The unformatted htm text below is from

Copyright 2004 by Leonard Evans

Information about

How to purchase

Table of Contents



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 

2 Data sources

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.

Quantification is at the core of scientific understanding, and quantification requires data. Few subjects provide more data than traffic crashes. Nearly all the world's countries record and classify crashes. Since the beginning of motorization, this has resulted in the collection of information on about a billion traffic crashes. Despite so much data, many questions remain unanswered because factors of interest have not been recorded or the data are not sufficiently reliable, complete, or conveniently accessible.
     A characteristic common to all data sets is that they include only crashes that meet specified criteria. The data set therefore is only a subset of the total reality. It follows that even reliable inferences from data sets do not necessarily provide useful information about real-world phenomena. For example, it is straightforward to estimate the most common crash severity in any data set which records crash severity. However, it is incorrect to conclude that this is also the most common crash severity. The most common crash severity for real crashes is just marginally above zero, but such low severity crashes do not get recorded in data sets.
     In general, the more serious the outcome of a traffic crash, the more likely the crash is well documented in data sets. Below we discuss different sources of data, starting with data on the most severe crashes.
Most of the literature on traffic safety, including this book, has a strong emphasis on fatalities. Not only are fatalities the most serious and permanent consequence of traffic crashes, but fatality data are vastly more reliable and readily interpretable than data for any other level of harm. However, fatality data are still subject to uncertainties. As was noted in discussing the Titanic, even a simple count may contain mistakes of both omission and incorrect inclusion.
What is a traffic fatality?
The definition of a traffic fatality is far from simple. The problems are readily illustrated by the definition used in the Fatality Analysis Reporting System (FARS). This data set defines a traffic fatality as a person who dies within 30 days of a crash on a US public road involving a vehicle with an engine, the death being the result of the crash. If a driver has a non-fatal heart attack that leads to a crash that causes death, this is a traffic fatality. However, if the heart attack causes death prior to the crash, then this is not a traffic fatality. If a victim dies many days after a crash, a difficult judgment may be required to decide whether it is a traffic fatality. For example, a frail person may die from pneumonia during hospitalization to treat crash trauma. As we all have some chance of dying at any moment, some people die within 30 days of even the most minor crash.
     The 30-day inclusion criterion is by no means universal. The National Safety Council uses one year. Their estimate of total US traffic fatalities typically exceeds the FARS total by about 4%, suggesting that a similar percent of crash victims die between one and 12 months after their crashes. As a victim may die from crash injuries decades after a crash, no feasible selection criterion can guarantee complete inclusion. The choice must be a compromise between completeness and timeliness. With a 30-day criterion, the complete data for (say) 2000 includes events through 30 January 2001. The administrative tasks required to produce the data set prevent it from being available until much later. Thus results derived by analyzing data cannot provide direct information about what is currently happening, only what happened in the past.
Fatality Analysis Reporting System (FARS)
   The FARS1 data set that provides many of the results in this book is maintained by the National Highway Traffic Safety Administration (NHTSA), part of the US Department of Transportation. It is a census of all US fatal crashes occurring since 1 January 1975. The FARS system now provides information on more than a million Americans killed in traffic, another reminder of the enormous harm from traffic crashes. The information is based mainly on police completing forms providing details in three categories:
            1. Crash (date, time, roadway category, etc.)
            2. Vehicles (number involved, types, model year, etc.)
            3. People (age, gender, alcohol use of involved persons including pedestrians and
                                all occupants of all vehicles, belt use by vehicle occupants, etc.)
     Each crash has more than 100 data elements coded by FARS analysts who may make some judgments based on the information available. For most fatally injured drivers, Blood Alcohol Concentration (BAC) is measured in an autopsy. In seeking mainly objective data elements, FARS lacks information on many factors of interest. The speed limit on the road on which the crash occurred is noted, but the police officer has no way to know what speed vehicles were traveling prior to the crash, or the impact speed. Likewise, fault is not indicated. Some variables are of uncertain reliability, including belt use of surviving occupants, which is based largely on what they tell police.
     A strange and needless deficiency in FARS is that cases in which deliberate intent, such as suicide, can be definitively identified are excluded. Thus FARS abandons, for no good reason, the goal of being a census of those killed in traffic. By excluding some small but unknown percent of traffic suicides it makes the file less useful for investigating traffic suicides. All traffic deaths should have been coded, and if deliberate intent was suspected or confirmed, this should have been noted in an additional data element. Hopefully FARS can correct this deficiency. Indirect methods applied to Finnish data indicate that as many as 5.9% of traffic deaths may be suicides (p. 225).
Some characteristics of fatal crashes
     Table 2-1 shows basic information from FARS for 2002. Logically, the number of fatal crashes cannot be larger than the number of fatalities or the number of involved vehicles. A crash in which anyone is killed is a rare event among crashes, just as a crash is a rare event among trips. A crash in which more than one person is killed is a rare event among fatal crashes. 9% of the fatal crashes killed more than one person. 13% of the two-vehicle fatal crashes killed more than one person. 22,086 of the fatal crashes, or 57.7%, were single-vehicle crashes. 23,639 of those killed, or 55.4%, were killed in single-vehicle crashes.

Table 2-1. US fatal crashes in 2002.1

     Note that the majority of people involved in fatal crashes are not themselves killed. For example, consider a car with four occupants crashing into two pedestrians, killing one and injuring the other. Assuming no car occupants are killed, this crash will have six involved people, one being killed. On the other hand, 13,339 of the fatal crashes (35.0%) involved only one person - the unaccompanied driver of a vehicle striking a fixed object, overturning, etc. without impacting any other person or vehicle. These crashes account for 13,339/42,815 = 31.3% of all fatalities.
     Although not part of the FARS data set, it is estimated by other methods that about 65 fetuses are killed annually in traffic crashes.
Non-fatal injuries
While fatal injuries conceptually involve only a yes or no determination, non-fatal injuries lie along a severity continuum, from minor scratches to near death, and apply to different regions of the body. The question "How many injuries?" has little meaning in the absence of some defined level of injury. Generally, the less severe the injury, the more frequently it occurs, so the total number of injuries increases steeply as the threshold for inclusion is lowered.
The Abbreviated Injury Scale (AIS)
Because categorizing injuries is so complex, no single classification coding scheme has achieved universal acceptance. The Abbreviated Injury Scale (AIS), developed by the Association for the Advancement of Automotive Medicine, is the most widely used and accepted scale. The AIS classifies injuries by body part, specific lesion, and severity on a 6-point scale in terms of the threat to life of a single injury. The scale is ordinal, meaning that an AIS 2 injury is greater than an AIS 1 injury, but is in no sense twice as great. It is therefore formally incorrect to apply arithmetical operations to sets of AIS values. The AIS level is determined by comparing injuries diagnosed by a physician to those listed in the detailed documentation that defines the scale. Other scales are also used, such as the International Classification of Diseases, which includes injuries as a subset of all illnesses. There is ongoing research to compare results from different scales in order to expand and improve them.
The AIS level is determined soon after the crash, and not by final outcome. As a consequence, it is possible for injuries at any AIS level to prove fatal later, although the observed risk of death increases steeply with increasing AIS level, as illustrated in Table 2-2. The values given for probability of death are not part of the AIS level definitions, nor are they expected to be closely replicated in general. They are the observed values in one study and are presented to indicate more clearly how the potential for injuries to be life-threatening increases with increasing AIS. If an examination uncovers no injury at a level matching any on the scale, this is recorded as AIS = 0. The scale is based exclusively on medical criteria - it does not reflect how the same injury can generate vastly different degrees of impairment for different individuals. An AIS 3 injury to a finger may have little effect on the life of a singer, but may end the career of a violinist.
Injury victims often sustain injuries to more than one body region. For many analyses it is convenient to use only one measure of injury severity, which is the maximum AIS, or MAIS. A victim with three injured regions of the body, all at AIS 1, would have a MAIS of 1; a victim with one region injured at AIS 1 and one at AIS 2 would have a MAIS of 2.
KABCO classification
The AIS is known for very few cases, because it requires a physician to examine the victim and then input the findings into the appropriate data file. A simple classification that has proved useful for many studies is the "KABCO" scheme, where K = killed, A = incapacitating injury, B = non-incapacitating injury, C = possible injury, and O = no injury. These classifications can be made at the crash scene by police officers, leading to their inclusion in large data files. KABCO values are coded for 99% of the 100,968 people included in FARS for 2001. A comparison of KABCO and AIS coding for the same crashes found that 49% of those coded as A ( = incapacitating injury) had no more than minor injuries (AIS = 0 or 1).
Disability-adjusted and quality-adjusted life year (DALY and QALY)
The disability-adjusted life year (DALY) is a measure that reflects the total amount of healthy life lost to all causes, whether from premature mortality or from some degree of disability during a period of time. One DALY is defined as one lost year of healthy life due to premature death or disability. While the World Health Organization estimates that traffic fatalities will be the sixth leading cause of death in the world in 2020, they are expected to be the third leading cause of DALYs lost.
Another metric often used in benefit-cost studies is the quality-adjusted life year (QALY). A year in perfect health is considered equal to 1.0 QALY. The value of a year in ill health would be discounted. For example, a year bedridden might have a value equal to 0.5 QALY.
Number of US injuries at different injury severities
Deciding how to code injuries is only one step along the way to answering such questions as how many injuries of a specified severity occur. There are too many injuries for all of them to be documented in detail, as is done for fatalities. Instead, a number of data sets have adopted sampling schemes. In most cases, the more severe the crash and the injuries it produces, the more likely it is to be selected for the expensive scrutiny necessary to determine injury levels and vehicle damage. Data sets including useful estimates of injuries include the Crashworthiness Data System (CDS), the General Estimates System, and the National Automotive Sampling System (NASS), and others.
The National Automotive Sampling System Crashworthiness Data System (NASS CDS) is a stratified probability sample of all US crashes involving a passenger vehicle that required towing due to damage. The probability that a crash is included depends strongly on crash characteristics. For example, the more severe the crash the more likely it is to be included (otherwise the system would be swamped with minor crashes). This makes the raw data unsuitable for most studies. Instead, the sampled crashes are scaled up to national estimates based on the structure of the sampling protocol. Such a process necessarily injects substantial additional uncertainty.
The estimates in Table 2-3 showing the numbers of people injured at different levels in the US in 2000 were obtained by synthesizing information from different data sets.14 Fatalities are included as a separate category to avoid double counting. Totals for all MAIS levels, especially the higher levels, would be larger if those who subsequently died were kept with the MAIS totals. Over 5 million injuries are estimated, with the vast majority being minor (MAIS = 1).

Table 2-3. The number of people suffering different levels of injury from traffic crashes in the US in 2000

The monetary cost of US traffic crashes
The estimates of total monetary cost in Table 2-4 include cost of lost productivity, medical costs, legal and court costs, emergency service costs, insurance administration costs, travel delay, property damage, and workplace losses. Converting all costs into the one metric of dollars has important advantages, but can be done only by invoking many assumptions. The authors of the report stress that economic costs represent only one aspect of the consequences of motor vehicle crashes, and do not reflect the pain, loss of function, disfiguration, emotional stress, and other suffering to the victims and immediate families.14 The lifetime economic cost to society for each fatality is estimated at just under a million dollars, over 80 percent of which is attributable to lost workplace and household productivity. The difficulties in estimating the cost of a fatality is succinctly captured in an essay titled: And how much for your grandmother? (p 245) The fact that most victims are young crucially affects the estimated cost of a fatality, which must necessarily be based on many assumptions.

Table 2-4. Monetary cost of motor-vehicle crashes in the US in 2000 (in billions of 2000 US dollars).

     The largest dollar cost is property damage. This includes property damage from all crashes, including those also involving injury. The largest contribution is from the 13.5 million non-injury crashes. The $231 billion cost of motor vehicle crashes represents $820 for each person in the US, and is 2.3 percent of the US gross domestic product.
Crash severity - damage to vehicles
     FARS contains a variable extent of deformation in four categories; none, minor, moderate and severe, based on visual inspection of the vehicle by police. Even though it is coded for 98% of the vehicles in FARS, it tends to be of limited use because more than 90% of the vehicles in which the driver is killed are, as one would expect, coded as having severe deformation.
     Ideally, one would like to know in detail how the forces on occupants varied in time during crashes to better understand how crash and vehicle characteristics influence injuries. Such information is available only for anthropometric dummies in instrumented laboratory crash tests. An overall measure that
relates to forces during a crash is the change in vehicle speed due to the crash (delta-v or Dv). A vehicle traveling at, say 60 km/h, crashing head-on into an immovable barrier would have Dv = 60 km/h. If it crashed into a stationary car of similar mass, each vehicle would have Dv = 30 km/h. It is found that injury outcomes in real crashes are related to delta-v. Such relationships arise only because the time during which the crash occurs is relatively similar for different crashes. Arguably, a safely landing airliner has Dv = 800 km/h. However,
this change from cruising speed to stationary takes place over 20 minutes, imparting minimal forces on occupants. A delta-v of 60 km/h occurring in 70 ms generates an average deceleration of 238 m/s2, or 24 times that due gravity (often written 24 G's) with associated potentially lethal forces. Falling from the same height onto a cushion or onto concrete produces the same delta-v. The cushion causes the delta-v to occur over a longer time, thereby reducing injury forces.
     The data plotted in Fig. 2-1 are for crashed vehicles with unbelted drivers, derived from weighted NASS data. The number of crashed vehicles increases very steeply with decreasing delta-v, reaching a peak at Dv just under 20 km/h. There have been many comments to the effect that the most common crash delta-v is some value, say about 20 km/h. This is not so. The peak in the top graph in Fig. 2-1 is a characteristic of the data set, not a characteristic of crashes. There are compelling reasons to believe that more crashes occur with Dv in the range 0-1 km/h than occur in the range 1-2 km/h, and so on, with the number of crashes increasing systematically with decreasing severity. At below about 20 km/h, the probability that a crash is recorded in the data set declines reaching essentially zero for Dv = 1 km/h, thus producing the observed pattern in the recorded data. The straight line fitted to the main body of the data in the top plot in Fig. 2-1 estimates about 9 times as many crashes in the range 0-1 km/h as in the range 19-20 km/h.
     The middle plot shows how the risk of death and severe injury increases with delta-v. At high values of delta-v there are few cases, which leads to the noisy pattern.
     The bottom plot is the number of crashes times the probability that the crash causes a severe injury or death. The peak values here are real and have important implications for occupant protection, because when different occupant protection devices are adjusted to protect best at a particular severity, their effectiveness will be less at other severities. The goal is not to provide the greatest protection for the greatest number of crashes (low Dv), nor is it to provide the greatest protection where risk is highest (high Dv), but to provide the greatest protection at the value of Dv at which the most harm occurs. This value depends on balancing large numbers of minor injuries against small numbers of more severe injuries and fatalities.

How reliable are injury reports?
The omission of large numbers of low-severity crashes from the data used to produce the top plot of Fig. 2-1 is a feature built into the data set - only crashes above a specified level of severity were supposed to be included (indeed, they were all tow-away crashes). The missing cases were not supposed to be included. Because almost no injuries are expected in even very large numbers of sub-threshold crashes, their omission is of no material importance. Real problems do arise when cases that should be included are omitted, and when cases are included when they should not be.
It is unlikely that examining the content of a data set can reveal missing values, or plausible entries that should not have been included. However, a number of different types of investigations shed light on the reported numbers
of injuries.
Fatalities compared to reported injuries from Irish data
Fig. 2-2 shows the number of traffic fatalities per million population versus road user age for Northern Ireland and for the Republic of Ireland for 1990-1992. Northern Ireland, which is a province of the much larger United Kingdom, and the Republic of Ireland, an independent nation, share the same small island of Ireland, not always amicably. As physical environment, climate, vehicles, and general human behavior are similar in the two jurisdictions, it is not surprising that fatalities show similar characteristics. However, reported injuries do not, as indicated by 1991 data for each jurisdiction:
Northern Ireland: 6.9 reported injuries per thousand population.
Republic of Ireland: 2.7 reported injuries per thousand population.
The authors of the report providing the data comment on the dramatic difference in reported injury rates compared to the lack of difference in fatality rates as follows:
The most likely solution to this conundrum is that reporting practices are very different in the two jurisdictions, with "minor" injuries likely to go unrecorded in the Republic but to be "over-reported" in the North.17
The more generous British social welfare benefits available in Northern Ireland provided monetary compensation for genuine injuries. However, the same benefits were available for reporting injuries even if none occurred. Such benefits being less available in the Republic at the time covered by the study may have led some real injuries to go unreported, because those injured did not feel it worth the time and trouble to report them.
Another finding was that even at the height of the political violence in Northern Ireland in the early 1990s, traffic fatalities still remained the major cause of sudden violent death.17
     The term "whiplash" refers to injuries associated with occupants' heads moving rearward relative to their bodies when vehicles in which they are traveling are struck in the rear by other vehicles. Late whiplash syndrome refers to symptoms that persist, or arise, long after the crash. Unquestionably many injuries occur in rear-impact crashes, many of which cause major pain and disability. Such injuries can be difficult to diagnose by objective medical tests, so patients' reports of neck pain are often the only basis of diagnosis.
     There are innumerable published estimates of more than a million whiplash injuries in the US each year, with some estimates being as high as 4 million. The total monetary cost is estimated to be 29 billion dollars per year. For Western Europe over a million whiplash injuries are reported, and estimated to cost 8 billion euros a year.
     It is common knowledge in the US and Western Europe that a reported whiplash injury can lead to monetary compensation. It is likewise well known that a rear-impact crash has a very high probability of being followed by claims of whiplash injury. The expectation that such injuries are a near inevitable consequence of a rear-impact crash may generate genuine symptoms that, absent such expectation, might not occur.
     How widespread would reports of whiplash injuries be if people did not expect to suffer them after rear-impact crashes, or could not receive payment for claiming symptoms? This question was addressed by two studies using similar methodology conducted in Lithuania. In Lithuania, few car drivers and passengers were covered by insurance, and there was little awareness among the general public about the potentially disabling consequences of a whiplash injury.
     In the first study, 202 occupants of cars that had been struck in the rear were interviewed 1-3 years after their crashes. A control group of 202 individuals matched in age and gender who had not been involved in any type of traffic crash completed the same questionnaire. Members of the study and control groups were asked to report symptoms associated with whiplash, with the results summarized in Table 2-5. The authors report that no one in the study group claimed disabling or persistent symptoms as a result of the crash.

Table 2-5. Comparison of reported whiplash symptoms by occupants of cars struck in the rear 1-3 years earlier to people not involved in traffic crashes.

     The second study used 210 subjects in cars struck in the rear, and 210 crash-free subjects matched in age and gender. Unlike the earlier investigation, study subjects were mailed questionnaires soon after the crash to obtain information about short-term effects. Follow up questionnaires were sent to the study subjects two months after their crashes, and one year after their crashes. A follow up questionnaire was sent to the control subjects a year after they were first identified. The results are summarized in Table 2-6. The authors conclude:
In a country were there is no preconceived notion of chronic pain arising from rear end collisions, and thus no fear of long term disability, and usually no involvement of the therapeutic community, insurance companies, or litigation, symptoms after an acute whiplash injury are self limiting, brief, and do not seem to evolve to the so-called late whiplash syndrome.22

Table 2-6. Comparison of reported whiplash symptoms by occupants of cars struck in the rear and respondents not involved in traffic crashes.

     NHTSA estimates that about 1.5 million vehicles are struck in the rear annually in the US. The more than a million reported cases of whiplash injury implies that a rear-end crash has about a 67% chance of generating a reported whiplash injury, so that samples of over 200 occupants struck in the rear would be expected to produce about 130 cases of whiplash. The data in Tables 2-5 and 2-6 convincingly reject any possibility that whiplash injuries are nearly that common. In fact, there are no more than minor differences between the self-reported symptoms of occupants of vehicles struck in the rear and people not involved in any type of traffic crash. The conclusion is inescapably clear. It is insurance compensation and litigation that is responsible for most of the whiplash injuries reported in the US and Western Europe, not crash forces.
Injuries per fatality
     In Canada from 1970 to 2001 the number of traffic fatalities decreased by 45%, but the number of injuries increased by 24%. A number of explanations have been offered to explain this dramatic contrast. These include the suggestion that occupant protection has made enormous strides in preventing fatalities, but not in preventing injuries. This is unconvincing. There is no reason to suppose that measures that reduce the forces on the human body in a crash will particularly alter the distribution of injuries by severity. All injury levels are expected to decline by comparable proportions. Such evidence as there is suggests occupant protection improvements will reduce injury risk more than fatality risk. For example, safety belts are probably more effective at preventing injuries than fatalities (p 283). Another suggestion is that improved trauma care reduces fatalities, but an injury remains an injury even if given better medical treatment. This is qualitatively correct. But, as more than half of fatalities in FARS 2002 died within an hour of their crashes, the quantitative effect of improved trauma care, while an important contributor, cannot explain more than a small portion of the enormous divergence between the fatality and injury trends.
     There are general reasons why the ratio of injuries to fatalities is expected to be fairly robust, and to not depend much on country, safety policy (for example, belt wearing laws) or vehicle design, and to change only gradually in time. Even if vehicle factors did somehow influence the ratio, the effect from year to year could be no more than a percent or so, because 90% of the vehicles on the road in a given year are the same vehicles that were on the road in the previous year.
The data in Fig. 2-3 defy any plausible interpretation in terms of engineering or medical factors. The number of injuries per fatality should be similar in Canada and Britain, and change only slowly, and similarly, in each country. What Fig. 2-3 appears to be reflecting is not changes in the risk of injury, but changes in the probability that an injury is reported. The reporting probability depends on politics, medical policy, insurance policy, and law, all factors that can change quickly, and differ from country to country, and from era to era.
     In Britain in the Second World War years 1942-1944 there were 20 reported injuries per fatality, compared to 34 in the pre-war years 1935-1938. After the war the number of reported injuries per fatality increased, but stabilized at close to 50 during the prolonged period from 1950 and 1970. This period was just after the introduction of the National Health Service in 1948. Everyone requesting health care received it free of cost, but opportunities for additional compensation for injuries were generally unavailable. Beyond the 1970s, opportunities for monetary compensation expanded. Fig. 2-3 shows a marked increasing trend in the number of reported injuries per fatality after 1970.
In Canada in the 1960s, when medical care was largely paid directly out of patients' pockets, the number of reported injuries per fatality was substantially lower than in Britain. However, it later increased rapidly as Canadian provinces moved more in the direction of public payment for medical care, and later opportunities expanded for compensation in addition to medical care.
     There are two effects that can make the number of reported injuries increase even if the actual number of injuries remains constant. First, in the past injuries occurred but were not reported. Direct out-of-pocket expenditures discourage reporting. Because of increased emphasis on health care, someone suffering a

Figure 2-3. The number of reported injuries divided by the number of reported fatalities in Canada and Great Britain. Data from Refs. 25 and .

cut, scratch, or bump today is more likely to seek medical care than in the past even if cost is not a consideration.
     The second way that reported injuries might depart from actual injuries is through injuries being reported when none is present. Providing rewards for reporting injuries encourages such behavior. Transport Canada defines injuries to "include all those who suffered any visible injury or complained of pain" (bold added).
A broader message
Data from a number of countries and sources show consistently that reported injuries can depart from actual injuries by large systematic amounts. This finding teaches two principles important to traffic safety. First, clear effects observed in data sets do not necessarily imply real phenomena, but may instead be due to data selection and definition. The second principle is more universal, and is well understood by economists, but often ignored, or even hotly denied, by others. The principle is that as the cost of an activity increases, less of it occurs, while as the reward for an activity increases, more of it occurs. The empirical data show that this principle explains variations in reported injuries per fatality. The same principle applies to many traffic safety topics. If the cost of crashing increases, fewer crashes occur. If the cost decreases, say, because of insurance, more crashes occur. Any policy that increases the cost of drunk driving, such as increased alcohol taxes, reduces drunk driving.

Summary and conclusions (see printed text)

References for Chapter 2 - 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 see pdf version of Chapter 1 or pdf version of Chapter 16.

[1] Fatality Analysis Reporting System (FARS) Web-Based Encyclopedia. Data files and procedures to analyze them at

[2] National Safety Council. Injury Facts (prior to 1999 called Accident Facts). Itasca, IL: published annually.

[3] Tessmer JM. FARS analytic reference guide 1975 to 2002. Washington, DC: National Highway Traffic Safety Administration, Department of Transportation.

[4] Weiss HB, Songer TJ, Fabio A. Fetal deaths related to maternal injury. J Am Medical Assoc. 2001; 286: 1863-1868.

[5] Association for the Advancement of Automotive Medicine. The abbreviated injury scale. AAAM; 1990.

[6] National Center for Health Statistics. International Classification of Diseases, Ninth Revision, Clinical Modification, Sixth Edition.

[7] Malliaris AC, Hitchcock R, Hedlund J. A search for priorities in crash protection. SAE paper 820242. Warrendale, PA: Society of Automotive Engineers; 1982.

[8] Farmer CM. Reliability of police-reported information for determining crash and injury severity. Traf Inj Prev. 2003; 4: 38-44.

[9] Homedes N. The disability-adjusted life year (DALY) definition, measurement and potential use. Worldbank Human Capital Development and Operations Policy working paper.

[10] World Heath Organization. The Injury Chartbook. Geneva; 2002.

[11] Graham JD, Thompson KM, Goldie SJ, Segui-Gomez M, Weinstein MC. The cost-effectiveness of airbags by seating position. J Am Medical Assoc. 1997; 278: 14181425.

[12] National Center for Statistics and Analysis (NCSA), National Highway Traffic Safety Administration, US Department of Transportation, Washington DC.

[13] Blincoe LJ, Faigin BM. The Economic Cost of Motor Vehicle Crashes. Report DOT HS 807 876. Washington, DC: National Highway Traffic Safety Administration, US Department of Transportation; 1990.

[14] Blincoe LJ, Seay AG, Zaloshnja E, Miller TR, Romano EO, Luchter S, Spicer RS. The economic impact of motor vehicle crashes, 2000. Report DOT HS 809 446. Washington, DC: National Highway Traffic Safety Administration, US Department of Transportation; May 2002.

[15] Adams J. Transportation Planning Vision And Practice. London, UK: Routledge and Kegan Paul, 1981.[16] Evans L. Safety-belt effectiveness: The influence of crash severity and selective recruitment. Accid Anal Prev. 1996; 28: 423-433.

[17] Leslie JC, Rooney F. Psychological factors in road traffic accidents: Statistical evidence and a study of the effects of viewing an anti-speeding film. Irish J Psychol. 1996; 17: 35-47.

[18] Whiplash and temporomandibular disorders: Medico-legal issues.

[19] Schmid P. Whiplash-associated disorders. Notfallzentrum, Chirurgie, Inselspital Bern.

[20] Whiplash prevention: A major car safety issue. .

[21] Schrader H, Obelieniene D, Bovim G, Surkiene D, Mickeviciene D, Miseviciene I, Sand T. Natural evolution of late whiplash syndrome outside the medicolegal context. Lancet 1996; 347: 1207-1211.

[22] Obelieniene D, Schrader H, Bovim G, Miseviciene I, Sand T. Pain after whiplash: A prospective controlled inception cohort study. J Neurol Neurosurg Psychiatry. 1999; 66: 279-83.

[23] NHTSA Performance Specifications: IVHS countermeasures for rear-end collisions.

[24] Department for Transport. Transport Statistics, Table 9.10 Road accidents and casualties: 1950‑2002.[25] Transport Canada. Canadian motor vehicle traffic collision statistics: 2001, footnote 4.