Lab studies have shown that marijuana can severely impair driving skills. Epidemiological studies, however, have been inconclusive regarding the contribution of marijuana use to crash risk. In the United States, case-control studies based on the merging of comparable crash Fatality Analysis Reporting System (FARS) and non-crash National Roadside Survey (NRS) data have been applied to assess the contribution of drugs to crash risk, but these studies have yielded confusing, even contradictory results. We hypothesize that such a divergence of results emanates from limitations in the databases used in these studies, in particular that of the FARS. The goal of this effort is to examine this hypothesis, and in doing so, illuminate the pros and cons of using these databases for drugged-driving research efforts. We took advantage of two relatively recent cannabis crash risk studies that, despite using similar databases (the FARS and the NRS) and following similar overall approaches, yielded opposite results (Li, Brady, & Chen, 2013; Romano, Torres-Saavedra, Voas, & Lacey, 2014). By identifying methodological similarities and differences between these efforts, we assessed how the limitations of the FARS and NRS databases contributed to contradictory and biased results. Because of its limitations, we suggest that the FARS database should neither be used to examine trends in drug use nor to obtain precise risk estimates. However, under certain conditions (e.g., based on data from jurisdictions that routinely test for drugs, with as little variation in testing procedures as possible), the FARS database could be used to assess the contribution of drugs to fatal crash risk relative to other sources of risk such as alcohol.
Keywords: Alcohol; Cannabis; Crash risk; FARS; NRS.