Journal of Public Health Policy

, Volume 39, Issue 2, pp 193–202 | Cite as

Road death trend in the United States: implied effects of prevention

Original Article


This study estimates road deaths prevented by U.S. vehicle safety regulations, state laws, and other efforts based on comparison of actual deaths to those predicted from temperature and precipitation effects on exposure, migration to warmer areas, population growth, median age of the population, and vehicle mix. Logistic regression of risk factors predictive of road deaths in 1961, prior to the adoption of federal vehicle safety regulations, state behavioral change laws, and other preventive efforts were used to predict deaths in subsequent years given the changing prevalence of the risk factors from 1962 to 2015. The included risk factors are strong predictors of road death risk. Without the preventive efforts, an additional 5.8 million road deaths would likely have occurred in the U.S. from the initiation of federal safety standards for new vehicles in 1968 through 2015.


Road death trend Vehicle safety regulations Seat belt laws Alcohol laws Temperature Emergency medical response 


Relying on graphs of national trends in miles traveled and road deaths in the United States (U.S.), in 2001 Richter et al. inferred that declines in road deaths per miles driven are the result of slower speeds related to increased congestion and that attempts to reduce deaths were a “public health failure” [1]. In 2004, Evans inferred “a dramatic failure in U.S. safety policy” based solely on the fact that the decline in U.S. road deaths, uncorrected for miles driven, was less than in other high-income countries. He claimed that occurred because the U.S. emphasized vehicle regulation over changing driver behavior. [2].

Actually all wealthy countries adopted safety standards for vehicles within a few years of one another in the 1960s and 1970s and the U.S. federal government spent billions on grants to U.S. states for driver behavior change efforts, safer roads, and emergency response. Funds used in support of high school driver education increased teenage driving without reducing risk per mile driven and resulted in increased deaths until the driver education grants were stopped. Reduced deaths were associated with other uses of the funds granted to the states, such as increased traffic law enforcement, road modifications, etc. [3]. By 2000, U.S. states were spending about $5.6 billion per year on traffic law enforcement and other safety efforts of which about $1.9 billion came from the federal government [4].

Nevertheless, in 2012 Evans was allowed to publish a paper in the American Journal of Public Health updating the international trends, asserting again that the differences were a result of variations in policies among the countries but citing no specific policy differences or evidence of their effectiveness [5]. Evans’ claims were not supported by a study of U.S. states and 43 high-income countries with similar populations, matched into four groups by climate, population density, and urbanization. No differences were found in road deaths per vehicle miles traveled among the groups with similar characteristics [6].

Ignored by claimers of policy failure is research showing otherwise. Combined lap and shoulder belts reduce vehicle occupant deaths about 45% when used. Belt use increased from about 15% before laws requiring belt use were enacted in the 1980s to more than 88% recently, based on observed belt use by vehicle occupants [7], a remarkable change in behavior. Alcohol at 0.08% or above by weight in U.S. drivers in fatal crashes declined from 35% in 1982 to 20% in 1997, rose slightly then dipped to 20% again in 2005. This reduction was associated with changes in state laws such as the legal minimum drinking age and the lowering of the legally acceptable blood alcohol concentration [8]. Licensed teenaged drivers are involved in fatal crashes about 3.5 times as often per mile driven as 30–70 year olds. Reductions of 16–21% of 16-year-old drivers’ fatal crashes were associated with laws restricting the conditions under which 16 year olds could be licensed, the percent reduction dependent on the specifics of the restrictions [9]. Involvement of drivers of 16–19 years old in fatal crashes declined from about 50 per 100,000 people in that age group in the late 1970s to fewer than 20 per 100,000 in 2014, due to additional declines in licensure [10]. That was perhaps affected by discontinuance of driver education in many schools, and increasing insurance rates.

After a 40% reduction in fatalities per registered passenger car that met the initial federal safety standards required of new cars sold to the government in 1966 models and all cars in 1968 and subsequent models [11], manufacturers phased in air bags as standard equipment in new cars during the late 1980s as a result of a federal standard for increased frontal crash protection in new passenger cars. Manufacturers also installed side air bags and improved energy management of other vehicle components in numerous makes and models in subsequent years. Energy transfer criteria in the government’s vehicle crash tests steadily improved year-to-year through 2010 [12]. A study of 1999–2005 model years of passenger cars, minivans, and sports utility vehicles assessed the effects of each of various vehicle safety features, controlling statistically for the effects of the others [13].

Vehicles with electronic stability control had about 42% fewer fatalities in crashes per vehicle than vehicles without the system. Vehicles that had injury criteria above average on the government’s side crash tests had 19.4% fewer deaths. Those that scored best on the Insurance Institute for Highway Safety’s offset frontal crash tests had 8.6% fewer deaths. Electronic stability control installations increased from 29% in ‘light vehicles’ of the 2006 model year to 71% of the 2011 model year. All such vehicles, including pickup trucks, sold after 11 September 2011 are required by a federal safety standard to have electronic stability control.

A review of studies of alteration of the built environment to reduce pedestrian deaths indicates effects from 25 to 75% [14] but the extent that these measures have been deployed in the U.S. is unknown. States that implemented state-wide trauma response systems had about a 9% lower fatality rate, controlling for other factors, than those that did not [15].

The substantial percentage reductions in road deaths related to the noted efforts indicates policy success, but because of possible overlapping effects, they cannot be added together to assess their collective effect. For example, the correlations of death reductions from seat belt use and from driving while intoxicated are each reduced when considered simultaneously because intoxication is associated with lack of belt use [10].

Given the well-documented effects of specific laws, regulations, and other efforts, why was the decline in U.S. road deaths not steeper? Research has identified several major factors that would potentially offset preventive efforts to reduce road deaths in the U.S. Warmer temperatures are strongly related to increased road use and death rates [16, 17]. The temperature effect was likely accelerated by the mass migration to warmer areas of the country in recent decades [18]. The large growth in sales of pickup trucks and truck-based “sports utility vehicles” that are heavier than most passenger cars increased the deaths of occupants of other vehicles and of pedestrians substantially [19]. They have higher rollover death rates than passenger cars because of a high center of gravity relative to track width [20]. Motorcycles, pickup trucks, and busses are substantially more likely to be involved in pedestrian deaths per mile driven than passenger cars [21]. Motorcycles have an occupant death rate 34 times that of other vehicles per mile traveled [22]. Their registrations increased in the 1960s and 1970s, declined in the late 1980s and 1990s, and increased again through 2015. A factor that would increase road deaths in the 1970s and reduce them subsequently is the aging of the large cohort of people born in the decade post-World War II (so-called “baby boomers”) who were in their teens and twenties in the 1960s and 1970s, but were near retirement age in 2015.

The purpose of the research reported here is to estimate the number of road deaths that would have occurred during 1968–2015 given the correlation of deaths per population to temperature, precipitation, median age of the population, and vehicle mix among U.S. states in the early 1960s and the changing prevalence of these factors. The difference between predicted and actual deaths is an estimate of the overall effect of the various mentioned state and federal government prevention policies and other efforts.


The study is confined to the contiguous 48 U.S. states because weather data are not available for all the years in Alaska and Hawaii. State-wide data were used because data on vehicle registrations are not available for smaller geographical areas within states. The author obtained counts of road deaths per year in each state from Vital Statistics Mortality Tables [23] and the WISQARS search system (Web-based Injury Statistics Query and Reporting System) [24] for 1961–2015. He searched the online files of the Fatal Analysis Reporting System (FARS) from the time it became available [25] to count road deaths per month in order to relate them to average temperature per month. FARS includes road deaths only when a death occurred within 30 days of the crash. Road deaths counted in vital statistics are about 1–5% more than in FARS from year-to-year.

The author copied truck, motorcycle, and total vehicle registrations, estimated miles driven and linear miles of roadway by state from Highway Statistics for each year [26]. He obtained resident population counts per state and median age of the population from the U.S. Census and estimated the numbers in non-census years by linear extrapolation except for the 2015 estimates obtained from the U.S. Census Bureau’s American Community Survey [27]. He downloaded files of monthly average temperature and precipitation by state from the National Oceanic and Atmospheric Administration (NOAA) [28]. NOAA uses different numerical codes for states than the Census Bureau and the death files, so the author matched the data from the various files using a computer routine.

Because temperature varies somewhat among smaller geographic areas within states, depending mainly on elevation and aridity, some fidelity of measurement is lost using state averages. The weather stations that record the temperature and precipitation data are concentrated in populated areas within states. Thus extreme temperatures that occur in thinly populated mountainous and desert areas have less weight on the averages [29]. NOAA adds and deletes a small percentage of the 7000 monitored weather stations from time to time, creating controversy about whether trends are affected, but NOAA officials insist that trends in the data are unaffected [30].

The author examined the relationship of average monthly temperature and road deaths per population exposed to temperatures monthly in detail before using the state yearly averages as predictors. He used least-squares regression to estimate the relation of deaths per population exposed to specific average temperatures per month in 1975 and repeated in each year 1976–1980 to establish reliability of the coefficients.

One measure of exposure to death risk is miles traveled per vehicle, recently found to be related to temperature in urban U.S. counties [17]. In 1994, the Federal Highway Administration began reporting estimated miles traveled separately in U.S. states based on traffic count data at 4000 locations [31]. The author correlated miles traveled per registered vehicle in 1994 through 1999 to average annual temperature and precipitation using least-squares regression to examine reliability of the coefficients using state-level data.

The author employed logistic regression to assess the association of deaths per population with average annual temperature, total annual precipitation, median age of the population, registered vehicles per population, and percent of registered vehicles that were motorcycles, trucks, or busses, using data for each of the 48 states in 1961. He affirmed the reliability of the coefficients by noting the fit of predicted deaths to actual deaths in 1962 through 1965 before the noted prevention efforts were adopted. He used the regression coefficients to predict the number of deaths that would occur given temperature and precipitation averages and migration to warmer areas, as well as vehicles per population and vehicle mix separately in each state through 2015 [32]. He compared the sum of predicted deaths across the states to the actual number of deaths nationally.


Changes in the risk factors during the period studied were remarkable. Nationally, population increased from 178 million to 314 million (76.4%) and vehicles per 100 population increased from 45 to 94 during 1961–2015. Percent of trucks among all registered vehicles increased from 19 to 55 during the study period and percent of busses declined from 0.4 to 0.3%. Percent of motorcycles of registered vehicles rose from 0.8 to near 4% during 1961–1984, declined to 2% in the mid-1990s, and rose to 3.7% in 2015. Average temperature among the states rose about 2° (F) from 1961 to 2015 with a few percent year-to-year variations about the trend line in interim years. Precipitation fluctuated but did not show much trend. The major effect of temperature is due to the migration of populations from cooler to warmer states. For example, the 8 states bordering Mexico and the Gulf of Mexico gained 252.7% in population (41.2–104.1 million) from 1960 to 2015 while the 13 states bordering Canada gained only 31.4% (59.8–78.6 million).

The association of road deaths per million person months of exposure to given monthly average temperatures is shown in Fig. 1. Each point represents the sum of deaths that occurred in a month in any state where the average temperature was at specified degrees divided by the sum of the population in the state every time the temperature was at the same degrees. The data in the graph are similar to those found when deaths on specific days were divided by the number of persons exposed to the temperature on a given day in the previous research of the 100 most populous counties in the U.S. [16]. The average death rate per exposed population at average temperatures near 80° (F) is about twice the risk at temperatures in the teens and the rate increases linearly, on average, as temperature increases. Table 1 shows the regression coefficient of the association seen in Fig. 1 compared to that in each of the subsequent 5 years. The 1975 coefficient is the second to the lowest and the 1976 coefficient is an outlier, but R2 is very strong in each year. Therefore, there does not appear to be much loss in fidelity of the correlation of temperature and road death rate per population using the average of monthly temperatures in each state.
Fig. 1

Deaths Per Million Person Months At Each Average Monthly Temperature: 48 Contiguous U.S. States, 1975

Table 1

Least-squares regression coefficients and 95% CI of the relation of road deaths per million person months and average temperature per month, 48 contiguous U.S. States



Temperature coefficient

95% CI

R 2































The regression of vehicle miles traveled per vehicle in relation to average annual temperature during the first 6 years that miles were measured more accurately shows consistently more miles driven per vehicle registered at warmer temperatures (Table 2). For each higher degree (F) of temperature, vehicles were driven about 110 more miles per year with some year-to-year variation. The relatively low R2 may be due to the small sample of sites per state to measure mileage. The results suggest that discretionary driving is more frequent in states with warmer temperatures.
Table 2

Least-squares regression coefficients and 95% CI of the relation of vehicle miles traveled per vehicle and average annual temperature, 48 contiguous U.S. States



Temperature coefficient

95% CI

R 2































Table 3 displays the logistic regression coefficients and 95% confidence intervals derived from the analysis of the noted risk factors among the states in 1961. Each of the predictor variables is strongly correlated to death rates per population among the states.
Table 3

Logistic regression coefficients relating road deaths per population to temperature, precipitation vehicle types, and median age of state populations, 48 contiguous U.S. States, 1961



Lower 95% CI

Upper 95% CI

Per degree (F) of average monthly temperature




Per inch of precipitation

− .00386

− .00508

− .00264





Percent motorcycles of registered vehicles




Percent trucks of registered vehicles




Percent busses of registered vehicles




Median age of the population

− .0228

− .0287

− .0169


− 9.0688


When a proxy for congestion (registered vehicles per linear miles of road surface in each state) was included in the predictor variables, the equation did not accurately predict the deaths in 1962–1966. The proxy was correlated with median age of the population (R = 0.48) and inversely correlated with percent trucks of registered vehicles (R = − 0.67). In 1961, older populations lived in states with higher congestion and trucks were more prevalent in states with low congestion. Because such collinearity distorts regression coefficients, congestion was dropped from the equation. Number of vehicles per population reflects congestion to a degree but the correlation with risk of death is positive, not negative, as claimed by the congestion notion.

The equation in Table 3 predicted slightly fewer deaths than actually occurred in 1962–1966, indicative of a conservative estimate of effects, but predicted deaths increased rapidly relative to actual deaths in subsequent decades except during the late 1980s and 1990s when the baby boomers were leaving their higher risk years and motorcycle registrations declined dramatically (Fig. 2). By 2015, the deaths expected without interventions were 8.8 times the actual number of deaths. The model predicted about 8 million deaths during 1968–2015 but 2.2 million actually occurred, suggesting that about 5.8 million road deaths were avoided by preventive efforts.
Fig. 2

Sum of Expected and Actual Road Deaths: 48 Contiguous U.S. States, 1961–2015


Contrary to the claims of failure of U.S. road safety laws, vehicle safety regulations, and other preventive efforts [1, 2, 5], these results suggest that they averted a public health disaster. Deaths would likely have been in the hundreds of thousands rather than tens of thousands per year in the twenty-first century without the demonstrably effective federal and state policies and other efforts mentioned in the introduction.

This estimate does not account for factors unmeasured here that are known to affect risk. In the 1960s, vehicle manufacturers engaged in a ‘horsepower war,’ competing to make and more powerful engines, which may explain why deaths during that period exceed the predictions. Although the driver error rate probably did not change much, if any, during the study period, distraction from portable cell phones and other electronic devices that increases risk when used while driving grew in numbers rapidly in the 1990s and early 2000s resulting in increased risk [33]. Lower gasoline prices correlated to increases in road deaths during 1983–2000 [34]. Historic gasoline prices are not available for each state, so they could not be included in this analysis. In 2015 dollars, the national average price per gallon was $1.83 in 1965 and $2.45 in 2015 with large fluctuations in the interim [35]. U.S. gas prices per gallon are less than half of those in most other wealthy nations [36], an additional reason for differences in road death rates among countries. Opioid drug use increased substantially in recent decades as indicated by rapidly growing overdose deaths, but a large case–control study found no significant effect of various drugs other than alcohol on driver crash risk when other risk factors were controlled statistically [37].

The claim that congestion explains all of the declines in road deaths [1] is not supported by the data in this study. While a decline in road deaths per miles traveled occurred before as well as after the mentioned prevention efforts, that fact is not a sufficient basis for concluding that the prevention efforts were ineffective. Many studies not referenced by the perpetrators of that inference indicate otherwise. The results of this study indicate that the factors that influence the trend in death rates differ in prevalence at different points in time and no single factor, such as congestion, can explain the entire trend.

The inverse correlation of precipitation and deaths per population suggests the probability that some people, especially pedestrians, pedal cyclists, and motorcyclists without a protective shell, also avoid being on roads when the weather is adverse, more than offsetting the increased risk of slick roads [38]. This finding about precipitation is opposite to that noted in the study of urban counties during 2014 [17]. While seasonality of precipitation is fairly consistent in some sections of the U.S., it is quite variable in others [39]. Observation of frequency of road use by type of users under different temperature and precipitation conditions at rural and urban locations would shed light on reasons for the difference. The finding that road use with accompanying deaths increases as temperature rise is consistent between the studies and suggests that continued warming of the atmosphere related to the emission of greenhouse gasses from motor vehicles and other sources [40] does not portend well for the future of road safety.



Thanks to Jennifer Brady and Claudia Tebaldi of Climate Central for their help in identifying the appropriate temperature and precipitation files.


  1. 1.
    Richter ED, Barach P, Ben-Michael E, Berman T. Death and injury from motor vehicle crashes: a public health failure, not an achievement. Inj Prev. 2001;7:176–8.CrossRefGoogle Scholar
  2. 2.
    Evans L. Traffic safety. Bloomfield Hills: Science Serving Society; 2004. p. 381–411.Google Scholar
  3. 3.
    Robertson LS. Federal funds and state motor vehicle deaths. J Public Health Policy. 1984;5:376.CrossRefGoogle Scholar
  4. 4.
    Federal Highway Administration. Our Nation’s Highways. 2000. Accessed 16 Jan 2018.
  5. 5.
    Evans L. Traffic fatality reductions: United States compared with 25 other countries. Am J Public Health. 2012;104:1501.CrossRefGoogle Scholar
  6. 6.
    Kahane CJ. Comparison of 2013 VMT fatality rates in U.S. states and in high income countries. Washington, DC: National Highway Traffic Safety Administration Report Number HS 812 340, 2016.Google Scholar
  7. 7.
    National Highway Traffic Safety Administration. Seat belt use in 2015—Use in the states and territories. Traffic Safety Facts. 2016. Accessed 29 July 2017.
  8. 8.
    Dang JN. Statistical analysis of alcohol-related driving trends. Washington DC: National Highway Traffic Safety Administration Technical Report #DOT HS 810 942, 2008.Google Scholar
  9. 9.
    Chen L, Baker SP, Li G. Graduated driver license programs and fatal crashes of 16 year-old drivers: a national evaluation. Pediatrics. 2006;118:56–62.CrossRefGoogle Scholar
  10. 10.
    Robertson LS. Injury epidemiology. 4th ed. Raleigh: Lulu Press; 2015.Google Scholar
  11. 11.
    Robertson LS. Automobile safety regulations: rebuttal and new data. Am J Public Health. 1984;74:1390–4.CrossRefGoogle Scholar
  12. 12.
    National Highway Traffic Safety Administration. Trends in HIC, chest G, and injury risk, NHTSA crash tests 1979–2010, 2015.Google Scholar
  13. 13.
    Robertson LS. Prevention of motor vehicle deaths by changing vehicle factors. Inj Prev. 2007;13:307.CrossRefGoogle Scholar
  14. 14.
    Retting RA, Ferguson SA, McCartt AT. A review of evidence-based traffic engineering measures designed to reduce pedestrian-motor vehicle crashes. Am J Public Health. 2003;93:1456.CrossRefGoogle Scholar
  15. 15.
    Nathens AB, Jurkovich GJ, Rivara FP, Maier RV. Effectiveness of state trauma systems in reducing injury-related mortality: a national evaluation. J Trauma Infect Crit Care. 2000;48:25.CrossRefGoogle Scholar
  16. 16.
    Leard B, Roth K. Weather, traffic accidents, and exposure to climate change. Resources for the future working paper, 2016. Accessed 28 Aug 2017.
  17. 17.
    Robertson LS. Climate change, weather and road deaths. Inj Prev. Accessed 31 Aug 2017.Google Scholar
  18. 18.
    Cebula RJ, Nair-Reichert U, Coombs CK. Total state in-migration rates and public policy in the United States: a comparative analysis of the great recession and the pre- and post-great recession years. Reg Stud. 2014;1:1.Google Scholar
  19. 19.
    White MJ. The “Arms Race” on American roads: the effect of sport utility vehicles and pickup trucks on traffic safety. J Law Econ. 2004;47:333.CrossRefGoogle Scholar
  20. 20.
    Snyder RG, McDole TL, Ladd WM, Minahan DJ. On-road crash experience of utility vehicles. Ann Arbor: University of Michigan Highway Safety Research Center; 1980.Google Scholar
  21. 21.
    Paulozzi LJ. United States pedestrian fatality rates by vehicle type. Inj Prev. 2005;11:232.CrossRefGoogle Scholar
  22. 22.
    Lin M, Kraus JF. A review of risk factors and patterns of motorcycle injuries. Accid Anal Prev. 2009;41:710.CrossRefGoogle Scholar
  23. 23.
    National Center for Health Statistics. Vital Statistics of the United States—Mortality. Washington DC: U.S. Department of Health and Human Services, 1961–1998.Google Scholar
  24. 24.
    Centers for Disease Control and Prevention. WISQUARS. 2017. Accessed 30 Nov 2017.
  25. 25.
    National Highway Traffic Safety Administration.. Fatality analysis reporting system. 1974–2015. 2017. Accessed 28 Feb 2017.
  26. 26.
    Federal Highway Administration. Highway statistics. Washington DC: U.S. Department of Transportation, 1961–2015.Google Scholar
  27. 27.
    U.S. Census. 2015 American Community Survey. 2017. Accessed 13 Jun 2017.
  28. 28.
    National Oceanic and Atmospheric Administration. 2017.; files climdiv-pcpnst-v1.0.0-20170705 and climdiv-tmpcst-v1.0.0-20170705. Accessed 20 Jun 2017.
  29. 29.
  30. 30.
    United States Government Accountability Office. 2011. Climate Monitoring. U.S. Congress.Google Scholar
  31. 31.
    Federal Highway Administration. Travel monitoring. 2017. Accessed 22 July 2017.
  32. 32.
    Selvin S. Statistical analysis of epidemiologic data. New York NY: Oxford University Press; 1991. p. 205–6.Google Scholar
  33. 33.
    Wilson FA, Simpson JP. Trends in fatalities from distracted driving in the United States, 1999–2008. Am J Public Health. 2010;100:2213.CrossRefGoogle Scholar
  34. 34.
    Grabowski DC, Morrissey MA. Gasoline prices and motor vehicle fatalities. J Policy Anal Manag. 2004;23:575.CrossRefGoogle Scholar
  35. 35.
  36. 36.
  37. 37.
    Compton RP, Berning A. Drug and alcohol crash risk. Traffic Safety Facts. National Highway Traffic Safety Administration. 2017. Accessed 20 Sept 2017.
  38. 38.
    Eisenberg D, Warner KE. Effects of snowfalls on motor vehicle collisions, injuries and fatalities. Am J Public Health. 2005;95:120–4.CrossRefGoogle Scholar
  39. 39.
    Finkelstein PL, Truppi LE. Spatial distribution of precipitation seasonality in the United States. J Clim. 1991;4:373–85.CrossRefGoogle Scholar
  40. 40.
    Tebaldi C, Adams-Smith D, Kenwood A. Warming winters: U.S. temperature trends. 2017. Accessed 15 Aug 2017.

Copyright information

© Macmillan Publishers Ltd., part of Springer Nature 2018

Authors and Affiliations

  1. 1.Epidemiology and Public HealthYale UniversityGreen ValleyUSA

Personalised recommendations