Effects of Precipitation Type on Crash Relative Risk Estimates in Kansas

There has been considerable interest in documenting the impacts of rain and winter precipitation on motor vehicle crashes, as well as a push towards improving the accuracy of precipitation-type prediction and identification. However, currently there is no documentation of the impacts of specific precipitation types on crashes. In an effort to bridge the gap between the two strands of research, vehicle crash data from Kansas for the years 1995-2014 are used to estimate the crash relative risk during rain, snow, sleet, and freezing rain. A matched-pair analysis approach is used to compute relative risk as the odds of a crash occurring during a precipitation-type event compared to the odds of a crash occurring during a similar non-precipitation control period 1 or 2 weeks before or after the event. Variable-length event periods for each precipitation type are defined through both crash report-identified precipitation types and nearby Automated Surface Observing System (ASOS) and Automated Weather Observing System (AWOS) precipitation reports. Newly developed methods to extract precipitation type beginning and end times within ASOS/AWOS reports allow for event durations as short as a single minute, and provide the most accurate account of precipitation type and crash relative risk estimates.

Crash relative risk estimates and their 95% confidence interval were computed for each precipitation type, where estimates of 1.00 indicate that the risk of a crash occurring during the event period is equivalent to the risk during the control period. Crash risks are higher during precipitation than during normal dry conditions, regardless of precipitation type; however, there exists a hierarchy of risk based on precipitation type. The relative risk of crash during freezing rain (3.09 mean) is significantly higher than for snow (2.30 mean), which is in turn significantly higher than rain (1.69 mean). The 95% CI for sleet (2.55 mean) encompasses values similar to those of both snow and freezing rain. This result is attributable to two reasons: the inability of ASOS/AWOS to detect sleet automatically resulted in a lower number of event-control matches used to compute the relative risk for sleet, and sleet often occurs with additional precipitation types so it is likely that the estimates include the impacts of mixtures. Both reasons likely contributed to the larger CI for sleet. Regardless, there is evidence that freezing rain poses the greatest crash relative risk out of all precipitation types, which is consistent with expectations, but here is quantified for the first time.