Using Complimentary Set Analysis to Validate the Underlying Assumptions of Quasi-induced Exposure

In the recent decades quasi-induced exposure has enjoyed increasing popularity with applications in the traffic safety analysis. However, issues have been raised that the majority of the relevant studies do not particularly attempt to verify the validity of the induced exposure technique prior to its adoption. In an effort to validate the critical not-at-fault assumption at the core of the applications, complimentary set analysis (a technique to test whether a driving cohort is randomly selected by its complimentary set of drivers of the same classification) is used and tested. The paper supplements the technique with a comprehensive statistical testing framework, which will enable the validation of the assumption to be conducted for various driver-vehicle characteristics (>2) at much more finely-disaggregated levels. The main findings of the research include: 1) at the most aggregated level, statistical testing does not support the hypothesis that one innocent driver-vehicle combination in the driving population is randomly impacted by the culpable parties of the same classification, mainly due to data aggregation and exposure data irregularities; 2) statistical results demonstrate an increasing trend of p-values when data are finely disaggregated in a stepwise manner, confirming the random-selection assumption of quasiinduced exposure; and 3) an important phenomenon inherent in the exposure matrix is that a driving cohort has a higher probability to collide with the same driving type as opposed to others of the same classification. Through the study it has been verified that complementary set analysis is a straightforward, convenient, and effective technique to check the validity of quasi-induced exposure.