Wait, Wait... Don't Deceive Me
A week or so ago, the popular NPR news quiz show "Wait, Wait...Don't Tell Me" asked guests to fill in the blank on the following statement:
Contrary to popular belief, new research suggests (BLANKING) can actually improve driver safety.
Answer: TALKING ON THE PHONE.
What!? How could this be?!
Well, for starters, our friends at "Wait, Wait Don't Tell Me" didn't get anything wrong. After all, that was one of the stories in the news that week. But the rest of media definitely blew it. Here's how it happened.
The report never claimed that using a cell phone actually improved driving safety. The research in question (which is available online, here) actually claimed that cell-phone use did not increase crash risk. Of course, this claim goes against several hundred scientific studies, surveys of drivers, and common sense. The claim simply flies in the face of everything that is known about the neuroscience of attention. A central tenet of cognitive neuroscience is that you cannot do two attention-demanding activities at the same time without impairments to one or both tasks. So just what is the basis for this claim and is it reasonable?
The authors of the report based their claim on a sample of California cell-phone subscribers who increased their phone usage at 9:00 P.M. based on cellular pricing plans in place in 2005. They found that there was a 7% increase in call usage at 9:00 on Monday through Thursday, but a much smaller increase on Fridays (4%) or weekends (2.5%). The study looked to see if there was an increase in crashes at 9:00 P.M. on Monday through Thursday. In fact, there was a slight increase in crashes at 9:00 PM.
Because the increase in crashes may have been due to rounding in police reports (e.g., saying the accident happened at 9:00 instead of 8:57), they decided to use a comparison interval based on 1998 crash data. The 1998 crash data also showed a slight increase in accidents at 9:00 P.M., but the authors assume that since the increase in crashes was about the same for 1998 and 2005 it means that cell-phone use has no effect on accident rates.
So their argument boils down to something like this: There were more cell phones in use by drivers at 9:00 P.M. in 2005 than there were at 9:00 P.M. in 1998 and the increase in crashes at 9:00 P.M. was about the same in 2005 as it was in 1998 - hence adding more cell-phone drivers to the mix didn't increase crashes.
But before we accept this conclusion (and discard the hundreds of studies published in the scientific literature), there are several assumptions that need to be considered.
First, the authors had to guess if the people using a cell phone at 9:00 were actually driving. They assumed that drivers with cell phones in 1998 were not on the phone (there is little evidence to support this assumption).
The report also relied on cell-tower records to determine if cell-phone users were driving. If the records indicated that a cell call switched from one tower to another, then the caller was assumed to be driving. Of course, if a passenger (or someone in a taxi, bus, or train) was using his phone, this would mistakenly classify the phone user as a driver. It is difficult (but not impossible) for someone who is not driving to cause an accident, so this assumption would make cell phones appear safer than they really are.
Also problematic in this assumption is the fact that cell-tower records are not reliable indicators of driving - calls switch towers based on call volume as well as location so that even if a caller is sitting in her house, her call may switch between towers. (I have seen deposition testimony from a provider in California which states that you cannot reliably use signal tower information to determine the location of a caller.) So the precision of the authors' estimates for cell-phone usage while driving is unknown for both 1998 and 2005. If you cannot reliably determine if someone is or is not driving, then it is difficult to estimate the crash risk from cell-phone use.
Another assumption is that a 7% increase in call volume at 9:00 P.M. would result in a meaningful increase in the overall proportion of drivers using a cell phone. Accurate statistics are simply not available for cell-phone usage at 9:00 P.M. But daytime estimates suggested that approximately 10% of drivers on the roadway were using their cell phones in 2005. A 7% increase in usage during the daytime would result in a 0.7% increase in the total number of drivers using their cell phones (i.e., 7% of 10% = 0.7%). There is no evidence that this small increase in cell-phone drivers would be detectable in crash statistics (and, as mentioned above, this 7% increase is likely to be an overestimate as is the cell-phone usage of drivers at 9:00 P.M.).
Another assumption is that drivers, driving conditions, and percentage of drivers using a cell phone at 9:00 P.M. are representative of all drivers and conditions across the nation. For example, our research indicates that the impairments from using a cell phone increase as traffic increases. If the traffic density is lower at 9:00 P.M. than during the daytime, then this would underestimate the risks of driving while using a cell phone.
Another fly in the ointment is that there were more crashes reported in their sample in 2005 than there were in 1998, a trend that is different from the national statistics (nationally, crashes actually declined from 1998 to 2005). It is difficult to know what to make of this anomaly and how it might have contaminated the estimates.
So, there are assumptions about cell-phone usage while driving, accident rates, and representation of the data that limit the conclusions that can be reached by these authors. Trying to make strong inferences based on this sort of data is fraught with problems (there are simply too many questionable assumptions).
Even with all the troublesome assumptions, the upper bound estimates of crash risk from concurrent use of a cell phone in the report was three times higher than normal. Remember that there are hundreds of scientific studies, using a variety of methods that show that using a cell phone increases crash risk. Probably the best estimates of risk come from epidemiological studies that obtain the cell-phone records from people who have been in crashes to determine the odds of getting into a crash when the driver is on the phone. These studies find that drivers are four times more likely to get into a crash when they are using their cell phones. The same crash risk that is found when drivers are legally drunk (.08 in all 50 states) .
So, you should be very skeptical of the media reports that talking on a cell phone makes you a safer driver - it just isn't true.