Numerous studies have examined the influence of rainfall on the relative risk of crash, and they all agree that rainfall leads to an increase in relative risk as compared to dry conditions; what they do not agree on is the magnitude of these increases. Here we consider three methodological decisions made in computing the relative risk and examine their impacts: the inclusion or exclusion of zero total events (where no crashes occur during event or control periods), the temporal scale of analysis, and the use of information on pavement and weather conditions contained with the crash reports to determine relative risk. Our analyses are based on several years of data from six U.S. states (Arkansas, Georgia, Illinois, Maryland, Minnesota and Ohio). Zero total events in the context of weather related crash studies typically provide no information on the actual crash odds and greatly alter the distribution of relative risk estimates and should be removed from the analysis. While the use of a daily time step provides an estimate of relative risk that is not significantly different from an hourly time step for the majority of rural counties in our study area, the same is true of only 39% of the urban counties. Finally, the use of pavement and weather condition information from the crash reports results in relative risk estimates that are lower than the standard approach, however this difference decreases as rainfall totals increase. By highlighting the influence of methodological choices, we hope to pave the way towards the potential reduction in uncertainties in weather-related relative risk estimates.
Keywords: Crash reports; Odds ratio; Rainfall; Relative risk.
Copyright © 2018 Elsevier Ltd. All rights reserved.