Article Text

Download PDFPDF
Climate change, weather and road deaths
  1. Leon Robertson
  1. Department of Epidemiology and Public Health (retired), Yale University School of Public Health, Green Valley, Arizona, USA
  1. Correspondence to Dr Leon Robertson, 3551 S. Via de la Grulla Green Valley, AZ 85622, USA; nanlee252000{at}yahoo.com

Abstract

In 2015, a 7% increase in road deaths per population in the USA reversed the 35-year downward trend. Here I test the hypothesis that weather influenced the change in trend. I used linear regression to estimate the effect of temperature and precipitation on miles driven per capita in urbanizedurbanised areas of the USA during 2010. I matched date and county of death with temperature on that date and number of people exposed to that temperature to calculate the risk per persons exposed to specific temperatures. I employed logistic regression analysis of temperature, precipitation and other risk factors prevalent in 2014 to project expected deaths in 2015 among the 100 most populous counties in the USA. Comparison of actual and projected deaths provided an estimate of deaths expected without the temperature increase.

  • motor vehicle occupant
  • motorcycle
  • pedestrian
  • bicycle
  • public health

Statistics from Altmetric.com

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Introduction

Road deaths per population in the USA declined about 50% during the 35 years through 2014. The reduction was mainly associated with the imposition of Federal Safety Standards for new vehicles sold in the USA as well as vehicle crashworthiness and crash avoidance features introduced voluntarily by manufacturers1, safety belt use laws2, laws addressing maximum legal blood alcohol concentration of drivers with increased sanctions for driving with concentrations above the maximum,3 minimum alcohol purchase age laws4 and teenage graduated licensure requirements.5

In 2015, the road death trend in the USA abruptly reversed—an increase of 7% in deaths compared with 2014. Some safety officials attributed the reversal to increased cell phone use while driving.6 While cell phone use is associated with an increased risk of severe crashes, cell phone ownership increased gradually and approached saturation among the US population several years before 2015.7 A national probability sample of drivers observed in traffic during 2014 and 2015 found no increase in use of all hand-held devices and a slight non-significant decrease in hand-held cell phone use from year to year.8 Use of such devices while driving likely dampened the rate of reduction in death rates in recent decades but it is unlikely to account for the abrupt change in road deaths in 2014–2015. Here I report an analysis of an alternative hypothesis.

Anecdotal reports of road deaths that occur during periods of ice and snow imply that colder climates increase risk. But seasonal differences in fatalities suggest the opposite. In the USA, there are hundreds more deaths per month in summer compared with winter.9 The per cent increase in deaths from 2014 to 2015 was disproportionately among road users unprotected by a vehicle shell—bicyclists (13%), pedestrians (9.5 %) and motorcyclists (8%). Slick roads undoubtedly increase the probability of collisions10, but weather also influences the extent and types of exposure to risk of road deaths—driving, walking or bicycling.

I hypothesised that warming of the atmosphere increases vehicle miles driven and fatality rates per population, other things being equal, and the unusually large increase in average temperature in 2015 contributed to the reversal in the trend of road deaths per population.

Methods

I assembled two data sets—one to assess the association of temperature and precipitation on annual miles driven per person in urban areas of the USA and the other to measure the relationship of weather and other factors to death risk per population at the county level. The Federal Highway Administration reports miles driven in urbanised areas in the USA periodically11 but the urbanised areas covered are not the exact same geographic areas as those designated as urban by the U.S. Census Bureau. State governments determine which extra census tracts to include. The Fatal Analysis Reporting System that I used to count road fatalities identifies location of the deaths by county but not census tract so it is not possible to match exactly mileage with deaths at the local level.

The report of 2014 data on miles driven uses 2010 population numbers that may have changed substantially among some of the areas in the interim between 2010 and 2014. Therefore, I used the 2010 data on miles driven and population size to estimate miles driven per capita in relation to weather for ‘urbanised areas’ that had populations of ≥300 000.

I obtained average annual temperature (F) and inches of precipitation in 2010 from local weather stations posted in the archives of Weatherunderground.com.12 The weather data for each area are usually collected at a local airport but in some cases at another local weather station. I merged these data with the mileage and population data and used least-squares regression to estimate the average miles driven per person associated with temperature and precipitation.

In the second data set, I included data on all road deaths in each of the 100 most populated counties in the USA during each year 2014–2015. These deaths I counted from the online files of the Fatal Analysis Reporting System, a national census of motor vehicle fatalities collected by the National Highway Traffic Safety Administration13 that includes state and county codes for each death. Average daily temperatures (°F) and average daily precipitation in each year for each area separately in 2014 and 2015 I included for each county. I obtained population in each county during each year, 2014–2015, from the American Community Survey conducted by the U.S. Census Bureau.14

To specify whether risk of death per population is related to average daily temperature, I matched the state, county and date of death with the average temperature on that date. I divided number of deaths on days of a given temperature by the total population exposed to the temperature on that day to get a death rate per person year.

Larger counties generally have more rural roads that have increased risk of fatalities.15 Higher incomes increase the probability that members of a household drive later model vehicles that have more safety features.16 Lower speed limits reduce death risk.17

To control statistically for differences among the counties in the prevalence of these risk factors, I obtained proxy measures of the variables. The Census Bureau website provided the area of each county in square kilometres and median income per household.18 I gathered speed limits on urban freeways in each state where the county is located from the Insurance Institute for Highway Safety website.19

I estimated the contribution of each of the mentioned factors to the variation in odds of road death per population among the counties during 2014 using logistic regression. I then used the regression coefficients to calculate the expected deaths in each county as a function of average annual temperatures and precipitation in 2015 and compared these to the actual number of deaths to assess the reliability of the coefficients derived in the cross-sectional analysis.

Results

Table 1 presents the regression coefficients and 95% CIs for the association of temperature and precipitation with annual miles driven per vehicle during 2010. Vehicles were driven an average of about 60 more miles per year per person for each increase in a degree (F) of temperature. They were driven an average 66 more miles per year per person for each one inch increase in precipitation. Temperature and precipitation were not significantly correlated (r=−0.06) so the regression coefficients were not distorted by collinearity.

Table 1

Least-squares regression estimates of the increment in annual average miles per capita associated with temperature and precipitation in urbanised areas of the USA with populations of ≥300 000, 2010

Figure 1 displays the correlation of roads deaths per billion person-days of exposure to the average temperature on the days of the deaths. On average, the higher the temperature, the higher the death rate (r2=0.51, p<0.001). The larger variance at the extremes of high and low temperature days is due at least partly to the relative rarity of those temperatures and thus lower numbers in the denominators.

Figure 1

Deaths per billion person-days, 2014.

The logistic regression coefficients on the hypothesised predictors of fatality risk among the 100 counties in 2014 are displayed in table 2. Deaths per population are higher in warmer areas. They are also higher in counties with more precipitation, higher speed limits on urban freeways and larger land area. Counties with more median income per household have lower death risk.

Table 2

Logistic regression estimates of the association of average annual temperature, total precipitation, state speed laws, area size and median family income to road fatality risk among the 100 most populous counties in the USA, 2014

I tested the reliability of the equation by applying the 2014-based logistic regression coefficients to temperature and precipitation data in 2015, leaving speed limits, county area and median income unchanged. The distribution of predicted deaths and actual deaths among the counties in 2015 were not significantly different (χ2=13.8, df=99, p<0.99), indicative of a good fit of the data.20

The average annual temperature among the 100 counties increased 1.5°F from 2014 to 2015. The 100 counties experienced 11 459 deaths in 2014 and 12 175 in 2015, a 6.2% increase similar to the 7% total increase nationally. To assess how much of that increase was associated with temperature change, I used the regression model to predict the number of deaths in 2015 if there were no temperature changes but included the precipitation in 2015. The sum of the expected deaths among the counties in 2015 if there were no temperature changes was 11 575, 5.2% below the actual number of deaths. Comparison of the expected deaths if there were no temperature changes and actual deaths among the counties indicated that the differences would not likely have occurred by chance (χ2=3135, df=99, p<0.001).

I also note that deaths on freeways can be reduced by lowered speed limits. I calculated the expected deaths in 2015 if all urban freeways in the counties studied had a 55-mile per hour speed limit, the lowest among the counties studied. The total is 10 169, a reduction of 20% from the actual number even as temperatures increased as much as they did.

Discussion

Most of the increase in fatalities from 2014 to 2015 is likely related to road users’ activities encouraged by warmer temperatures. The finding that vehicles are driven an average 60 more miles per person for each degree increase in temperature may not seem like much but applied to the sum of the total populations in the areas included, the increase is substantial. There were 152 115 894 people in the urbanised areas included during 2010. Assuming that the findings in 2010 are generalisable to the later years, the 1.5° average temperature increase that occurred from 2014 to 2015 would add almost 13.6 billion miles of travel in those areas (152 115 894×59.52×1.5) assuming no population growth.

And that does not include the likelihood of increased walking and bicycling in warmer periods. Although this study is limited by the lack of data on the extent to which people walk and use bicycles in relation to temperature, it is plausible that they do so in periods of greater warmth. A review of the literature on physical activities concluded that people are less active during adverse weather.21 The increased miles associated with increased precipitation may reflect more use of motor vehicles rather than walking during periods of rain or snow.

The strong correlation of inter-day, inter-year and inter-county variation in temperature does not mean that the well-known risk factors for road deaths are less than thought. But for those risk factors to be in play, people must be out on the roads. It is possible that changes in other risk factors during 2014–2015 influenced the increase in road deaths. For such changes to account for the change associated with temperature increase, they would have to be strongly associated with temperature changes day by day, year by year and county by county, an implausible scenario.

Loosening of prohibitions on marijuana use for medicine or recreation and the increased use of prescription and non-prescription opioids, reflected in increasing deaths from overdoses, may increase road injury risk. Only 5 of the 100 counties in this study were in states that allowed recreational use of marijuana during the study period—too few to include a variable for such laws in the analysis and too few to have an influence of any consequence on the findings. A large case–control study of the risk of vehicle crashes in relation to drugs measured in drivers found only small RRs that were not statistically significant when adjusted for other risk factors.22

Reductions in vehicle occupant death rates should continue to occur as older, less safe vehicles are scrapped. The magnitude of the 2015 temperature increase was unusual for a single year but few scientists question the evidence that indicates continued warming of the atmosphere for the foreseeable future from the combustion of fossil fuels.23 The evaporative effect of warming oceans increases precipitation but where and when it will fall is difficult to predict.24 Since increased vehicle miles result in increased CO2 emissions, the effect of temperature on miles driven suggests a classic feedback system—more miles, more CO2, more warming, more miles.

Conclusion

As temperatures continue to increase from heat-trapping gases in the atmosphere, road deaths will likely increase more than expected unless there are major mitigating countermeasures. Numerous extensively researched preventive measures to reduce road deaths and severe injuries are available.25

What is already known on the subject

Slick roads increase vehicle crash risk.

What this study adds

Warmer weather and precipitation are independently related to miles driven. More road deaths occur on warmer or wetter days. The reversal in the downward trend in road deaths per population in the USA is mainly associated with an unusual rise in temperature from 2014 to 2015.

References

Footnotes

  • Competing interests None declared.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data sharing statement A table showing the lack of correlation of temperature to environmental risk factors is available as mentioned in the manuscript.