.3). For each race and gender comparison, drivers aged 16-22 have rates far higher than those for drivers aged 50 or older (by factors between 3.3 and 11.2).
     The number of traffic tickets a driver gets in a year reflects more than just the propensity to violate traffic law. Drivers in each of the 12 cells in Table 9-1 travel different distances at different times, and therefore necessarily have different probabilities of being observed by police. What is a more controversial and difficult issue is that all individuals may not have identical probabilities of receiving tickets even for identical behavior in identical circumstances. The probability that a police officer issues a ticket might be prejudicially affected by the gender, age, or race of the driver. The data alone cannot provide information on the extent to which sexism, ageism, or racism are present. The agreement of the gender and age dependence of ticket rates with casualty rates supports that a major reason why drivers receive tickets is that their behavior in traffic increases the risk that they will be in a fatal crash. However, contributions from other factors cannot be ruled out. Among young (aged 16-22) drivers, African Americans of both genders have lower citation rates than white drivers of the same gender and age. This may reflect less availability of vehicles, less driving, or perhaps young African Americans are driving more carefully because they believe that police are prejudicially targeting them. Such a belief, whether true or not, would save lives of those who believed it. The higher rates for African Americans in the other age groups may reflect a largely universal socioeconomic effect in which risk taking in a host of activities declines with increasing affluence.
Large data files
The only demographic factors recorded in US traffic crash files, such as FARS, are gender and age. Many results based on these variables have been presented in earlier chapters. Mortality from the National Center for Health Statistics contains more detailed information on those dying from any cause. By linking this data set with FARS, together with information on distance of travel from the 1990 Nationwide Personal Transportation Survey, fatality rates for the same distance of travel were estimated for factors in addition to those included in FARS. The study examined occupant fatality risks for children (aged 5-12) and teenagers (aged 13-19 years), and compared the risks for Hispanics, non-Hispanic blacks, and non-Hispanic whites. For each billion vehicle miles of travel, 14 non-Hispanic black children were killed, compared to 8 Hispanics, and 5 non-Hispanic whites. For teenagers the rates were 45 for Hispanics, 34 for non-Hispanic blacks, and 30 for non-Hispanic whites. Unquestionably, socioeconomic status makes a large contribution to any such differences, as well as to the differences in Table 9-1. However, a later study of adult risk that does incorporate a measure of socioeconomic status concludes that this variable cannot account for all of the differences.
Self-reported behavior in questionnaires
Questionnaire, or survey, studies have the advantage of producing information on many factors for large numbers of respondents. A disadvantage is that self-reports of behavior may depart substantially from actual behavior, particularly if the behavior is considered to be socially undesirable or is illegal. Even in the unlikely event that drivers knew how often they speeded, they might not respond truthfully. Comparing self-reports to observed behavior shows that speeders tend to underreport their speeding. Still, responses reflect some mix of true speeds, the responder's ability to estimate true speeds, and the self-image the responder wishes to convey. These are all factors relevant to safety.
Responses were obtained from 1,095 licensed drivers in Scotland in a 20-minute in-home interview. Table 9-2 shows the percent characterized as speeders, defined as reporting that they had been stopped by the police for speeding during their driving career, or had been flashed by a speed camera in the previous three years (the driver is aware when a flash photograph, indicating speeding, is taken). At all ages, male rates exceed female rates by factors between 1.6 and 2.4. The data do not provide information on speeding versus age because, as drivers age, their life-long exposure to the risk of a speeding offense can only increase.
The same survey asked drivers how they would adjust their speed under various driving conditions. Differences reported to be statistically significant are listed in Table 9-3, again demonstrating that males are associated with riskier behavior. The greater tendency of males to slow down in light rain could possibly reflect greater appreciation of the relationship between braking and wet roadways, or (more likely) driving closer to the margin on dry roads.

Table 9-3. How Scottish drivers of different genders indicated they would respond to changed conditions.30

Observational studies
Driver propensity to take risk, as indicated by chosen speed or by following headway (gap between a driver's vehicle and the one in front), was measured in a series of observational studies in which oncoming cars were photographed from freeway overpasses. , The license-plate number, read from the photograph, was used to extract from state files the driving record, gender, and age of the registered owner of the vehicle. A photographed driver judged not to differ in age or sex from the registered owner was assumed to be the owner.
The studies found that speed on a rural two-lane road decreased with increasing driver age. The proportion of vehicles observed following the vehicle in front at dangerously short headways (tailgating) declined with increasing driver age. The risk taking behaviors of speeding and tailgating were more prevalent among male drivers. These results are summarized in more detail elsewhere.13(p 137-138)
Personality factors and crash rates
The above relationships between risk-taking in traffic and gender and age apply to averages of large groups. The behavior of individuals within a given demographic category vary more than differences between demographic categories - there are many young males who drive at lower risk than many older females. There has been much effort to attempt to explain such variation in terms of personality.
The first study indicating "we drive as we live"
One of the earliest studies to examine the relationship between crash involvement and personality was published in 1949. Characteristics of 96 Toronto taxi drivers who had four or more crashes were compared to those of 100 taxi drivers with no crashes. The two groups, matched in age and driving experience, provided the results in Table 9-4.

Table 9-4. Frequency of contact with social agencies by 96 Toronto taxi drivers with four or more crashes compared to 100 with no crashes.33

The authors conclude:
It would appear that the driving hazards and the high accident record are simply one manifestation of a method of living that has been demonstrated in their personal lives. Truly it may be said that a man drives as he lives. If his personal life is marked by caution, tolerance, foresight, and consideration for others, then he would drive in the same manner. If his personal life is devoid of these desirable characteristics then his driving will be characterized by aggressiveness, and over a long period of time he will have a much higher accident rate than his stable companion.33(p 329)
    
Although the methodology of the 1949 study can be criticized on many counts, it was the first study to provide specific evidence of a strong link between broad personality characteristics and crash involvement, and introduced the concept a man drives as he lives. A major deficiency of the study is that much of the interpretation is based on psychiatric-type interviews conducted while riding in the taxis. This procedure is subjective in nature, and given the extreme differences between the groups of drivers, it was not possible for the interviewer to remain unaware of the group to which the driver most likely belonged, thus raising the possibility of bias. The comparison is between extremes (some of the drivers with high crash rates verged on the psychopathic), so it could be argued that the results may not necessarily be applicable to a more moderate degree of crash over involvement.
Psychiatric profiles of fatally injured drivers
    
An imaginative technique was applied to obtain psychiatric profiles of 25 deceased drivers judged to be at fault in the crashes in which whey were killed in Houston, Texas, from 1967-1968. The profiles were produced by conducting in-depth interviews with family members and associates of the deceased. These were compared to profiles of 25 control subjects selected from the same voter precincts in which the deceased had lived, and matched in such characteristics as age (all were males). Many criticisms of this study are possible. The information gathering processes were necessarily quite different for the deceased and control subjects, and the sample sizes were small. However, the differences found are much larger than any that appear likely to be due to possible biases in the technique. Personality disorders were associated with 75% of the fatally injured drivers, compared to 8% for the control sample. Even the few abnormal personalities among the control population were found to be less deviant, and to have more adequate coping mechanisms that helped compensate for their psychiatric liability.
Difficulties in researching the role of driver personality in crashes
    
Although personality factors play a central role, they are difficult to investigate. One problem is that crash rates depend strongly on factors, such as gender, age and alcohol use, so that any study of personality factors must control for these. This may require large sample sizes. Yet measures of personality are not available in large data sets, but must instead be determined by the investigators by, for example, administering one of many available personality instruments as components of questionnaire studies.
     Large compared to normal departures from the norm. The large effects of personality discussed above were measured by comparing extreme cases, such as taxi drivers with four or more crashes to those with zero, or fatally injured drivers to typical drivers. Since the comparison is between such extremes, it could be argued that the results may not be valid when interpolated to the more moderate degrees of over involvement that are of such importance in traffic safety. This problem is parallel to the dose-response problem in toxicology. Does a large, easily measured, deleterious effect associated with a massive dose of some substance support the inference that one tenth of the dose would still produce some deleterious effect, perhaps about one tenth the effect of the large dose? Or is there some threshold below which the substance produces no deleterious effect? Given that most crashes involve drivers not as far from the average as those in the studies discussed, it is important to know whether moderate variations within the normal ranges of behavior can explain variations in crash rates. As in the toxicology case, the smaller the dose, the more difficult it is to measure the response, but the more people are involved.
     Above-average crash risks are still small. There are methodological problems inherent in comparing those with moderately above-average crash rates to those with below-average crash rates. To illustrate, assume a hypothetical population of identical drivers each with the same average crash rate of 0.086 crashes per year used in Table 1-1. Column 4 of Table 1-1 (p. 15) shows that in a ten-year period 42% of the drivers have zero crashes, 36% one crash, and 21% two or more crashes. Suppose one compared driver characteristics of groups of safe (zero-crash) and risky (two or more crashes) drivers. Because membership of either group is determined by the pure randomness of a Poisson process, driver characteristics in each group are identical. Such a finding could be misinterpreted to conclude falsely that drivers who crash do not differ from those who do not crash.
     Now suppose there is a group of genuine risky drivers who are twice as likely as average drivers to crash (that is, their crash rate is 0.172 crashes per year) and a group of low risk drivers who are half as likely as average driver to crash (that is, their crash risk 0.043 crashes per year). In a ten-year, period 18% of the risky drivers will still be crash free, whereas 7% of the low risk drivers will have two or more crashes. Likewise, any group of subjects selected on the basis of having no observed crashes will contain many high risk drivers, and any group selected on the basis of having two or more observed crashes will contain some low risk drivers. The high and low risk groups of subjects will indeed contain drivers with higher and lower average propensities to crash, but the difference will be but a fraction of the real factor of four difference. This makes it more difficult to detect relationships based on comparing high and low risk drivers, and the effects measured in any relationships that are found will systematically underestimate the magnitude of the true effects by substantial amounts.
    There does not appear to be a way out of this dilemma. Statistical theory, together with various assumptions, can be used to estimate the most likely real effect from an observed difference, but as the observed difference is small even when the real difference is large, the estimate will lack precision. One cannot use very long periods to accumulate larger numbers of crashes because crash risk, which depends so strongly on age, is not stable over long periods.
Other studies on the relationship between personality and crash risk
    
Despite the difficulties, a large number of studies on personality factors, mainly using interview methods, have been performed. Reviews13, mention studies that report that higher risk drivers are less mature, less intellectually oriented, less academically successful, less interested in aesthetic matters, lower in aspiration, poorer in attitudes toward the law, and generally less well adjusted socially. They are more emotionally unstable, unhappy, asocial, anti-social, impulsive, and under stress. They are more likely to smoke and to have personality disorders and paranoid tendencies. They have less happy childhoods, a tendency to express open feelings of hostility, increased sensation seeking, low tension-tolerance, increased aggressiveness, and are more likely to seek prestige and social roles oriented towards authority and/or competition. They are more likely to have family histories, and current family relationships, reflecting higher degrees of disruption and conflict.
      More recent studies augment and extend these findings. Higher crash rates are related to high hostility in combination with poor self-esteem, high job stress, and self-reported tendencies to speed and to disregard traffic rules. Type-A personality drivers self-reported being involved in more crashes and displaying more aggression on the road than average drivers. Survey respondents reacting to behavior of other drivers in different scenarios were more likely to attribute hostility to male drivers. Given all the correlates that have been identified with over involvement in crashes, it is not surprising that over involvement in crashes relates to a whole host of other unhealthy behaviors, such as smoking. Credit histories have been shown to be good predictors of crash involvement. The auto-insurance industry is using the credit ratings available in large data files to set premiums.
Emotional stress. Personality denotes stable character traits that do not change over short time periods. Emotional stress may produce short or medium term departures from an individual's long-term average driving behavior. The risk of a driver being involved in a crash or violation is found to increase just prior to divorce proceedings. Drivers killed in crashes are more likely than control drivers to be undergoing periods of personal stress.34 A clustering of child pedestrian deaths around the time of the child's birthday is reported and attributed to the birthday excitement overriding the child's normal caution. It is not possible to investigate this further using FARS data because date of birth is not recorded in FARS, just age to the nearest year.
Non-transport motives
     
Transportation is not the only goal of driving. In an extreme example, racing drivers are not going anywhere, but are driving for a mix of pleasure, excitement, glory, and prize money. While regular driving has little in common with racing, it still includes motives in addition to utilitarian transportation. Most drivers find pleasure in driving fast and accelerating rapidly. We are all aware of the use of vehicles to show off, and to display competitive prowess in order to impress peers and attract members of the opposite sex. While the behavior of nearly all drivers is influenced at some times by non-transportation motives, this is particularly prevalent in young drivers, especially young male drivers. Such behavior is enshrined in our culture, and figures in many classic youth cult movies, such as Rebel Without a Cause and American Graffiti. Cars are used as an outlet for the independence, rebelliousness, and peer acceptance needs of newly licensed adolescents, a manifestation of a broader adolescent problem behavior syndrome. These non-transportation motives seem to be at the very core of the problem of traffic crashes, because it is risky behavior, especially speeding, that is the major determinant of the number of casualties. It is unlikely that adolescents and young men do much of their speeding in order to keep appointments with strict deadlines. They are motivated much more by the enjoyment and thrill of traveling fast, breaking the law, and defying society. The importance of these non-transportation motives in safety underlines that the problem is not lack of skill or knowledge, and countermeasures ignoring such effects are unlikely to be successful.
Suicide
    
The US recorded 23,458 male and 5,741 female suicides in 1999, giving an overall male-to-female ratio of 4.1 to 1. For ages 20-24 the ratio is 6.4, and declines systematically with age, reaching 3.1 for ages 45-54.
Given the large number of suicides, it would be remarkable if traffic crashes were not used for some of them, especially as use of a vehicle provides a convenient, undiscoverable, and honorable method of self-destruction. It minimizes guilt in those left behind, and avoids insurance and religious complications. Not only can crashes be used for premeditated suicides, but they may just happen to be available at the instant of a momentary, and perhaps otherwise temporary, impulse towards self-destruction. It is a near-perfect instrument with which to indulge the death instinct postulated by Freud.
     The possibility that some traffic crashes are suicides has been discussed for decades. A 1973 book, Accident or Suicide? Destruction by Automobile, offers a plausible narrative supporting a strong connection, but provides no convincing evidence that any specific traffic fatality was really a suicide. It is, of course, intrinsically difficult in most cases to determine definitively if a fatality is a suicide, so more indirect approaches are required.
A novel method examined time-series of suicides and crashes to see if traffic crashes increased after widespread reports of suicides of famous people. , An increase was observed, and attributed to suicides, based on research showing that media coverage of suicides leads to copycat increases in suicides in general. , Another empirical indication that some traffic fatalities are suicides was based on observed similarities in the day-to-day variations of suicides and traffic fatalities in the US.
     A more direct approach was possible using in-depth information available for all fatal traffic crashes in Finland. Based on preliminary identification by crash investigation teams, and further selection by two forensic pathologists, 84 drivers killed in Finland were classified as suicides. This led to the conclusion that suicide accounted for 5.9% of driver fatalities. Head-on crashes with heavier vehicles were more common than single-car crashes. In 4% of cases, the crash led to the death of another person. This study found traffic suicides to be strongly related to alcohol abuse, a factor that plays a major role in suicides in general. Further analyses of additional years of data from the same sources indicated that the percent of all traffic deaths that were suicides was increasing in time.
It is particularly difficult to use FARS data to analyze suicide because known suicide cases are excluded (p. 21).
Knowing the fraction of traffic fatalities attributable to suicide is important because most countermeasures are unlikely to have much influence on this component of traffic fatalities. The presence of suicides in data used to evaluate countermeasures will result in systematic underestimation of effectiveness for non-suicidal road users.
Family influence
    
It is well established that one of the largest factors determining whether young people smoke is if their parents smoke. The corresponding question of whether parents who crash have children who crash is more difficult to address empirically. Propensity to crash is estimated by a small number of crashes in a necessarily long time period, and the estimate departs from the true propensity because of randomness. People can be adequately characterized as smokers or non-smokers by simply asking them. The solid relationships for smoking suggest possible, but more difficult to determine, relationships for crashes. Two studies do indeed consistently find large effects for traffic crashes.
Driving records of young drivers (age 18-21) were compared to the records of their parents using the driver history file for North Carolina. It was found that children's driving records in the first few years of licensure are indeed related to the driving records of their parents. Children whose parents had three or more crashes on their record were 22% more likely to have had at least one crash compared with children whose parents had no crashes. Likewise, children whose parents had three or more violations were 38% more likely to have had a violation compared with children whose parents had none.
     A 1970 study found that sons of fathers with one or more traffic convictions (in a six-year period) were 58% more likely to have one or more traffic convictions than sons of fathers with no traffic convictions.
Earlier we showed that studies comparing drivers with many crashes to drivers with zero crashes systematically underestimate real effects because randomness will lead to some low risk drivers being included in the high risk category, and vice versa. This effect will be present also for violations, but because of the higher numbers of violations, it will have less influence. This interpretation is consistent with the 22% effect found for crashes compared to 38% for violations.53
The magnitudes of the effects found, even in the face of processes that inherently lead to their underestimation, suggest that parental influence has a large influence on a person's propensity to crash. Pursuing the smoking analogy suggests that the general social environment in which someone is embedded will strongly influence their tendency to crash. The main determinant of adult smoking is smoking as a teenager - once acquired, the behavior tends to persist. Once behaviors such as speeding, drunk driving, non-wearing of belts, or general risk-taking in traffic are acquired in youth, they will tend to persist. It seems likely that drivers who are more risky than average for their age when they are young will remain more risky than average for their age as they grow older, even though their absolute risks will decline as they age. I believe that the analogy with smoking is helpful for cases in which research on drivers is infeasible. Major progress has been achieved in smoking cessation, which further offers guidance for reducing crashes, as discussed in Chapter 13.
Crime rates and crash rates
    
There is much evidence that individuals who are involved in traffic crashes are more likely to commit crimes. A study conducted in the Netherlands obtained information on a random sample of 903 traffic crashes. From a separate file of criminal involvement, the researchers obtained the crime histories of the 1,181 male and 350 female drivers involved in the crashes (a male-to-female ratio of 3.4 to 1). The aggregate data showed that the drivers involved in crashes were more likely to have criminal records than the general public, by 31.0% compared to 15.2% for males and 11.4% compared to 3.5% for females. The traffic-crash file was examined to distinguish between drivers who displayed risky behavior prior to their crashes and those who did not. This was important since most of the crashes involved two vehicles. The drivers displaying risky behavior were more involved in various types of crimes than the other drivers - by factors of 2.6 for having a police record for violent crime, 2.5 for vandalism, 1.5 for property crime, and 5.3 for having been involved in a traffic crime.
     An examination of the driving records of 114 jailed criminals found that, compared to the general public, the criminals had 3.25 times as many citations for traffic violations, 5.5 times as many property damage and injury producing crashes, and 19.5 times as many involvements in fatal crashes. Some of the criminals in the sample may have been in jail for traffic offenses, which would lead to an overestimation of effects. A large sample of respondents to a survey in the Netherlands showed relatively strong relationships between crime and involvement in traffic crashes as non-drivers, including while walking and riding a bicycle.
The age dependence of involvement in crime has been described as follows:
     The propensity to commit criminal acts reaches a peak in the middle to late teens and then declines rapidly throughout life. Further, this distribution is characteristic of the age-crime relation regardless of sex, race, country, time, or offense. Indeed, the persistence of this relation across time and culture is phenomenal.
     Speeding, running red lights, tailgating, drunk driving, and not wearing safety belts are themselves criminal activities, and all correlate with crash-involvement rates. So it is perhaps not all that surprising that crash involvement relates to crime in a more general way. Research further suggests that more serious traffic offenses correlate with more serious non-traffic crimes.55-57
Gender differences in risk taking by babies and children
    
Some extremely rare events offer insights into how risk depends on gender from the earliest ages. Despite the rarity of the events, the 1.24 million traffic fatalities documented in FARS 1975-2002 provide usable samples.
Child "driver" fatalities
Table 9-5 shows the number of children in FARS 1975-2002 satisfying the following criteria:
- Child was the only occupant of a vehicle.
- Vehicle was moving when it crashed.
- Child was coded as the driver.
- Child was killed.
     One baby boy in the first year of life (age = 0) satisfied these conditions, no baby girls; for age 1, one boy, no girls. 16 boys aged 5 or under died as drivers compared to no girls. For drivers 10 and under, there were 761 boy fatalities compared to 69 girl fatalities, for a male-to-female ratio of 11 to 1. These risk ratios are larger than those encountered for adult risk taking. As age approaches the legal age of licensure, the male-to-female ratio declines to the values, still high, associated with young drivers.
    FARS data do not provide information about the circumstances of crashes. However, it seems plausible that babies, infants, and children were left alone in passenger seats of vehicles with engines running, or with ignition keys left inserted. It appears that in circumstances like these, boys were dramatically more likely than girls to venture into the driver seat and set the vehicle in motion. With increasing age it seems plausible that they were attempting to drive by copying observed adult behavior.
Gender and pedestrian fatalities
    
Figure 9-4 shows the number of male pedestrian fatalities per capita divided by the number of female pedestrian fatalities per capita in FARS 1975-2002. For ages less than 50 this ratio differs only slightly from the simple ratio of male pedestrian fatalities to female pedestrian fatalities. It is only after about age 50 that the number of females in the population begins to diverge appreciably from the number of males of the same age; by age 90 there are more than twice as many women as men.
    If the probability of becoming a pedestrian fatality were independent of gender, the data would distribute randomly around the dashed line indicating equal risk. The data depart dramatically from such a pattern. At every age in one-year increments, the male rate exceeds the female rate.
     The raw fatality data shown in Table 9-6 for the youngest ages show that the differences are clear and reliable from the earliest ages. In the first year of life, 106 baby boys died as pedestrians, compared to 75 girls, for a gender ratio of 1.41 ± 0.21. By age one, sample sizes increase substantially, leading to increasingly precise estimates that show pedestrian risk is unmistakably greater for male infants than for female infants. The ratios in Table 9-6 are similar to, but not identical to, the values plotted in Fig. 9-4 because the raw fatality numbers in the table do not include the normalization by population.
While the male risk is about 40% above female risk in the early years, the difference begins to soar to more than 200% at an age that looks suspiciously like the onset of puberty.
Crashes, crimes, and testosterone
The three graphs (Figs 9-5 to 9-7) show similar dependence on gender and age. Yet they plot quantities that are seemingly unrelated.

Figure 9-5. Estimated involvement in severe single-vehicle crashes per capita. Inferred from FARS 2002 data by same process that produced Fig. 7-8, p 155.

Figure 9-5 shows involvements in single-vehicle severe crashes per million population, calculated by applying Eqns 6-8 and 6-9 to FARS 2002 data in the same way as was done in Chapter 7. Single-vehicle crashes are selected because they depend on the actions of only one driver. Thus Fig. 9-5 reflects mainly the contribution of driver behavior, with the roles of other drivers and the dependence of survivability on gender and age removed.
Figure 9-6 has nothing to do with traffic. It shows the number of arrests per thousand population derived from FBI Uniform Crime Reports. Only arrests for crimes unrelated to driving are included. Figs 9-6 and 9-5 both show incidents per capita per year.
Figure 9-7 shows chemical measurements of testosterone in a large sample of people, as measured in a simple saliva analysis.
Safety and behavior as influenced by gender and age
    
The similarities between the graphs for crashes and crime suggest that they each share a common origin. The similarities between both these graphs and the one for the hormone testosterone suggest that the common origin is testosterone. Testosterone at the individual level also relates to the same pattern of personality characteristics found associated with high crash rates.60 Sensation-seeking individuals have high crash risks and high testosterone levels. The two small peaks for males in Fig. 9-7 occur around the midtrimester of pregnancy and a few months after birth. This may help explain the higher number of boy than girl pedestrian deaths in the first year of life, and the one boy driver compared to no girl drivers killed in the first year of life.
The three graphs add emphasis to a central theme in traffic safety, the dominant role of behavior rather than knowledge or skill. No one would suggest that the lower arrest rate for 40-year-olds compared to 20-year-olds occurs because the 40-year-olds have at long last learned how to not commit crimes. Similarly, lower crash rates for 40-year-old compared to 20-year-old drivers does not mean that 40-year-old drivers have simply learned how to not crash.
Criminality is not produced by insufficient knowledge about ethics or the law. Lack of knowledge about correct driving procedures is not the primary source of traffic crashes by young male drivers, which is why driver education has little effect on safety.
Gender differences and socialization. Large gender-dependent differences pervade traffic safety findings. Recall from Chapter 3 that among the 1.2 million worldwide annual traffic fatalities, 2.3 males are killed for every female killed.
It is sometimes claimed that these effects are the result of socialization processes that constrain the genders to fulfill assigned roles prescribed for them by society. It is unquestionably true that in every society males and females experience different socialization. In some societies the differences are sanctioned in religion, law, and custom, and accordingly exercise an overwhelming influence that is bound to generate large gender-dependent differences in all types of behavior. For example, a society that effectively prohibits women from driving is going to observe a vastly larger gender ratio for driver fatalities than any reported here. Even when gender equality is official policy, there can still be many subtle, unconscious, or unobserved ways in which females are treated differently from males, or are expected to behave differently.
      The findings for baby drivers and baby pedestrians consistently imply that gender-dependent behavior differences appear at the earliest ages - right from birth. It seems to me implausible that the large gender differences in baby-driver and baby-pedestrian fatalities could be due to socialization. It is difficult to imagine that baby boys would be left alone in vehicles or placed near traffic at rates different from girls to a degree that would explain the differences in fatality risks observed. The conclusion seems inescapable that at some fundamental level, right from birth, males and females behave differently in relation to risks in traffic. The testosterone measurements, which have nothing to do with socialization, leave little doubt that the differences in risk taking are in part innate rather than learned behavior. The magnitude of the male-to-female risk ratio is influenced by many social factors, but that it is greater than one appears to be an immutable fixture of the human condition.
Implications for interventions. In Chapter 7 we found that, when subject to the same physical impact, women were 28% more likely to die than men, a difference attributed to basic physiological differences between the genders. Here we find that the gender and age dependence of behavior central to traffic safety appears to originate, in large part, in basic hormonal differences. To realize the origins of these effects is to also realize that policy cannot make males as safe as females, or females as likely to survive crashes as males. Women live longer than men, but the goal of medicine is not to strive for equal longevity, but to increase everyone's lifespan. The goal of traffic safety is not the unattainable one of making risks equal for everyone, but the achievable one of reducing risks for everyone.

Summary and conclusions (see printed text)

References for Chapter 9 -  Numbers in [ ] refer to superscript references in book that do not correctly

show in this html version.  To see how they appear in book see the pdf version of Chapter 1.

[1] Näätänen R, Summala H. Road-User Behavior and Traffic Accidents. Amsterdam, Netherlands: North Holland; 1976.

[2] Cahill RA. Collisions and their Causes. London, UK: Nautical Institute; 1997.

[3] Häkkinen S. Traffic accidents and professional driver characteristics: A follow-up study. Accid Anal Prev. 1979; 11: 7-18.

[4] Murray W, Dubens E, Darby P. Virtual fleet risk manager. Napier University, Edinburgh, Scotland; 19 January 2004.

[5] Williams AF, O’Neill B. On-the-road driving records of licensed race drivers. Accid Anal Prev. 1974; 6: 263-270.[6] Kloeden CN, McLean AJ, Moore VM, Ponte G. Travelling speed and the risk of crash involvement. Volume 1– Findings. CR 172. Canberra, Australia: Federal Office of Road Safety; November 1997.[7] Solomon D. Accidents on main rural highways related to speed, driver and vehicle. Washington, DC: Department of Commerce & Bureau of Public Roads; July 1964.

[8] Kloeden CN, Ponte G, McLean AJ. Travelling speed and the risk of crash involvement on rural roads. Report No. CR 204 Department of Transport and Regional Services. Australian Transport Safety Bureau; July 2001.

[9] Nilsson G. Reduction in the speed limit from 100 km/h to 90 km/h during summer 1989: Effects on personal injury accidents, injured and speeds. Report No. 358A. Linkoping, Sweden: Swedish Road and Traffic Research Institute; 1990.

[10] Taylor MC, Baruya A, Kennedy JV. The relationship between speed and accidents on rural single-carriageway roads. TRL Report TRL511. Prepared for Road Safety Division, Department for Transport, Local Government and the Regions; 2002.

http://www.trl.co.uk/static/dtlr/pdfs/TRL511.pdf

[11] Finch DL, Kompfner P, Lockwood CR, Maycock G. Speed, speed limits and accidents. Project Report 58. Berkshire, UK: Transport Research Laboratory; 1994.

[12] Richter ED, Barach P, Friedman L, Krikler S, Israeli A. Raised speed limits, speed spillover, case fatality and road deaths in Israel: A five-year follow-up. Am J Public Health. 2004; 94: 568-74.

[13] Evans L. Traffic Safety and the Driver. New York, NY: Van Nostrand Reinhold; 1991.

[14] Insurance Institute for Highway Safety. Status Report, Special Issue: Speeding. Vol 38 (10); 22 Nov. 2003. http://www.iihs.org/srpdfs/sr3810.pdf

[15] Patterson TL, Frith WJ, Povey LJ, Keall MD. The effect of increasing rural Interstate speed limits in the United States. Traf Inj Prev. 2002; 3: 316-320.

http://www.ipenz.org.nz/test/ipenztg_cd/cd/2002_pdf/11_Patterson.pdf

[16] Joksch HC. Velocity change and fatality risk in a crash – a rule of thumb. Accid Anal Prev. 1993; 25: 103-104.[17] Evans L. Safety-belt effectiveness: The influence of crash severity and selective recruitment. Accid Anal Prev. 1996; 28: 423-433.

[18] Hyde F. The German Way. New York, NY: Passport Books (division of NTC/McGraw-Hill); 1997.

[19] International Road Traffic and Accident Database (OECD). Selected risk values for the year 2001. http://www.bast.de/htdocs/fachthemen/irtad/english/we2.html

[20] Cirillo JA. Interstate system accident research study II, interim report II. Public Roads. 1968; 35: 71-75.

[21] Hauer E. Accidents, overtaking and speed control. Accid Anal Prev. 1971; 3: 1-12.

[22] Cameron M. Estimation of the optimum speed on urban residential streets. Monash University Accident Research Centre. Report available from Australian Transport Safety Bureau, GPO Box 967, Canberra, ACT 2608, Australia; October 2000.

http://www.general.monash.edu.au/muarc/rptsum/optspeed.pdf

[23] Cameron M. Potential benefits and costs of speed changes on rural roads. Monash University Accident Research Centre. Report CR 216, available from Australian Transport Safety Bureau, GPO Box 967, Canberra, ACT 2608, Australia; October 2003.

http://www.atsb.gov.au/road/pdf/cr216.pdf

[24] Carlsson G. Cost-effectiveness of information campaigns and enforcement and the costs and benefits of speed changes. Paper presented to European Seminar in Luxembourg, Cost-effectiveness of road safety work and measures; 26 November 1997.

[25] All cars – cars are our life. http://www.allcars.org/top_lists_topspeed.html[26] Zingraff MT, Mason HM, Smith WR, Tomaskovic-Devey D, Warren P, McMurray HL, Fenlon RC. Evaluating North Carolina State Highway Patrol data: Citations, warnings, and searches in 1998. Report submitted to North Carolina Department of Crime Control & Public Safety and North Carolina State Highway Patrol; Nov. 2000.

http://www.nccrimecontrol.org/shp/ncshpreport.htm

[27] Baker SP, Braver ER, Chen LH, Pantula JF, Massie D. Motor vehicle occupant deaths among Hispanic and black children and teenagers. Arch Pediatr Adolesc Med. 1998; 152: 1209-12.

[28] Braver ER. Race, Hispanic origin, and socioeconomic status in relation to motor vehicle occupant death rates and risk factors among adults. Accid Anal Prev. 2003; 35: 295-309.

[29] Corbett C. Explanations for ‘understating’ in self-reported speeding behaviour. Trans Res Part F: Psych Behav. 2001; 4(2): 133-150.[30] Campbell M, Stradling SG. Factors influencing driver speed choices. Transport Research Institute, Napier University, Edinburgh, Scotland; 14 October 2003.

http://www.dft.gov.uk/stellent/groups/dft_rdsafety/documents/source/dft_rdsafety_source_024721.doc

[31] Evans L, Wasielewski P. Risky driving related to driver and vehicle characteristics. Accid Anal Prev. 1983; 15: 121-136.

[32] Wasielewski P. Speed as a measure of driver risk: Observed speeds versus driver and vehicle characteristics. Accid Anal Prev. 1984; 16: 89-103.

[33] Tillmann WA, Hobbs GE. The accident-prone automobile driver. Am J Psychiatry 1949; 106: 321-331.

[34] Finch, JR, Smith JP. Psychiatric and legal aspects of automobile fatalities. Springfield, IL: Charles C. Thomas; 1970.

[35] Shinar D. Psychology on the Road -- the Human Factor in Traffic Safety. New York, NY: John Wiley; 1978.

[36] Norris FH, Matthews BA, Riad JK. Characterological, situational, and behavioral risk factors for motor vehicle accidents: A prospective examination. Accid Anal Prev. 2000; 32: 505-15.

[37] Perry AR, Baldwin DA. Further evidence of associations of type A personality scores and driving-related attitudes and behaviors. Percept Mot Skills. 2000; 91: 147-54.

[38] Yagel D. Interpersonal antecedents of drivers’ aggression. Trans Res Part F: Psych Behav. 2001; 4: 119-131.[39] DiFranza JR, Winters TH, Goldberg RJ, Cirillo L, Biliouris T. The relationship of smoking to motor vehicle accidents and traffic violations. NY J Med. 1986; 86: 464-466.

[40] McMurray L. Emotional stress and driving performance: The effects of divorce. Behavioral Research in Highway Safety. 1970; 1: 100-114.

[41] Keeling JW, Golding J, Millier HKGR. Non-natural deaths in two health districts. Archives of Disease in Childhood. 1985; 60: 525-529.

[42] Jessor R. Risky driving and adolescent problem behavior: An extension of problem-behavior theory. Alcohol, Drugs, and Driving. 1987; 3: 1- 11.

[43] Anderson N. National vital statistics reports. Deaths: Leading Causes for 1999, Volume 49, Number 11, Center for Disease Control, October 2001. Table E. Deaths and percent of total deaths for leading causes of death by sex: United States.

http://www.cdc.gov/nchs/data/nvsr/nvsr49/nvsr49_11.pdf

[44] Tabachnick N. Accident or suicide? Destruction by Automobile. Springfield, IL: Charles C. Thomas; 1973.

[45] Philipps DP. Suicide, motor vehicle fatalities, and the mass media: Evidence towards a theory of suggestion. Am J Sociology. 1979; 84: 1150-74.

[46] Bollen KA, Phillips DP. Suicidal motor vehicle fatalities in Detroit: A replication. Am J Sociology 1981; 87: 404-412.

[47] Bollen KA, Phillips DP. Imitative suicides: A national study of the effects of television news stories. Am Sociological Review. 1982; 47: 802-808.

[48] Jamieson KH. Can suicide coverage lead to copycats? Am Editor. 2002; 824: 22-3.

[49] Bollen KA. Temporal variations in mortality: A comparison of U.S. suicides and motor vehicle fatalities, 1972-1976. Demography 1983; 20: 45-59.

[50] Ohberg A, Penttila A, Lonnqvist J. Driver suicides. Brit J Psychiatry. 1997; 171: 468-72.

[51] Hernetkoski K; Keskinen E. Self-destruction in Finnish Motor Traffic accidents in 1974-1992. Accid Anal Prev. 1998; 30: 697-704.

[52] Soteriades ES, DiFranza, JR. Parent’s socioeconomic status, adolescents’ disposable income, and adolescents’ smoking status in Massachusetts. Am J Pub Health. 2003; 93: 1155-1160.[53] Ferguson SA, Williams AF, Chapline JF, Reinfurt DW, De Leonardis DM. Relationship of parent driving records to the driving records of their children. Accid Anal Prev. 2001; 33: 229-34.

[54] Carlson WL, Klein D. Familial vs. institutional socialization of the young traffic offender. J Safety Res. 1970; 2: 13-25.

[55] Junger M, West R, Timman R. The relation between crime and risky behavior in traffic. J Res Crime and Delinquency. 2001; 38: 439-459.

[56] Haviland CV, Wiseman HAB. Criminals who drive. American Association for Automotive Medicine, 18th Annual Proceedings, Toronto, Ontario; 12-14 September 1974, p. 432-439.

[57] Junger M, Terlouw G-J, van der Heijden PGM. Crime, accidents and social control. Criminal Behaviour and Mental Health. 1995; 5: 386-410.

[58] Gottfredson M, Hirschi T. The true value of lambda would appear to be zero: An essay on career criminals, criminal careers, selective incapacitation, cohort studies and related topics. Criminology. 1986;24:213-234.

[59] US Department of Justice, Federal Bureau of Investigation, Uniform Crime Reports. Arrests for males and females at:

http://www.fbi.gov/ucr/cius_02/html/web/arrested/04-table39.html

http://www.fbi.gov/ucr/cius_02/html/web/arrested/04-table40.html

[60] Dabbs JM, Dabbs MG. Heroes, Rogues, and Lovers: Testosterone and Behavior. New York, NY: McGraw-Hill Professional; 2000.