Sunday, January 27, 2008

Cell Phones and Driving, Once More

Almost two years ago I wrote about research conducted by the National Highway Traffic Administration and Virginia Tech's Transportation Institute which finds, unsurprisingly, that inattention is a main cause of traffic accidents. Further,
[t]he most common distraction for drivers is the use of cell phones. [T]he number of crashes and near-crashes attributable to dialing is nearly identical to the number associated with talking or listening.... [D]ialing a hand-held device (typically a cell phone) [increased the risk of a crash] by almost three times.
Moreover, as the American Psychological Association points out,
[p]sychological research is showing that when drivers use cell phones, whether hand-held or hands-off [emphasis added], their attention to the road drops and driving skills become even worse than if they had too much to drink. Epidemiological research has found that cell-phone use is associated with a four-fold increase in the odds of getting into an accident [see below] – a risk comparable to that of driving with blood alcohol at the legal limit....

David Strayer, PhD, of the Applied Cognition Laboratory at the University of Utah has studied cell-phone impact for more than five years. His lab, using driving high-fidelity simulators while controlling for driving difficulty and time on task, has obtained unambiguous scientific evidence that cell-phone conversations disrupt driving performance. Human attention has a limited capacity, and studies suggest that talking on the phone causes a kind of “inattention blindness” to the driving scene.

In one study, when drivers talked on a cell phone, their reactions to imperative events (such as braking for a traffic light or a decelerating vehicle) were significantly slower than when they were not talking on the cell phone. Sometimes, drivers were so impaired that they were involved in a traffic accident. Listening to the radio or books on tape did not impair driving performance, suggesting that listening per se is not enough to interfere. However, being involved in a conversation takes attention away from the ability to process information about the driving environment well enough to safely operate a motor vehicle....

Disturbingly, forthcoming research [since reported in "A Comparison of the Cell Phone Driver and the Drunk Driver" and "Cell-Phone Induced Driver Distraction"] will show that talking on a cell phone (even hands-free) hurts driving even more than driving with blood alcohol at the legal limit (.08 wt/vol). When talking on a cell phone, drivers using a high-fidelity simulator were slower to brake and had more “accidents” than when they weren't on the phone. Their impairment level was actually a little higher than that of people intoxicated by ethanol (alcohol).

The studies at Virginia Tech and the University of Utah rely on instrumented vehicles and simulators. Some skeptics dismiss the results of such studies because of their "artificiality." But the results are consistent with after-the-fact analyses of the role of cell-phone use in actual accidents. See, for example, "Association between Cellular Telephone Calls and Motor Vehicle Collisions," by Donald A Redelmeier and Robert J. Tibshirani (New England Journal of Medicine, February 1997), and "Role of mobile phones in motor vehicle crashes resulting in hospital attendance: a case-crossover study," by Suzanne P. McEvoy et al. (BMJ, a journal of the British Medical Association, July 12, 2005).

Redelmeier and Tibshirani analyzed 26,798 cell-phone calls over a 14-month period and found that
[t]he risk of a collision when using a cellular telephone was four times higher than the risk when a cellular telephone was not being used (relative risk, 4.3; 95 percent confidence interval, 3.0 to 6.5). The relative risk was similar for drivers who differed in personal characteristics such as age and driving experience; calls close to the time of the collision were particularly hazardous (relative risk, 4.8 for calls placed within 5 minutes of the collision, as compared with 1.3 for calls placed more than 15 minutes before the collision; P<0.001);> risk, 5.9) offered no safety advantage over hand-held units (relative risk, 3.9; P not significant).
McEvoy et al. queried "456 drivers aged ≥ 17 years who owned or used mobile phones and had been involved in road crashes necessitating hospital attendance between April 2002 and July 2004." The results:
Driver’s use of a mobile phone up to 10 minutes before a crash was associated with a fourfold increased likelihood of crashing (odds ratio 4.1, 95% confidence interval 2.2 to 7.7, P < p =" 0.003)." n =" 21)">All of the studies cited above are microscopic; that is, they examine the behaviors of specific drivers and/or the causes of specific accidents (or simulated accidents). They are also remarkably consistent in their findings: Using a cell phone while driving is risky -- about as risky as driving while drunk.

Saurabh Bhargava and Vikram Pathania (hereafter B&P) are graduate students in economics at UC Berkeley who claim to have refuted the kinds of findings summarized above. Their effort is documented in "Driving Under the (Cellular) Influence: The Link Between Cell Phone Use and Vehicle Crashes" (AEI-Brookings Joint Center for Regulatory Studies, Working Paper 07-15, July 2007). As B&P explain, they
investigate[d] the causal link between cellular usage and crash rates by exploiting a natural experiment induced by a popular feature of cell phone plans in recent years—the discontinuity in marginal pricing at 9 pm on weekdays when plans transition from “peak” to “off-peak” pricing. We first document[ed] a jump in call volume of about 20-30% at “peak” to “off-peak” switching times for two large samples of callers from 2000-2001 and 2005. Using a double difference estimator which uses the era prior to price switching as a control (as well as weekends as a second control), we [found] no evidence for a rise in crashes after 9 pm on weekdays from 2002-2005.
What B&P found, in fact, is a slightly negative relationship between the rise in call volume and the accident rate. (See tables 5 and 6 on pages 28 and 29, and related discussion.) How could that be, if it is inherently reckless to use a cell phone while driving?

B&P's paradoxical results flow from serious shortcomings in their analysis:
  • The actual use of cell phones by drivers isn't known very well; B&P cite only broad averages based on survey samples.
  • The extent to which cell-phone use by drivers actually rises or falls at the switch-over certainly isn't known.
  • The results rest on differences in accident rates between two periods: 1990-98 (before the introduction of "off-peak" pricing) and 2002-04 (after the introduction of "off-peak" pricing). But those two periods differ in potentially significant ways: the incidence of younger persons (i.e., more reckless drivers) in the population, the per capita consumption of alcohol, and the design of motor vehicles and highways. B&P acknowledge the second and third factors, but address none of them quantitatively. (See tables A1, a summary of data sources, and table A2, which gives summary statistics.)
  • B&P conduct three additional analyses (page 30) that, they claim, confirm their "basic results." First, they find (unsurprisingly), a negative correlation between accidents and cell-phone ownership over time, but they merely acknowledge "that there are unobserved variables which are correlated with the growth in cell phone ownership across regions and time." Second, their examination of the relationship between accident rates and cell-phone ownership across areas of varying population density (metropolitan, urban/suburban, rural) is unnecessarily convoluted and, therefore, unconvincing. Third, they trot out the apparent ineffectiveness of legislative bans on cell-phone use with fatal-accident rates in five jurisdictions, but they offer no statistics about the level of enforcement efforts that accompanied or followed the bans.
The bottom line is that B&P's analysis fails to control for time-related variations in critical variables. For reasons detailed in the addendum to this post, time-series analysis is inadequate to the task at hand.

B&P expose some relevant cross-section data, but neglect its implications in their haste to exonerate cell-phone use as a cause of accidents. Figures 2 and 3 (page 4) give indices of cell-phone calls and fatal crashes in 2005, in 10-minute bins from 8 p.m. to 10 p.m. A set of observations for a single year offers the advantage of controlling for time-related factors (proportion of young persons in the population, per capita alcohol consumption, and automobile and highway design). B&P do not divulge the data underlying figures 2 and 3, but -- given the scale on which the figures are drawn -- the data are readily discernible. Regression analysis yields this result:
Index of fatal-accident rate =
- (0.074 x number of minutes after 8 p.m.)
+ (9.787 if weekend, zero if weekday)
+ (0.199 x index of outgoing cell-phone calls)

The t-values of the intercept and coefficients are 2.469, -3.423, 6.183, and 2.670, respectively (all significant at the 0.95 level or higher). The adjusted R-squared of the equation is 0.695. The mean values of the dependent and explanatory variables are 49.692, 60, 0.5, and 130.385, respectively. The standard error of the estimate (3.984)/the mean of the dependent variable (49.692) = 0.080. The equation is significant at the 0.99 level.
The signs of the intercept and the variables are intuitively correct. One would expect (a) a positive "baseline" rate of fatal accidents; (b) a negative relationship between the lateness of the day and the accident rate, as the number of vehicles on the road diminishes and the use of cell phones shifts from the highway to the home; (c) a higher accident rate on weekends, when there is more "partying," especially among younger (i.e., more reckless) drivers; and (d) a positive relationship between cell-phone use and accidents.

In fact, at the mean values of the variables, a 1-percent rise in aggregate cell-phone use leads to a 0.26-percent rise in the index of fatal accidents, which is equivalent to a 0.52-percent rise in the rate of such accidents. Putting it another way, cell-phone use accounted for about 50 percent of fatal accidents during the hours of 8 p.m. to 10 p.m. in 2005. That may overstate the contribution of cell-phone use to fatal accidents, but (given the evidence cited earlier in this post) I have no doubt that it points in the right direction. For example:
  • If Redelmeier and Tibshirani (see above) are right about the relative of risk of collision arising from cell-phone use (relative risk of 4.3 = 3.3 x baseline rate), and
  • about 15 percent of drivers are on cell phones between 8 p.m. and 10 p.m., and
  • fatal accidents rise in proportion to total accidents, then
  • an estimate of about 50 percent is not unreasonable.
Despite having statistically exonerated cell-phone users as a menace to others, B&P concede the opposite. This is from the UC Berkeley press release announcing their paper:
The economists [B&P] don't dispute that using cell phones while driving can be dangerous. Bhargava conducted his own personal experiment, talking on his cell phone while driving in Minnesota this summer. Acknowledging that he doesn't often drive, much less drive and talk on the cell phone at the same time, Bhargava said he almost crashed twice on that trip.

"Our research should not be viewed as an endorsement to use cell phones in a negligent way," he said. "It certainly may be risky for a marginal user."

Pathania added another cautionary note: "Since we know that certain demographic groups such as teenagers frequently call and text while driving, and that they are also risky, inexperienced drivers, further research is needed in this area. Laws banning cell phone use in cars for such groups may well have some merit."

Reality trumps cock-eyed statistical analysis every time.

The moral of the story is that cell phones and driving don't mix. I am sticking with the bottom line of my earlier post:

[F]or the vast majority of drivers there is no alternative to the use of public streets and highways. Relatively few persons can afford private jets and helicopters for commuting and shopping. And as far as I know there are no private, drunk-drivers-and-cell-phones-banned highways. Yes, there might be a market for [such] highways, but that's not the reality of here-and-now.

...I can avoid the (remote) risk of death by second-hand smoke by avoiding places where people smoke. But I cannot avoid the (less-than-remote) risk of death at the hands of a drunk or cell-phone yakker. Therefore, I say, arrest the drunks, cell-phone users, nail-polishers, newspaper-readers, and others of their ilk on sight; slap them with heavy fines; add jail terms for repeat offenders; and penalize them even more harshly if they take life, cause injury, or inflict property damage.

See the addendum at Liberty Corner II.