Programs & Labs
Subhead H2
Esse molestie consequat. vel illum dolore eu feugiat? Est usus legentis in qui facit eorum claritatem.
Fall 2003
Putting limits on the speed-versus-safety debate
Gary Davis
A new study may help put limits on the debate about the role speed plays in car crashes.
Although it may seem obvious that driving slower is safer, there is a lack of consensus in the public and policy arena about the connection between the speed one chooses to drive and the risk of being involved in a crash. The issue is further muddied by several recent studies that suggest only a weak connection between driving speed and crash risk.
Associate Professor Gary A. Davis of the University's Civil Engineering Department is hoping his research will help clarify the issue and inject some facts into the debate.
According to Davis, problems with previous research on the subject have produced misleading results. For example, one commonly used research approach looks for associations between a road's average speed or speed variance and its crash rate, without looking at the speeds of vehicles actually involved in crashes. Unless carefully done, this approach can produce classic statistical misinterpretations, such as the ecological fallacy and Simpson's paradox. On the other hand, crash reconstruction techniques do attempt to look at the speeds of vehicles involved in crashes, but haven't accounted for inevitable measurement errors when estimating the speeds of vehicles. These errors in speed measurement can give rise to inaccuracies in statistical results.
A case-control approach
Rather than study broadly aggregated data, Davis and graduate students Sujay Davuluri and Jianping Pei decided to focus on how speed affects crash risk at the individual level. To avoid the measurement-error pitfalls of previous research, the team developed a method for quantifying the uncertainty in estimates made from crash reconstructions, and incorporated this into a case-control study of vehicle speed and crash risk. The "cases" were a sample of vehicles involved in fatal crashes, for which the researchers collected and analyzed data from the Minnesota Department of Transportation and the Minnesota State Patrol.
For each case, "control" speeds were collected on vehicles passing the crash site during conditions similar to those when the crash occurred. Crash reconstruction methods were used to estimate each case vehicle's speed. Rather than use a straightforward case-control approach, the heterogeneity in the crashes they had to work with required them to develop new methods for extracting information from very small samples, Davis says.
The researchers then conducted analysis of cases and controls, generating both parametric and non-parametric estimates of the relative risk of crash involvement as a function of vehicle speed using methods for modeling causal inference for artificial intelligence. The use of a computational technique called Markov Chain Monte Carlo simulation to compute probabilities, given evidence, helped solve previous estimation problems.
By entering the speed of a vehicle into an equation, the researchers can determine the probability of necessity of that speed for the crash to occur. If the speed of the vehicle is unknown, the probability of avoidance can be determined with a related equation and the range of possible speeds.
Taking chances with higher speeds The findings, Davis reports, show that higher speeds are associated with an increase in crash risk, and in some cases, the researchers could identify speed as a definite causal factor for a crash.
"Speed by itself doesn't appear to be sufficient to cause a fatal crash," Davis says. Rather, he adds, speed comes into play when a driver finds him- or herself in a crash-avoidance situation, resulting either from driver error or by a surprise event outside of his or her controla pedestrian stepping suddenly into the street, for example. In such a case, speed, while not by itself causing the crash, can make it difficult to avoid a collision.
"Contrary to what some have argued, speed is not benign," Davis says. "The question we all have to ask ourselves when selecting a higher speed is, 'Do we feel lucky?'"
Should we just slow down?
Davis says he doesn't expect the research results to dictate policy; instead, he hopes to inform the ongoing debate about the role of speed in determining crash risk. Aside from influencing what speed limits should be, the findings could potentially help law enforcement and traffic engineering agencies decide how much emphasis to put on speed limits and their enforcement. This in turn might lead to a need for automated enforcement methods or for non-enforcement approaches (such as in-vehicle driver feedback) for controlling speed. On another level, the information might spur further debate about the tradeoffs between the benefits of higher speeds (such as shorter travel times) and the costs associated with a potentially increased crash risk.
Davis does hope the research will impact individual drivers. Once a crash occurs, the role of speed in causing damage is clear. While the research did not find that speed was to blame for all crashes, speed remains a risk factor for crashes in certain situations. And unlike crash triggers, speed is a risk factor that is under the control of the driver. Davis says that when a trigger appears, driving at a reasonable speed may be the difference between being involved in and avoiding a crash.
"It is important to drive like the next accident trigger is just down the road," he says.