Effects of measurement assumptions on estimations of risk of collision

Abstract

This paper provides a replication and extension of previous research describing the relationship between risk of fatal collision and driver blood alcohol content. Risk of collision indices are derived through a comparison of the blood alcohol levels of nighttime Canadian drivers and nighttime driver fatalities. In the context of these data sets an assessment is made of the effects of different treatments of missing data upon estimations of risk of collision. Assumptions regarding the distribution of alcohol among non-respondents in roadside surveys, as well as among driver fatalities who are not tested for alcohol, are shown to have substantial implications for the magnitude of the risk factors derived. Nevertheless, although the magnitude of the relationship between driver blood alcohol and risk of fatal collision varies as a function of the assumptions utilized, the directionality of the relationship corresponds consistently to the results reported in previous investigations.