Tuesday, July 28, 2009

No Effect, No Cause?

Using a cell phone while driving is A Very Bad Thing. Studies clearly show the accident risk while texting is eight times greater than when keeping both eyes on the road, and all ten thumbs on the wheel. People talking on cell phones are four times more likely to cause a crash.

No doubt.

That undoubtfulness leads to a hypothesis: Since cell phone use while driving is risky, and cell phone usage has become increasingly common since the mid-1990s, then there must be an increase in accident rates over the period.

Turns out that isn't the case.

Number of US highway crashes (1000s) and miles driven (1,000,000s): 1990 - 6,471 (2,144,362); 1995 - 6,699 (2,422,696); 2000 - 6394 (2,746,925); 2005 - 6,159 (2,989,430); 2006 - 5,973 (3,014,116).*

With the minor exception of 1990 to 1995, the number of accidents per year has been decreasing. The rate, based upon miles driven, has decreased throughout the period.

There are two ways of looking at this. It could be that improvements in driver training**, car technology, and road design from 1995 - 2006 lowered the overall accident risk faster than cell phone usage increased it.

Alternatively, researchers had no real idea what they meant by risk. "Cell phone usage increases the risk of an accident by four times." If I increase my risk by four times, does my mishap rate go up by the same amount? If it does not, then how does risk have any meaning? If risk does not reflect in rate, then risk becomes a cause without an effect.

Impressionistically, I am find the former explanation unconvincing. ABS and ASC systems have made cars more controllable. However, most drivers have no idea how to use ABS, and ASC is too new in mass market cars to have made any measurable impact.

So I'll go with the latter. The researchers just know that using a cellphone while driving is more dangerous, so they stopped at demonstrating the "cause".

This reminds me of Warmenism. The Cause is obvious, so obvious that very often no ink is spilled looking for the effect. The Arctic icecap has been melting; well, until recently, anyway. Clearly, obviously, that is due to Global Warming, aka Climate Change. Has anyone seen a story on this subject that cited any actual sea or air temperatures?

*2006 is the latest year available.
** Yeah, right.

12 Comments:

Blogger Mark Frank said...

I am afraid the first explanation is almost certainly correct. And you can't prove anything by looking at overall accident rates.

Whatever your doubts about vehicle technology there is some factor lowering accident rates. This applies in many developed countries round the world - the UK being another one. Maybe it is driver training, drink drive campaigns, better roads, more time in traffic jams. In the UK better roads and drink driving are proven factors.

The increasing number of cell phones is most unlikely to show up in increased accident rates because:

1) The number of driver miles has also increased dramatically - although probably not as much as cell phone usage.

2) On average people only spend a small proportion of the time they are driving using a cell phone. So while it may raise their risk while they are using it, it will not significantly affect their overall risk.

Meanwhile it seems most unlikely that the Virginia Tech Transportation Institute and other reputable researchers are incompetent and the mobile phone manufacturers have failed to pick up on this.

July 29, 2009 1:24 PM  
Blogger Hey Skipper said...

Mark:

You are right -- something is lowering accident rates, and it is applies in many, if not all, developed countries. However, in listing the potential causes, you left one very substantial possibility from the list. (That's okay -- so did I when I wrote the post.)

Demographics.

All developed country populations are aging. Since it is beyond argument that young drivers have a higher accident rate, proportionally fewer of them will, ceteris paribus, guarantee a lower accident rate over time.

In fact, I'll bet there is a not particularly difficult calculation that, given the change in age distribution and the accident rate for each age range, will yield an expected change in auto accident rates.

This, in turn, could be compared to actual accident rates. The difference, if any, is then open to further reasoning about causes of that difference. It may well be that the actual accident rate is higher than that what should be the result of demographic changes. Alternatively, the difference could be the other way around, or even statistically nil.

This is one of my two real gripes with this kind of reporting, and the studies that underly it. The deductive consequences of the cell phones are risky hypothesis are not subtle. As well, the lower accident rate is an accomplished fact. It is a sure sign of religious belief that the hypothesis is taken at face value.

The increasing number of cell phones is most unlikely to show up in increased accident rates because:

1) The number of driver miles has also increased dramatically - although probably not as much as cell phone usage.


Presuming all miles driven are equally risky, then rate eliminates quantity. It just doesn't matter (well, within reasonable ranges) how many vehicle miles there are.

2) On average people only spend a small proportion of the time they are driving using a cell phone. So while it may raise their risk while they are using it, it will not significantly affect their overall risk.

That's a good point. However, it is only as good as the underlying assumption, and also risks denying the underlying numbers.

From your cite, the UK ownership of cell phones has gone from nearly zero to roughly 45 million in twenty years. If there is real risk involved with cell phone usage, how can such a huge change over a short time not show up in accident rates?

Particularly when noting the stats on the first page: according to the study, cellphone usage while driving is common. (Oddly though, there is apparently a study able to determine from outside observation when hands-free users are on the phone. How the heck?)

July 30, 2009 1:59 AM  
Blogger Mark Frank said...

I think you underestimate the quality of academic and government statisticians. I am sure they can and do take into account demographics etc when estimating the effect of various factors on accident rates. I haven't the time to look up the actual research but this kind of thing is fairly routine for a competent statistician.

The cell phone (aka mobile phone here) link is well researched looking at the specifics of just how using cell phones increase the risk. Just look at the ROSPA report to which I linked.

You can see how much the explosion in mobile ownership might be expected to influence overall accident figures by making just a few assumptions. Suppose in 1989 the accident rate is X (accidents/drive mile). Suppose in 2009 every driver owns a cell phone (the highest possible assumption) and using a cell phone while driving increases the rate to 4X per passenger mile. Now suppose that on average a driver uses a mobile phone for 5% of the time they are driving (a very high assumption). The total increase in accident rate due to every driver getting a phone would be 4/20 or 20% over 20 years - roughly 1% a year (actually a bit less because it should be compounded).This is much smaller than the underlying trend which you identify of around 4% improvement per year.

July 30, 2009 6:38 AM  
Blogger Bret said...

"I think you underestimate the quality of academic and government statisticians..."

Is that even possible? :-)

July 30, 2009 7:52 AM  
Blogger Mark Frank said...

In my experience there is nothing wrong with government and academic statisticians. The problem is the spin that politicians and the press put on the figures.

July 30, 2009 9:21 AM  
Blogger Bret said...

Did you miss the smiley face at the end of my last comment?

However, consider:

A: All humans are political.

B: Statisticians are human

Therefore,

C: Statisticians are political

While B may be debatable by some :-), my point is the mere fact that the statistician works in academia or government implies a political bias that is radically different than mine. Therefore, I don't put a lot of faith in much of it (though I'm always happy to quote their statistics when it proves my points :-).

July 30, 2009 10:00 AM  
Blogger Hey Skipper said...

Clearly, I have a bunch of problems with this.

However, they all have to do with the "studies" (scare quotes intentional). I doubt there was a true statistician among them.

It may well be that cellphone usage actually leads to more crashes. However, the "studies" do not get from premise to conclusion. Their assertion of risk is fatuous, because they have not gone from cause to effect.

Kind of like Warmenism, or any other religion.

July 30, 2009 4:01 PM  
Blogger Peter Burnet said...

Skipper, quite apart from the study failing to read in things like the decline in drunk driving, car safety, the ommission of fender-benders from the stats, etc., there is another issue at play here, one that distinguishes this from Warmenism, etc. Hardly anybody is qualified to build or interpret climate models. We are all beholden to what the experts say. Even if a layman sets out to study the competing claims of climate scientists carefully, he still is pretty much restricted to comparing theories/conclusions and has a hard time critiquing their underlying methodologies systematically by measuring them against experience.

But everybody drives and it is quite reasonable for them to bring their daily first-hand experiences to bear on this without worrying too much about what the boffins say about the big picture. I think most folks, certainly in urban areas, would say that driving today is a more intense experience demanding focus and presence of mind. There are a lot more vehicles on the road. People drive impatiently or even angrily, pass on the right, ignore lane-change signals, etc. Driving on a busy Interstate means huge trucks bearing down on you one after the other and passing with little room to spare. I don't think this is just a function of aging or nostalgia either. You just have to visit a small town to compare.

So people intutitively know that the distracted or unfocussed driver in the next lane is a risk to them (and their vehicle) here and now, just as mothers know sending their toddler outside in winter without a hat is a health risk even if it makes no sense according to germ theory and the experts will say cold and wind can't cause ear infections.

July 31, 2009 2:08 AM  
Blogger Hey Skipper said...

Peter:

Skipper, quite apart from the study failing to read in things like the decline in drunk driving, car safety, the ommission of fender-benders from the stats, etc., there is another issue at play here, one that distinguishes this from Warmenism.

Where this is like Warmenism is the a priori assumption that the thing one is attempting to demonstrate is true. When the study says that using a cellphone while talking is 8 times riskier then not, they really should be able to demonstrate how such a difference in risk -- cause -- yields an effect in accidents.

But they didn't. Now, there are two possibilities why not. One is falling prey to the a priori assumption. The second is that they can't. It could be the system is too complex to model what the accident rate would be without cellphones (which should be cautionary to Warmenism, because the traffic system is plug simple compared to climate). It could also be that people compensate for increased risk, which the study mentions in passing, with the consequence that people while using cellphones don't do certain inherently risky things, like changing lanes for instance.

Just because people intuitively know something doesn't make it true. If the underlying agenda is to criminalize behavior (or drive energy costs through the ceiling), then the least we should expect is that people doing these studies actually finish the damn things.

Or that reporters ask the critical questions. Ooops, never mind, wrong planet for that.

July 31, 2009 9:20 AM  
Blogger David said...

Hey Skipper:

I think that you're being unnecessarily hard on this study, although it's a little hard to tell because it hasn't been reviewed or published yet.

Going from the Virginia Tech press release, this was a field study of professional drivers who were monitored for cell phone use and for keeping their eyes on the road. What the study found is that just talking on a cell phone doesn't increase the risk of an accident by much, but that texting or dialing -- in other words, taking your eyes off the road for as long as 5 seconds -- is much riskier than simply driving. So, driving while your eyes are not on the road is riskier than driving while watching the road; not really an astonishing finding.

Now, as you say, because of other counter-balancing trends, increased use of cell phones to text might not result in increased deaths overall, but that's not what the study's about.

More problematic is that this is what we call a field study, rather than an experiment or even a field experiment. That is, the researchers are just watching what happens to a particular population, not manipulating the population. As such, it's completely ambiguous as to causation and is subject to a variety of weaknesses. Given that taking your eyes off the road to text is so obviously stupid, it might be that something about the drivers (stupidity) caused them to text, and that stupidity also causes accidents. In other words, the relationship between texting and accidents is spurious or at least mediated by stupidity.

August 05, 2009 9:39 AM  
Blogger Hey Skipper said...

What the study found is that just talking on a cell phone doesn't increase the risk of an accident by much, but that texting or dialing -- in other words, taking your eyes off the road for as long as 5 seconds -- is much riskier than simply driving.

What do you mean by "risk"?

I remember a study of some six years ago that presumed to demonstrate that additional safety features (ABS, air bags, etc) did not, in fact, produce more safety. The authors of the study studied NYC taxi drivers, and concluded that more safety features were neutralized by riskier driving.

Fine.

The next step, which they, as here, did not take, was showing how riskier driving led to greater risk. If there were no more accidents with the studied group than the control group, then the presumed risk simply did not exist, no matter the appearance of risk.


increased use of cell phones to text might not result in increased deaths overall ...

To more thoroughly isolate risk of an accident, I ignored fatality rates. One can have an accident without a fatality, but one cannot have a fatality without an accident.

August 09, 2009 6:35 PM  
Blogger David said...

Risk here means that an accident is 1.3 times more likely to happen when the driver is talking on a cell phone than when he is not. The correlation in this study is a fact, which might be generalizable, or not. Causation is entirely unclear.

August 13, 2009 2:21 PM  

Post a Comment

Links to this post:

Create a Link

<< Home