What do the data say about traffic safety? It’s complicated.
by Rik Williams, Data Scientist, Policy Research & Economics, and Nadia I Anderson, Ph.D., Manager, Public Policy, Road & Traffic Safety
In October, some good news emerged: US road fatalities declined in 2017 and in the first half of 2018, according to data compiled by the National Highway Traffic Safety Administration (NHTSA). This is a welcome development, especially since there had been a troubling uptick in fatalities the two years prior. Nonetheless, motor vehicle crashes continue to present a major public health concern in the US, representing the leading cause of death in 2015 for children and young adults (and among the top 10 causes for all age groups under 65).
Understanding the factors that can increase (or mitigate) crashes is key for taking steps to reduce traffic fatalities, and experts generally agree on what works in a broad sense: engineering, education, enforcement and emergency response. Ridesharing is a small but growing part of the transportation system, and Uber is interested in understanding how — and doing our part — to make our roads and cities safer.
Despite what we already know about the big picture of improving road safety, we still have a lot to learn about the details in order to make evidence-based interventions; universities, governments, and private organizations therefore dedicate substantial resources toward understanding this subject, with NHTSA alone budgeting nearly $300 million for vehicle and highway safety research in 2016. Yet even when a research question may seem simple and intuitive (like whether access to ridesharing reduces the general public’s rate of driving under the influence of alcohol), the results aren’t always clear-cut. Why is this?
In the following, we’ll look at some of the challenges that face researchers trying to understand the underlying causes of (and solutions to) traffic fatalities. But first, let’s take a step back and look at the historic trend in US road deaths:
As cars and driving became more prevalent in the early 20th century, fatalities dramatically increased, but then declined overall since the 1970s. This isn’t because people stopped driving, but is generally attributed to improvements in vehicle safety features, road design, and legal frameworks: think seat belts, maximum speed limits, and implementation and enforcement of drinking and driving laws.
However, on top of these general trends there’s a great deal of short-term variability; as annotated on the graph, several sharp decreases coincide with economic shocks. When the economy shrinks or even grows at a slower rate, for a number of reasons, fatalities go down. A simple explanation for this may be that when economic activity is lower, people drive less.
In order to account for year-to-year variations in miles driven, researchers typically consider the fatality rate as opposed to total fatalities; in other words, the number of fatalities per total number of vehicle miles traveled (VMT). As the graph above shows, this rate has also broadly declined over the past 25 years both for overall fatalities and for those involving alcohol, even as shorter-term increases and decreases are still apparent. In other words, variability still remains even when VMT is taken into account, suggesting that the simple, “people drive less” explanation doesn’t fully explain the observed trends.
If we wanted to further refine this analysis to determine whether a single factor — like the presence of ridesharing or a change in a state’s law — is behind one of these short-term variations, the task at hand becomes far more complicated. Indeed, several studies on the subject of road safety have attempted to answer questions like these and found mixed results. What’s going on here? Below, we outline three primary reasons why this isn’t as easy as it might seem.
Controlling for all factors
The US transportation system is massive and complex, involving infrastructure, drivers, vehicles, and a variety of laws and regulations that govern the transportation sector; all of these components (and countless sub-components) impact traffic safety. A 1% decrease in traffic fatalities means hundreds of lives saved, so it’s crucial to understand how even minor factors influence the system as a whole. However, in order to measure the impact of one such component, all other factors that could have influenced fatalities at the “1% level” must also be considered. As an example, a 2016 peer-reviewed study explored the nuances of how the economy affects traffic fatalities beyond the simple fact that people drive more (or less) during economic booms (or busts) and looked at what else is behind the short-term variations that remain in Figure 2 above.
The following variables were included in their analysis:
- Circumstances of each crash: whether it involved a large truck, a speeding vehicle, a drunk driver, a single vehicle or multiple vehicles, and whether it occurred in an urban or a rural area
- Local unemployment rate
- Beer taxes
- Gas prices
- Handheld device bans
- Texting-while-driving bans
- Seatbelt law primary enforcement (i.e. whether someone can be pulled over solely because they aren’t wearing a seatbelt)
- Graduated driver licensing law for teenage drivers
By accounting for these factors, the authors reached the intriguing conclusion that changes in the vehicle mix (more large trucks on the road) and driver behavior (higher tendency to speed) during periods of economic growth appear to have a greater impact on traffic fatalities than the simple number of miles driven.
In some cases, however, properly controlling for a large number of variables turns up a “null result” — in other words, no significant connection is found between the variables of interest. This was the case with a recent study exploring the link between ridesharing availability and overall traffic fatalities (where the analysis accounted for ten other factors that could affect fatality rates). Other times, exploring the same question with different methodologies and control variables can lead to seemingly contradictory results; this may explain how one report found a significant link between marijuana decriminalization and traffic fatalities, while another (released around the same time) found none.
Correlation vs. causation
Understanding causality is a perennial challenge across all types of research. That is, if we observe two things happening at the same time, can we say that one thing directly led to the other? An example here is the growth of ridesharing in a city occurring at the same time as a decrease in DUI arrests. In this scenario, there are three possibilities: one thing directly contributed to the other; the change happened by coincidence; or both are being driven by some underlying third factor — for example, shifts in demographics, income, local laws, or travel preferences. Even with lots of high-quality data, it can be difficult to determine which of these scenarios is correct.
In some branches of science, controlled experiments are used to solve the correlation-causation conundrum. In transportation research we also rely on “natural experiments” — finding cases when a variable of interest changed suddenly, and looking at the subsequent effects. A recent study used this approach to investigate how the availability of Uber affects the rate of alcohol-involved crashes, with suspensions (and resumptions) of Uber’s service serving as the “experiment.” Ultimately they found mixed results: in two of the cities, Uber availability was linked to a significant decline in alcohol-related crashes, but no link was seen in the other two. The authors suggest that whether or not ridesharing has an effect could ultimately depend on “specific local characteristics,” again emphasizing the importance of understanding and accounting for all variables.
Data quantity vs. data quality
Finally, when conducting any type of research, there is often a trade-off between the quality and quantity of available data. In the case of road safety, we know a lot about each traffic fatality: data are collected according to standards set by the Federal Government, including information on the time, place, and circumstances surrounding each crash, and are centrally compiled by the U.S. Department of Transportation. However, these data are relatively limited in quantity in comparison to other roadway events. For instance, a mid-sized city like Portland, Oregon may experience 40–50 traffic fatalities per year, with substantial year-to-year fluctuations. It can be difficult or impossible to measure the causal effect of a change in the overall transportation system (say, the implementation of Vision Zero policies) on fatalities when the events being tracked are relatively rare.
It is also worth noting that the impact of road safety on lives and livelihoods extends far beyond fatalities. In fact, non-fatal crashes that cause injuries and property damage are over 100 times more common. This raises the question: would it be better to use the far more abundant data on non-fatal crashes to study traffic safety instead? Perhaps, but in these cases the larger quantity of data is offset by inconsistent quality. States have different reporting standards for non-fatal crashes and citations, and many minor collisions aren’t reported to police. Additionally, information on complicating factors like changes to traffic safety laws and their enforcement is often unavailable. Taken together, while there may be a lot of data out there, it can be extremely difficult to find a reliable and self-consistent data set that’s appropriate for the study at hand.
Uber’s commitment to road safety research, and getting to zero road deaths.
The above points highlight just some of the challenges that researchers face when studying the factors that affect road and traffic safety. However difficult this may be, understanding road fatalities is the first step towards ending them, and we are committed to doing our part — through data-driven product improvements, partnerships, and collaborations with independent researchers.
Uber first launched in San Francisco in 2010, and since then we’ve been working to understand not only how drivers and riders use Uber, but also the broader impact our presence is having on the cities and communities we serve. We wholeheartedly support the efforts of those in the road safety community who are working to make fatal crashes a thing of the past, and look forward to opening up new partnerships with researchers to better understand how we can help make this happen.
