Cell Phone Use and Traffic Accidents, Revisited
A recent study on cell phone use and driving behaviors (Does Cell Phone Conversation Impair Driving Performance?) found that having a cell phone conversation during a driving simulation impaired driving performance. Laboratory studies such as this have the advantage of being able to carefully control the administration of a hypothesized risk factor - exposure - such as a cell phone conversation. But laboratory studies are limited in their ability to assess the effect of cell phone use on the actual outcomes of greatest concern - traffic accidents and resulting injuries or fatalities. For such real-life conditions, epidemiology may provide a better picture of reality than can be achieved in the lab.
Most of the epidemiological studies to date have found indications of a link between cell phone use and auto accidents, but a causal connection has been extremely difficult to establish. This is in part due to two separate but related methodological issues: exposure assessment and confounding.
Exposure Assessment is often challenging in epidemiological research. First, the exposure must be defined in a measurable way. Is the presence of a cell phone in a car of interest? Ownership of a cell phone? Hours of phone use per month? The action of dialing a phone while driving? Being engaged in a potentially attention-diverting conversation? Each of these definitions of exposure carries with it certain assumptions, and accompanying threats to the validity of the study.
For instance, ownership of a cell phone may be easy to ascertain, but this broad definition of exposure leaves the analysis vulnerable to confounding. Confounding occurs when a third factor is associated with both the exposure and the outcome. If people who own cell phones have more car accidents than those who do not, how do we know if the higher accident rate is due to cell phone use while driving, as opposed to characteristics associated with cell phone ownership? These might include impatience, fast driving, or a tendency to engage in multitasking.
Defining the exposure more narrowly could help to reduce confounding. In this case, one could narrow the exposure definition to “phone use just prior to an accident.” The problem then becomes one of measurement. If we fail to correctly classify those who were on the phone at the time of an accident versus those who were not, we may lose the ability to detect an effect, or an effect might be exaggerated. This issue of misclassification of exposure was highlighted by Tom and Ray Magliozzi - “Click and Clack” - of National Public Radio’s “Car Talk.” They criticized a widely publicized study sponsored by the AAA’s Foundation for Traffic Safety that found cell phone use did not increase the risk of accidents as much as other driver-distractions. The radio hosts argued that the study’s use of self-reports by drivers - made to police at the accident scene - was likely to result in substantial misclassification of exposure.
In contrast, Redelmeier and Tibshirani (New England Journal of Medicine, Feb. 13, 1997) matched the times of cell phone use recorded on phone bills with the times of accidents given in police reports. This approach offers a good example of an attempt to reduce exposure misclassification. By completely separating the ascertainment of the exposure (the phone call) from the report of the event (the accident), it was less likely that the outcome could influence the reported phone use. But, there is still a risk of exposure misclassification. These researchers assumed precise recording of both the time of the accident and the time of the call. Misclassification could result if there were only a few minutes difference in the time standards used by the phone company and by the person reporting the accident. The figure depicts how the use of a cell phone immediately after an accident could end up being recorded by researchers as occurring before the accident.
Research into the association between traffic accidents and use of cell phones highlights both some strengths (a closer relation to real-life than can be achieved in the lab) and some weaknesses (difficulty in defining, measuring, and classifying exposure) of field research. Such methodological difficulties do not mean, however, that policy makers should dismiss the findings of epidemiological studies. The four-fold increase in risk found in the study by Redelmeier and Tibshirani is unlikely to be due to confounding and measurement error alone. But no single approach can determine the degree to which using a cell phone while driving poses a safety hazard. A clearer picture will result by examining studies using a range of approaches - epidemiological, on-road observation, and off-road simulation.