Estimating the prevalence of motor vehicle crashes involving drowsy drivers, United States, 1999 - 2008: Using multiple imputation to address missing data

Background: Researchers have speculated that driver drowsiness is underreported in databases derived from police reports of motor vehicle crashes. For example, in an annual census of fatal crashes that occur in the United States, 3.0% of crashes from 1999 through 2008 were coded as having involved a drowsy driver. Unfortunately, these data do not distinguish between drivers who were known not to have been drowsy and drivers whose drowsiness status was unknown. Thus, the proportion of drivers of unknown drowsiness status who likely were drowsy cannot be estimated from these data. However, another national crash database does make this distinction. Method: A multiple imputation method was used to estimate the proportion of drivers of unknown drowsiness status who likely were drowsy in a representative sample of crashes that occurred in the United States from 1999 through 2008 and involved a passenger vehicle that was towed from the scene (N=47,597). Results: In the original observed data, 3.9% of all crashes and 3.6% of fatal crashes involved a driver coded as drowsy; 59% of all crashes and 87% of fatal crashes involved no driver coded as drowsy but involved at least one driver whose drowsiness status was unknown. In the observed and imputed data, an estimated 7.0% of all crashes (95% Confidence Interval: 4.6% – 9.3%) and 17% of fatal crashes (95% Confidence Interval: 13% – 21%) involved a drowsy driver. Conclusion: Ignoring missing data on driver drowsiness likely results in substantial underestimation of its prevalence in motor vehicle crashes.