by Brian Smith, Kevin Vo, and Kel Yip
In 2016 the Department of Transportation released a harrowing statistic: 2,348 more people died on US roads in 2015 than the year before. This was the biggest single year jump since 1966, and in response the DOT released a call-to-action to help understand why 2015 was such a deadly year. As part of the Code for San Francisco Brigade we thought we might be able to contribute to this effort.
At first we searched for a root cause, like more people driving while distracted or higher temperatures, but after looking at a lot of factors we couldn’t find any smoking guns. Instead of assuming that there is a cause and trying to find it, we tried the opposite approach: assume that the roads were just as safe as the year before and try to disprove that. In the end we weren’t able to find any evidence that the roads were fundamentally less safe than they have been since 2009.
It may seem obvious that the roads became more dangerous in 2015 since we know a lot more people died, but because the economy was recovering and gas was cheap people were also driving a lot more. More miles driven means more accidents, even if the roads are just as safe per mile driven. When you account for total miles driven 2015 was still an increase over 2014, but it was a lower number of fatalities per mile than it was in 2012.
While eyeballing a chart is a good place to start, given the seriousness of the situation we also wanted to use more rigorous approaches. We applied two different statistical techniques to test if 2015 was an abnormally deadly year or fit in line with the overall trend since 2009. We also wanted to apply these techniques to some of the types of fatalities that increased the most in 2015, such as bicyclists and multi-vehicle crashes.
The first method looked at the overall correlation between fatalities and miles driven between 2009 and 2014 for each state and then compared that to the correlation in 2015. There was a high correlation between 2015 and earlier years for all the groups we looked at except for accidents involving under 16 drivers, which there are very few data points for. Below is a visual representation of this analysis, and you can find a full explanation of this method here.
The second method looked at whether 2015 was an abnormal near for fatalities per mile driven compared to the rates over the previous 5 years. It did this using an interrupted time series to establish a baseline rate from 2009 to 2014 for each state and then checked if 2015 was an outlier compared to that baseline. This wasn’t able to show that 2015 was a statistically significant outlier for any of the variables we looked at either. You can find a more information about this analysis here.
Overall we unable to find evidence that the number of traffic fatalities in 2015 was abnormally high once you take miles driven into account. We’re hoping to incorporate the 2016 data when it’s released so we can see if this result continues to be the case. It’s a little disheartening to see that traffic safety hasn’t been getting better since 2009, but at least it’s not getting worse. Hopefully as newer safety technologies such as automatic braking receive significant penetration the roads will start getting safer again.