« Ground rules for tomorrow's MR BookForum | Main | The Revolution is Four Years Old! »
Is cell phone use while driving really so dangerous?
Natural experiments have their uses. Saurabh Bargrava reports his joint work with Vikram Pathania:
...we document a 20-30% rise in cellular call volume during the time of the day-- 9pm on weekdays-- when cell phone providers systematically transition from "peak" to "off-peak" pricing. We then measure the resulting increase in fatal and non-fatal crashes during this period as compared to weekends and earlier periods which serve as controls. We find no evidence for a rise in crashes, and estimate small positive upper bounds for the effect size at 9pm (~1% for all crashes, and 2.4% for fatal crashes).
Here is the paper. Here is a related press release. You might wonder how this can be reconciled with all those studies showing that talking on a cell phone is as bad as driving drunk. The authors discuss the other work starting on p.8 and score some good points. Furthermore get this:
We confirm our results with three additional empirical approaches—we compare trends in cell phone ownership and crashes across areas of contiguous economic activity over time, investigate whether differences in urban versus rural crash rates mirror identified gaps in urban-rural cellular ownership, and finally estimate the impact of legislation banning driver cell phone use on crash rates. None of the additional analyses produces evidence for a positive link between cellular use and vehicle crashes.
No, I am not encouraging you to talk on your cell phone while driving, if only because so many of you already talk so much, and besides, what if these guys are wrong? But science must progress, and in that spirit I report these intriguing results.
Posted by Tyler Cowen on August 20, 2007 at 08:00 PM in Economics | Permalink
Comments
Neat. I wonder whether some of the effect can be explained by the hypothesis that people who tend to wait until after 9pm to make their calls also tend not to make calls while driving (say, because they're younger than 16).
I guess I'll go read the paper!
Posted by: zlguocius at Aug 20, 2007 8:17:36 PM
Maybe the majority of rush hour drivers are sensible and only talk on their phones when stuck in traffic jams, and not when weaving in and out of traffic going 60 miles an hour. I would expect cell phone use to have a bigger impact on the accident rate after high schools let out in the afternoons or in the evening when traffic is going faster.
Posted by: joan at Aug 20, 2007 8:29:41 PM
My contention has always been that regardless, it's just another difficult to enforce law. The kind of person who is so lazy in their daily habits to be a public nuisance with their cell phone, is also a nuisance because they're changing radio stations, eating a burrito, et cetera.
A catch-all law that provides a driver can be prosecuted for "distracted driving" if they do get in an accident is sufficient. This is because it is the act of talking that distracts the mind so much, regardless of where the eyes are focused. And with those tiny little blue-tooth ear pieces, all a driver does is put the device on their right ear, and chat away.
So bottom line is that it is eventually just another law on the books, right up there with spitting on the sidewalk, and jaywalking.
Posted by: Ray G at Aug 20, 2007 8:30:40 PM
Who gets drunk during the day? Perhaps cell phones account for crashes before 9 p.m. and after 9 p.m. (when people are more likely drunk), drinking accounts for the crashes then. In this way, we would not necessarily see an increase in crashes and still cell phones could be shown to have a negative impact on driving.
Posted by: philosophking at Aug 20, 2007 8:37:40 PM
My initial thought was that of the first comment. Are most of the folks calling at 9:00 waiting around at home to call Mom and Dad, or are they driving around when they think to call?
Personally I never think about calling home after 9:00 unless I'm already at home. I'm not as good of a driver when I'm talking on my cell .
Posted by: Shane Milburn at Aug 20, 2007 9:07:50 PM
My initial thought was that of the first comment. Are most of the folks calling at 9:00 waiting around at home to call Mom and Dad, or are they driving around when they think to call?
Personally I never think about calling home after 9:00 unless I'm already at home. I'm not as good of a driver when I'm talking on my cell .
Posted by: Shane Milburn at Aug 20, 2007 9:07:57 PM
Whenever I get pissed off at someone driving stupidly, as soon as I can see them they're invariably on a cell phone.
I believe the cell phone effect to be similar to the old person effect: you slow down instinctively to be safer, like you are subconsciously taking into account your decreased attention and compensating.
So, talking on a cell may not make you less safe, but nevertheless there is a great cost to people doing this - they slow everyone else down and make traffic less efficient. Probably this cost is less than the utility of conversations so banning it is probably a bad idea.
Posted by: Paul N at Aug 20, 2007 9:27:05 PM
About 9.0% of US wireless subscribers use a wireless phone as their primary phone(no land line). How do they determine where these calls are originating from?
Posted by: Pritesh at Aug 20, 2007 10:09:46 PM
Occasionally, statistical robustness is misplaced in policy decisions. Using a sample of one event I have made up my mind. I was hit on my bicycle last week by a cell phone wielding driver who made no effort to look away from his phone as he pulled out into the street.
One cell phone related accident is one to many.
Posted by: Ryan at Aug 20, 2007 10:48:02 PM
How is this a natural experiment? Transition from peak to off-peak pricing for cell phone calls does not happen randomly; it is determined endogenously by many factors which are correlated with (or at least are not orthogonal to) traffic, accidents, etc. Do they even address this in the paper?
Posted by: economister at Aug 20, 2007 11:26:01 PM
Though I'm generally a fan of the Bookings Institute, I think this paper is awful. From the start, there is a chart showing anti-correlation between accidents and cell phone penetration. There is also anti-correlation between accidents and total U.S. beer consumption.
Reasons for lower crash rates include (1) raised ages of drivers in some states, (2) better engineering of cars, (3) better roads, (4) dramatically reduced fraud, (5) better policies on truck driving, (6) toll roads, and (7) the other fourteen thousand factors I didn't think of first. The use of time as a significant variable in this paper is fatally flawed at best, and at worst suggests an engineering of results.
The paper quickly dismisses large amounts of prior research with a few belittling comments, but without substantial argument. I used to work in the actuarial department of an auto insurance company. We hated that people talked on their cell phones while they drove because, in our own records of crashes with our own clients, it was a huge factor in the crashes. By our estimates, greater than the correlation of alcohol consumption.
That was the late 90's, and a decade later, it may be that many people have learned to drive better talking on the phone. Maybe. I know for certain that I have become better at looking at the jerks around me doing it, and making space between me and them. Maybe people like me are suppressing the correlation.
I would also be willing to bet that people have started more frequently lying about being on the phone when they crash. I know my company was formulating a way to add points to drivers with phones built into their cars at the time. I don't know how that went because that was right about the time most car phones were replaced with ordinary cell phones. It's also true that a person talking on a cell phone when they harm another person is open to both civil AND criminal penalties -- so of course the amount of lying has gone up. This paper states that "Disentangling these effects" (of reporting cell phone use to police) is not possible, while completely ignoring an obvious sharp rise in fear of legal responsibility (not to mention higher insurance rates).
And hey, I'm well trained in statistical methods, and I really don't see how this 9 pm thought experiment demonstrates much of anything. I've got to accept a pile of assumptions to think this has anything to do with the drivers who wind up in the most accidents. A late hour like that is about the safest time to drive -- a time when relative inattention has the fewest pitfalls.
There is a lot of bullshit language in this article such as "This study can also be linked to..." with basically no qualification. Previous studies are critiqued for small sample sizes, but the total body of data on this issue is enormous, and the insurance companies have been aware of it (from their own unpublished stats) since the mid-90s.
Then there are claims that cell phones might alleviate boredom and fatigue for some drivers. This may be true, but even so, these distance drivers should be categorized differently from the morning/evening commuters and after-school kid drivers. This would in fact mean that for the rest of drivers, the cell phones were actually a cause of higher incidents of crashes. Perhaps it's rational to allow talking on the cell phones by drivers during off-peak hours only! Who could write a policy paper like this a miss such a suggestion?
I don't believe a lot of actuaries are shaking their heads up and down when they read this. They're shaking it side-to-side.
I've spotted dozens of other mistakes in this paper, but I don't intend to document them all in one grand rebuttal. But it's just plain terrible. It's one of those papers that makes me start to wonder who has what kind of political motive. The actuarial world will surely ignore it, and rightfully so.
Tyler, you usually spot a piece of crap when you see it. I believe you should reread this paper and take a whiff.
Posted by: infopractical at Aug 21, 2007 1:03:01 AM
economister:
9pm is fairly arbitrary compared to 8:30 or 9:30.
infopractical:
You didn't name a single mistake in the paper.
Posted by: Douglas Knight at Aug 21, 2007 1:32:09 AM
It's not a mistake to lead with the use of a graph that takes time into account as a variable when there are an abundant number of highly published reasons not to do so? That's not a fallacy? And is a fallacy not a mistake in a policy paper?
The math is all accurately computed, if that's what you're saying. And that doesn't mean much if the assumptions and logic are flawed.
Posted by: infopractical at Aug 21, 2007 1:40:12 AM
Looking back, I certainly did not articulate well the reasons I dislike this paper so much. Looking back, I am reminded that Tyler took this paper with a grain of salt. And so I wonder why -- did his BS detector move to yellow, not knowing whether it would move to green or move to red?
I believe provoking the yellow light reaction is the point of a paper like this. It gets shock value while at the same time telling "the villains" in question that they're doing the world a service, and don't let anybody tell you that you need a babysitter telling you how to drive.
Actually, I'm quite sympathetic to that last part, and I believe there are better solutions than some or all of the proposed legislation.
I do not believe that a room full of actuaries would be sold by this paper. In fact, I'd guess that said actuaries weren't interviewed for this paper. The list of reasons why that is the case is very long. A fair rebuttal to this paper would be longer than the paper itself. I listed a few thoughts already.
I recommend taking a look around the internet for papers and see if the literature on the subject is large and solid, or easily indicted. Google something like "effects of cell phones on driving." The collection of data paints a clear picture. Many of the studies include long lists of variables used to build large and compelling correlation matrices. It is the suggestion of a nearly flat effect that doesn't fit with the rest of the data.
Having read so many articles and papers in the literature (scores of them), it's very easy for me to focus in on the joke: even if a lot of assumptions were granted on my behalf for the benefit of the writers -- they pick a method of testing that centers around 9pm. Why 9pm? Don't think they didn't have a choice. They could find events centering around any given hour for the purposes of recording the data. They could have picked, I don't know, rush hour, which is on a weighted basis, the most important part of the time curve.
They might also pick two different hours before jumping to a conclusion? While it may have been a little more costly, it seems like the responsible second step before jumping to any conclusions. Extrapolation methods, such as assumptions of nearly convex regions on the effect-time graph, allow a great deal of certainty. But a single data point provides almost no information at all. When somebody with a statistics credential misses that fact, either their credential or their motive is suspect.
Posted by: infopractical at Aug 21, 2007 2:35:37 AM
This is all very interesting but the question from a policy perspective is why do we single out certain sorts of activity that increases risk for regulation while not others? It's my sense that these laws have almost nothing to do with cost/benefit analysis and almost everything to do with what people associate with the guy who cuts them off.
Posted by: TruePath at Aug 21, 2007 2:47:33 AM
Truepath, it is certainly a cost/benefit issue, but it's also a responsibility issue. Automobiles are dangerous, and when people make mistakes, those mistakes affect other people. Like when loaded guns are waved about.
Waving loaded guns is more immediately dangerous, but the overall effects by drivers are far greater. Would you really rather have no laws against drunk driving? How is this any different? Is that really the battle to pick against paternalistic government?
Posted by: infopractical at Aug 21, 2007 3:30:16 AM
Truepath, it is certainly a cost/benefit issue, but it's also a responsibility issue. Automobiles are dangerous, and when people make mistakes, those mistakes affect other people. Like when loaded guns are waved about.
Waving loaded guns is more immediately dangerous, but the overall effects by drivers are far greater. Would you really rather have no laws against drunk driving? How is this any different? Is that really the battle to pick against paternalistic government?
Posted by: Daisy at Aug 21, 2007 5:44:52 AM
They might also pick two different hours before jumping to a conclusion? While it may have been a little more costly, it seems like the responsible second step before jumping to any conclusions. Extrapolation methods, such as assumptions of nearly convex regions on the effect-time graph, allow a great deal of certainty. But a single data point provides almost no information at all. When somebody with a statistics credential misses that fact, either their credential or their motive is suspect.
http://www.cinatrade.com
Posted by: Daisy at Aug 21, 2007 5:45:45 AM
I haven't read the paper yet, but does it isolate the effect of other drivers (who are not using their cell phones) on vehicle crashes? It could very well be the case that the cell-phone users are driving worse, but are compensated for by the skills of other drivers. Or, it could be that drivers talking on the phone compensate for their reduced skills by deliberately driving more slowly and carefully, or only talk on broad stretches of highway which require less attention, or (as suggested above) when they are stuck in bumper-to-bumper traffic.
I recall an episode of Mythbusters devoted to this question; if I recall correctly, talking on the phone had roughly the same effect as a couple beers.
Posted by: Independent George at Aug 21, 2007 9:15:42 AM
One thought that occurs to me is that phone calls after 9 are mostly optional. I would never initiate a personal call in a car, but I constantly have to take work calls during the day in the car. I know that I am significantly worse as a driver when I am on the phone, but emergencies sometimes come up at times when I can't pull over.
The bottom line is that I think the work/personal call timing cuts across their logic.
I feel as others do, that this is one of those "counterintuitive" cases where your intuition is right.
Posted by: RobbL at Aug 21, 2007 9:28:19 AM
Wow... just about a perfect exemplar of what you need to get an economics paper some attention:
(1) A hot topic
(2) A result at odds with the literature and, to some extent, common sense
(3) Three hot techniques: A natural experiment, difference-in-difference estimators AND regression discontinuity
(4) A hot adviser -- David Card
(5) An implication of consistency with optimizing behavior, with an obligatory reference to Peltzman (1975). Admittedly, this one cuts both ways, since showing some behavioral blindspot would be eye-catching as well.
Posted by: Jonathan at Aug 21, 2007 9:32:22 AM
I use my cellphone on occasion, while driving.
Since I have never driven with any level of intoxication, I cannot directly compare the two but I would say that intoxication would be more distracting than cellphone usage. Being intoxicated does not allow for regulating. e.g. you cannot directly control your level of intoxication through the different types of driving situations.
I can use discretion when talking on the phone. I will usually make calls only when I am on a highway, where there is relatively less attention required from me. If I am navigating through local roads I would refrain from making a call.
Using a cellphone is distracting but generally the risks can largely be mitigated through regulated usage. While one may be able learn to drive well with self regulated cell phone usage. I'd still imagine that response times to sudden change in road conditions (e.g. deer) are reduced when using a cell phone. Additionally if one is not using a headset their vehicle control might be reduced as well since they don't have usage of both hands.
Ultimately I'd assume that cell phone usage is more akin to other newly available distractions in a car such as tvs, GPS systems but would top the category in terms of distraction. I do agree that there should continue to be penalties for engaging in intentionally distracting activities while driving.
Posted by: Josh at Aug 21, 2007 9:56:37 AM
I'm not totally up on all the methodology in any of these studies, but I did watch Mythbusters tackle the question. I didn't like that the cell phone use they used in their experiment was asking questions on a numbers and logic test and requiring the driver to answer, in my experience cellphone use during driving is mostly idle chatter and not brain bending calculation. I'm not saying that this is how any of the scientific research is conducted, just that I was dissapointed by the Mythbusters particual methodology in one episode (but it's still a great showand I watch whenever I see that it's on).
Posted by: Nathaniel at Aug 21, 2007 12:58:32 PM
I'm not totally up on all the methodology in any of these studies, but I did watch Mythbusters tackle the question. I didn't like that the cell phone use they used in their experiment was asking questions on a numbers and logic test and requiring the driver to answer, in my experience cellphone use during driving is mostly idle chatter and not brain bending calculation. I'm not saying that this is how any of the scientific research is conducted, just that I was dissapointed by the Mythbusters particual methodology in one episode (but it's still a great showand I watch whenever I see that it's on).
Posted by: Nathaniel at Aug 21, 2007 12:59:41 PM
I didn't read the paper thoroughly, but do they account for two things?
1) Whether the mix of car-based cell use and land-based cell use is the same around 9pm as at other times of the day? I suspect that more people are using phones at home at 9p than at other points in the day (e.g., rush hour). If it differs, that could skew the results.
2) The overall percentage of cell use in cars vs. not cars. If most of the calls around 9pm are from folks at home (intuitively this makes sense--you don't sit in your car waiting for cheap calling), then won't the number of calls drown out statistically any increase in accidents?
Posted by: ah at Aug 21, 2007 1:59:12 PM
